Académique Documents
Professionnel Documents
Culture Documents
courses
RELIABILITY AND RISK ANALYSIS
VOLUME 1
Safety of
Chemical Batch Reactors
and Storage Tanks
edited by
A. Benuzzi a n d J. M. Zaldivar
EURO
COURSES
A series devoted to the publication of courses and educational seminars
organized by the Joint Research Centre Ispra, as part of its education and
training program.
Published for the Commission of the European Communities, DirectorateGeneral Telecommunications, Information Industries and Innovation,
Scientific and Technical Communications Service.
The EUROCOURSES consist of the following subseries:
- Advanced Scientific Techniques
- Chemical and Environmental Science
- Energy Systems and Technology
- Environmental Impact Assessment
- Health Physics and Radiation Protection
- Computer and Information Science
- Mechanical and Materials Science
- Nuclear Science and Technology
- Reliability and Risk Analysis
- Remote Sensing
- Technological Innovation
Safety of Chemical
Batch Reactors
and Storage Tanks
Edited by
A. Benuzzi
and
J. M. Zaldivar
Commission of the European Communities,
Joint Research Centre,
Institute for Safety Research, Ispra, Italy
LJ
ff
KLUWER ACADEMIC PUBLISHERS
DORDRECHT / BOSTON / LONDON
ISBN 0-7923-1233-3
Publication arrangements by
Commission of the European Communities
Directorate-General Telecommunications, Information Industries and Innovation,
Scientific and Technical Communications Service, Luxembourg
EUR 13457
1991 ECSC, EEC, EAEC, Brussels and Luxembourg
LEGAL NOTICE
Neither the Commission of the European Communities nor any person acting on behalf of the
Commission is responsible for the use which might be made of the following information.
CONTENTS
1
19
49
79
99
125
147
161
201
227
255
285
299
311
355
371
391
409
INCIDENTS
REACT,IjNS
IN
THE
CHEMICAL
INDUSTRY
DUE
TO
THERMAL-RUNAWAY
CHEMICAL
1)
2)
Introduction
Thermal-"runaway" is characterised by progressive increases in rate of
heat generation, temperature and pressure (the latter generally caused by
components in the reaction mass vapourising and/or decomposing to yield
gaseous products at the elevated temperatures involved).
Thermal-runaway begins when the heat generated by a reaction exceeds the
heat removal capabilities of the hardware in which the reaction is being
carried out. At first the accumulated heat produces a gradual temperature
rise in the reaction mass which causes an increase in the reaction rate.
This self-accelerating process may finally lead to an explosion. The
problem is that an increase in temperature has a linear effect upon the
rate of heat transfer but has an exponential effect on the rate of
reaction and subsequently on the rate of heat generation.
Runaway is a major problem in unsteady-state batch reactors, since the
task of specifying the design, operation and control of an apparently
simple kettle reactor with stirrer, heating/cooling coils, possibly reflux
facilities, and emergency relief venting can be difficult, if all the
time-dependent parameters are considered. It is a task which requires a
systematic approach. The problem is often compounded because batch
reactors are frequently multi-purpose rather than dedicated to one
process. Due to economic factors a batch reactor may be used to carry out
many different chemical processes, and it is necessary to ensure that the
heat of reaction does not exceed the existing cooling capacity of the
vessel for each reaction.
A. Benuzzi and J. M. Zaldivar (eds.). Safety of Chemical Batch Reactors and Storage Tanks, 117.
1991 ECSC. EEC. EAEC. Brussels and Luxembourg. Printed in the Netherlands.
3
BPF document is specific to phenolic resins the general approach adopted
could be used elsewhere. It is perhaps significant that no phenolformaldehyde polymerisation incidents have been reported over the last few
years.
THE PRIME CAUSES
The prime causes which led to overheating and eventual thermal-runaway for
169 of the incidents (20 were without sufficient details for the
assignment of a prime cause) are classified below under the main
headings:
(a)
(b)
process chemistry;
plant design and operation.
(a)
PROCESS CHEMISTRY
(i)
Reaction
Chemistry/thermochemistry
1
1
1
34
(20%)
water contamination
other impurities
1
15
(b)
(i)
Temperature
Control
(9%)
are
1
32
(ii)
(19%)
Agitation
inadequate stirrer specification
2
17
(iii) Mischarging
of
Reactants
12
(10%)
wrong material
vindercharging
1
35
(iv)
-
Maintenance
equipment leaks (scrubber 1; valves 3;
cooling pipes/jacket 3)
unauthorised modifications
1
25
(21%)
(15%)
1
11
(6%)
INDUSTRIES INVOLVED
Batch reactors are ubiquitous in the chemical industry due to their
convenience and flexibility. The pattern of incidents, however, shows, as
might be expected, a preponderance to certain specific industries (Table
2).
RECENT INCIDENTS
The analysis of Barton and Nolan (2) covered the period 1962-1984. The
data covering 1985, 1986 and 1987 can be summarised:A total of 47 incidents were reported, 3 in 1985, 16 in 1986 and 28 in
1987. Either there was a real upsurge in incidents in 1986 and 1987,
which seems unlikely, or, which seems more probable, the impact of the new
reporting regulations (Reporting of Injuries, Diseases and Dangerous
Occurrences Regulations 1985 [RIDDOR]) has resulted in improved
reporting.
The prime causes (3 incidents in 1987 were without sufficient information
for the assignment of a cause) of the incidents follow the familiar
pattern:
8 (18%) (ca. average) were due to little or no study or research or
development work being done before scaling up and going into production.
14 (32%) (well above the average) were due to mischarging of reactants of
which 4 were due to overcharging (1 catalyst); 4 were due to addition of
the wrong material, e.g. drums of wrong material were stored with drums of
one of the reactants and were charged in error; 3 to too rapid addition; 1
to wrong sequence of addition; 1 to undercharging of a reactant and 1 to
improper control (use of a hosepipe).
4 (9%) were due to temperature control failures.
10
intermediates/mixtures can be found using laboratory techniques. A
variety of laboratory techniques are available for use to acquire this
knowledge. The Association of the British RTarmaceutical Industry
developed a laboratory scheme (5) for screening new products and
processes. More sophisticated techniques include use of accelerating rate
calorimetry (6) or other adiabatic calorimetry systems. The study of
reaction mixtures is ideally carried out by using a heat flow calorimeter
(7). These techniques will be described in more detail by other speakers
at the Symposium. A thermal hazards assessment strategy is discussed
below.
It is also possible to obtain information relating to changes in heat
transfer coefficients and control parameters, due to changes in properties
such as viscosity and specific heat as the reaction proceeds, using heat
flow calorimetry (8).
The laboratory studies can provide data on the onset temperature of and
magnitude of exotherms. The detected onset of an exotherm is scale
dependent i.e. the larger the reaction mass, the lower the onset
temperature. From such information and a thorough examination of previous
plant operating experience, it is possible to set safety margins and hence
select the operating temperature for the given reactor charge size.
The ensured quality of the raw materials is vital to safe operation. The
analysis shows that the presence of impurities, water in particular,
appears to present a problem. The presence of water can cause additional
heat evolution, raising the total heat output above the reactor cooling
capacity, leading to temperature rise and increased rate of reaction
causing subsequent further increases in heat generation.
With reference to the prime causes relating to plant design and operation,
it is obvious that heat removal rate is an important criterion for batch
reactor design, to which adequate agitation, eg stirrer speed, is related,
particularly with regard to scale-up from laboratory data. Numerous
correlations exist for heat transfer in agitated, jacketed vessels (9,10)
and it is possible to scale-up data on inside film heat transfer
coefficients from heat flow reaction calorimeters to industrial size batch
reactor plant (8). It is imperative that the cooling capacity of the
designed plant can cope with the heat generation from all the chemical
processes envisaged.
It is unusual for batch reactor plant to be designed to resist any
calculated pressure rise resulting from a runaway reaction. Ideally, of
course, the objective should be for process control to eliminate any
runaway potential. However, pressure relieving of the reactor or dumping
the contents or quenching the reaction should be considered in case of
emergency. If pressure relief venting is considered, attention must be
materials
and
unit
processes
which
are
potentially
(b) quantify the hazards which arise from these with a rujiimum of
testing.
It involves a sequential approach, which cavers thermochemical evaluation,
reaction calorimetry and the effects caused by scale, accumulation and
cooling/agitator failure.
A typical strategy is shown in Figure 1 (13,14). This is discussed more
fully in the references given.
The thermochemical evaluation consists of data on the thermal stability,
heat of reaction and total heat capacity of reactants of the desired
reaction, the expected adiabatic temperature rise and any general process
hazards, e.g. flammability and toxicity of reactants.
12
Reaction calorimetry, either in the form of heat flow or adiabatic Dewarbased calorimeters allows the measurement of many process variables
(agitation,
heating,
and
cooling
requirements)
and
reaction
characteristics (kinetics, reaction enthalpy, heat release rates and
reactants1 heat capacity) under known environmental heat loss conditions.
The reaction calorimetry stage of the assessment also allows for the
determination of adiabatic temperature rise and gas generation potential.
The heat release per unit mass or unit volume of reactants can be used
with the previously established plant cooling capacity to ascertain safety
margins for safe operation. It is also usually necessary to consider the
potential results following the failure of agitator and cooling systems,
along with the results from heat accumulation storage tests.
Conclusions
Despite the apparent knowledge which exists, the techniques which are
available and the commercial instruments on the market for the assessment
of potential runaway reactions, to aid process and plant design, control
and operation, incidents continue to occur due, in the main, to common
errors.
The hope is that more chemical manufacturers will introduce systematic
assessment procedures. A systematic approach should reduce the types of
common errors exemplified in the analysis. It is essential to have a
thorough understanding of the process chemistry and thermochemistry and
then to ensure adequate engineering design for heat transfer, adequate
control systems and safety back-up systems and adequate operational
procedures, including training.
An assessment strategy for chemical reaction hazards, has been outlined.
A need is perceived for coherent and concise guidance to be produced,
particularly for small and medium-sized companies, covering the areas of
thermal hazards assessment, venting, and a formalised approach to process
control. HSE has now initiated, and in part, sponsored, the production of
a User Guide on safety in exothermic reactions by I Chem E. Other
sponsors have come from industry. The publication is being written by an
Industrial Fellow reporting to a Steering Committee. It will seek to
bring together information produced in the last few years on all aspects
of the subject, including thermal hazards assessment, process design, heat
transfer problems, process control, vent sizing and operator training. It
will not be a full text-book but should alert smaller to medium sized
companies to the problems in these areas and point out where to go for
further help and advice.
REFERENCES
1
Vclon, P.F., and Barton, J.A., 1987. Some lessons from thermalrunaway incidents. Journal of Hazardous Materials 14, 233-239.
Brogli, F., Giger, G., Randegger, H., and Regenass, W., March 1981.
Assessment of reaction hazards by means of a bench scale heat flow
calorimeter in: Runaway reactions, unstable products and combustible
powders, I.Chem.E. Symp. Ser. 68, Chester.
Steele, C.H., Ph.D. thesis, Heat transfer characteristics and scaleup under isothermal and reflux conditions in batch reactors (in
preparation). South Bank Polytechnic.
10
Chilton, C.H., Drew, T.B., and Jebens, R.H., 1944. Heat transfer
coefficients in agitated vessels, Ind. Eng. Chem. 36, 510.
11
12
14
13
Cronin, J.L., Nolan, P.F., and Barton, J.A., June 1987. A strategy
for thermal hazards assessment in batch chemical manufacturing.
I.Chem.E. Symp. Ser. 102, UMIST, Manchester.
14
15
KLetz, T.A., 1986. HAZOP & HAZAN - Notes on the identification and
assessment of hazards, I.Chem.E (Loss Prevention), Rugby.
16
17
heat release
DESIRED PROCESS
reaction temperature
addition times
maximum holdup time
operating procedures
THERMAL
STABILITY OF
REACTION
GOMPONENTS
PLANT
(OPERATIONAL
DATA
cooling capacity
control parameters
MODEL PROCESS :
REACTION
CALORIMETRY
REDEFINE
CONDITIO N
PROCESS DEVIATION
REACTION
CALORIMETRY
magnitude of exotherm,
adiabatic temperature
increase,
gas generation
secondary reactions
product stability
residual cooling
requirement
PLANT FAILURES
&MALOPERATIONS
IDENTIFIED DURING
PROCESS ANALYSIS
PROCEED TO
PILOT PLANT
16
Table 1
Number of incidents per specified chemical process
Chemical Process
Number of Incidents
Polymerisation
(including condensations)
Nitration
Sulphonation
Hydrolysis
Salt formation
Halogenation
(Chlorination and Bromination)
Alkylation using
Friedel and Crafts Synthesis
Amination
Diazotisation
Oxidation
Esterification
64
15
13
10
8
8
5
4
4
2
1
134
Table 2
Specific manufacturing industries, in which reported batch reactor
runaway incidents have occurred during the period 1962-1987
Manufacturing Industry
Number of incidents
51
41
20
13
13
13
5
5
5
23
189
17
APPENDIX 1
IDENTIFYING HAZARDS
Among the better known formal systems are 'Hazard and Operability Study1
(HAZOP), used to identify hazards, and Hazard Analysis (HAZAN), used to
quantiry hazards (15,16).
Having identified a hazard it is still necessary to decide what to do
about it. Ways must be found to reduce the probability of a runaway
occurring.
Where consequences are judged to be severe, or where the causes giving
rise to the hazard are many or interrelated, it is recommended that a
'fault-tree' (17) is constructed, showing the way in which various events
or faults can give rise to a hazard. When constructed the tree can be
used to see where the most likely causes of an incident lie, and where
additional precautions can be introduced to minimise the risks.
For the most rigorous examination it is necessary to allocate
probabilities to each event in the fault tree, allowing the total
probability of the final event to be calculated (HAZAN).
Where companies are not able to carry out such examinations of their batch
processes alone, they can call on the services of consultant practitioners
to assist them.
Division
ABSTRACT. The circumstances leading to accidents are often very complex but most of them
could have been foreseen by the use of laboratory tests, hazard analysis and chemical reaction
engineering techniques. In this paper, different approaches to improve the safety of chemical batch
processes and storage tanks will be studied, as well as emergency procedures to minimize the
effects of a thermal excursion of the reaction mixture. The objective is give a global vision of the
diversity of aspects that must be covered and the basic concepts to deal with them.
1. Introduction
The rapid development of the Chemical Industry in the past decades has increased the complexity
of chemical plants, and the diversity of products. In parallel, there has been a corresponding
increase in the number of accidents and therefore, the quantity of human losses, material damages
and environment impact has augmented.
The study of accident case histories [1, 2] shows that the circumstances leading to accidents are
often very complex, involving human error, insufficient knowledge about the chemistry of the
process, poor training of the operator, inadequate instrumentation, etc. but it also shows, that the
accident probably could have been foreseen in a high percentage of the cases, by means of
laboratory tests, hazard analysis, and chemical reaction engineering techniques.
Loss of thermal control due to undesired or poorly controlled desired reactions, which can lead
to destruction and release of toxic materials, is a Chemical Engineering area, in which the main
contributions are to develop process concepts which prevent the loss of control of the reactions and
countermeasures to protect against runaway events. In any case, prevention and protection against
chemical reaction hazards is based on the understanding of the basic phenomena involved. In order
to achieve safety, the study of four different aspects is vital: the thermo-kinetic phenomena, the
physical and chemical properties of reagents and products, the equipment characteristics, and the
operating conditions.
In recent years, the search for inherent safety has been widely recommended [3]. A process is
inherently safe, in a rigurous sense, when no disturbance whatsover can cause an accident. In
practice, this is impossible to achieve. However, this concept should be an objective in process
design, since considerable reduction in the potential hazard can be reached at this stage, and even
19
A. Benuizi and J. M. Zaldivar (eds.). Safety of Chemical Batch Reactors and Storage Tanks, 19-47.
1991 ECSC. EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
20
the effects of disturbed operating conditions such as cooling system malfunction or agitator
stoppage can be assessed.
A detailed analysis on the probability of an incident and its severity cannot be effected in the
absence of data with regard to the phenomena involved. Then, this is an important condition that
should be fulfilled.
In this paper different approaches to improve the safety of chemical processes will be studied,
as well as emergency procedures to minimize the effects of a thermal excursion. The objective is
give a global vision of the diversity of aspects that must be covered and the basic concepts to deal
with them.
Heat generation
rate
Temperature
21
loss equals the rate of heat generation in unit volume. The upper and lower steady states, of the
intersection between the sigmoid function of heat generation rate and heat removal rate in figure 1,
are stable to temperature perturbations while the intermediate one is not.
According to figure 1, in a thermal explosion it is possible to distinguish the following phases:
1/ Auto-thermal behaviour : Initially, the exothermic reaction is stable and under control. If for
some re.i^ii iiie reaction becomes unstable, that is, it can no longer be held in check by normal
process control, then the temperature will gradually begin to rise.
2/ Initiation: The reacting mass reaches a temperature in which the heat generated is higher than the
heat disipated by the cooling system. Hence, there is a self-heating behaviour with an acceleration
due to the exponential dependence on temperature of the reaction rate.
3/ Acceleration: The reacting mass rises until it reaches a temperature that triggers off the
decomposition reactions, characterised by their high exotermicity and gas production. The
pressure of the system increases suddenly due to gas production and/or vigorous evaporation of
the liquid phase.
4/ Explosion or reaction auto-controlled: If the reactions continue to accelerate, the pressure
reaches the limit of wall resistance of reactor and an explosion occurs. Otherwise, the reaction rate
sometimes can be controlled by reactants consumption or by diffusion rates if the mass transfer
phenomena plays an important role (e.g. oxygen diffusion for combustion reactions).
Representing the evolution of temperature versus time (see figure 2), it is possible to define
different parts. At the first stage the system is stable and completely defined by the initial
conditions. The second and third stages are the early stages of an instability and it may be possible
to restabilize the reaction by taking unusual actions such as emergency cooling, addition of a
supressant, and quenching. At some point in time, no such restabilization method can bring the
reaction back under control. The reaction is said to runaway. The only recourse to avoid pressure
buildup and possible explosion is venting.
Temperature
time
PREVENTION
PROTECTION
Figure 2. Temperature-time history of a runaway reaction.
Depending on the type of reaction involved in the runaway initiation, it is possible to distinguish
between two different cases; the former is a production process which becames unstable, while the
later is an unwanted reaction that goes out of control:
a/ Loss of control of the desired reaction. The behaviour of the wanted reaction may become
unstable by different causes: high reactant accumulation, high sensitivity to impurities,
22
degenerating operational conditions (i.e. poor mixing, too high feed rates, wrong initial
temperature), failure of the cooling system, etc.
b/ Undesired reactions, fundamentally decomposition and oxidation reactions that are unwanted.
The main possible causes are: reactive compounds mixed accidentally (e.g. cooling water that
penetrates into the reactor), the temperature of the reaction mass increases until decompositions are
triggered off, low heat dissipation capacity that even in very weakly active undesired reaction
systems -in the long term- runaway can be produced (i.e. storage tanks).
Another clasiffication on types of thermal explosion was given by Bowes [9] in function of how
the unstable steady state point is reached. The first kind occurs at the point where the stable and
unstable steady states converge (see figures 4 and 5) and beyond which only high-temperatue
diffusion-controlled steady states can exist, and the second kind requires the self-heating system to
be forced through an unstable steady state with the aid of some other heat source for thermal
explosion to occur.
3. Safety measures against a runaway reaction
There are essentially two different types of measures and countermeasures that can be taken in
order to avoid runaway reactions depending on the region of the temperature-time history which
has been reached by the system (figure 3):
- Prevention measures: oriented to avoid situations that can lead to a runaway scenario.
- Protection measures: oriented to stop or to minimize the consequences of a thermal excursion
(region of abnormal behaviour).
/
Off-line^
PREVENTION
Calorimetric studies
Improving plant design
Analytical criteria
Simulation
New synthetic routes
- Instrumentation
On-llne< - Improving control techniques
- Detection of initiation of runaway
s^ Simulation
-Full cooling
Stopping the'
runaway
PROTECTION
Pressure
^ Quenching'
relief
Add an inhibitor
Add cold liquid
Dump
Containment
23
Clearly, the prevention measures are the most desirable ones because they do not affect the
integrity of desired products. These measures can be divided into off-line and on-line. The former
is part of the systematic process analysis before carrying out the process in the plant and aiming at
the obtention of basic data as a prerequisite for understanding the process behaviour and the risk
associated. The later tends to avoid the loss of control of the process when operating and to detect
possible deviations from the safe operation at an early stage.
The proic^iion measures have as an objective to restabilize the control of the reactor, by means
of active emergency actions such as full cooling, fast injection of a supressant or dumping the
reactor contents; or at least to reduce the damage for people, installation and environment, by
means of passive measures such as containment. These measures are taken when the process is
outside of desired conditions, that means when the rate of heat generation has surpassed the rate of
heat removal and the process is getting to a dangerous state.
4. Prevention measures
4.1. OFF-LINE AND ON-LINE TECHNIQUES
The logical way of obtaining knowledge about the process in order to prevent the potential hazard
of thermal explosion, begins with safety tests performed under laboratory conditions, using small
samples. Apart from basic data, such as physical properties and data related to process equipment,
the type of experimental information necessary for the evaluation of the thermal safety of a
chemical process, can be divided into different types [10].
The first type of tests to be performed concern the evaluation of the thermal stability of
substances and mixtures of substances. Typically, the mixed starting materials and samples from
intermediate process phases are investigated using mg. quantities (eg. DTA, Differential Thermal
Analysis).
For information on the desired reaction, heat flow reaction calorimetry has proved to be an
appropriate tool. The method provides information measured under conditions very similar to
industrial situations which permits the gaining of knowledge about the process and the influence of
the operating conditions on its behaviour [11].
Other required experimental data concerns the heat evolution dynamics of secondary reactions.
Unwanted exothermic reactions can be characterised [12] by a small-scale thermal stability test
(e.g. DTA or DSC) using mg. quantities; adiabatic test (e.g. Dewar, ARC, PHI-TEC); isothermal
test; and test for deflagration. Exothermic secondary reactions are a particularly difficult safety
problem. When a reaction of this type has been established, it is important to determine the
temperature at which this reaction can be observed [10]. When this temperature is more or less the
same or lower than the desired reaction temperature, then it is practically impossible to run the
reaction safely and new operating conditions or synthetic routes must be studied.
The experimental data, obtained by the procedures described above, can be used to apply the
criteria and rules for the safe design of the process (see chapter 4.2) and/or to feed numerical
simulators. The advantage of the former is that it is easy and quick to apply but the information
obtained is relatively reduced when compared to numerical simulation. The advantages of using
mathematical modelling in hazard analysis evaluation are: to interpret the experimental data, to
reduce the number of experiments needed to establish an acceptable degree of understanding , to
predict the dynamic behaviour of the reactor under conditions which are not easily achieved with
laboratory equipment and to perform the scale-up procedure [13].
24
Complementary to off-line techniques there are other types of procedures to prevent the potential
hazard. These are called on-line supervision techniques because they are carried out in real time.
Furthermore to improve instrumentation devices (ie. double sensors for important measures like
reaction mass temperature) and/or to improve control procedures (i.e. introduction of adaptive
strategic); an important on-line prevention measure consists on the detection of potential hazardous
situations with sufficient time in advance to take necessary countermeasures to avoid the thermal
excursion. This can also be done by means of some rule or criteria, or by means of model-based
techniques involving on-line numerical simulation.
4.2. ANALYTICAL CRITERIA
The mathematical modelling of the behaviour of compounds or mixtures that can lead to runaway
reactions has followed two different approaches. On one side, the theory of thermal explosion
which have been applied in the fields of combustion and explosion. On the other side and in
parallel, the concepts developed by this theory has been used in Chemical Reaction Engineering in
order to assess potentially dangerous situations and to design chemical reactors for safe operation.
In this chapter several main points of the thermal explosion theory will be presented. From the
simplest approach (Semenov Theory) that treats a system with uniform reaction mass temperature
and without reactant consumption, to more advanced theories that describe the effects of nonuniform reactant distribution temperature. A more detailed treatment of this theory is available
elsewhere [6,9].
4.2.1. Semenov Model. The Semenov model [4] assumes an uniform temperature distribution
within the reacting system. This is more or less the case of homogeneous systems in stirred tank
reactors.
The theory considers a pseudo zero order exothermic reaction (A > R) with an Arrhenius type
rate equation given by
dC.
A = -r. =-k-C. = - A e
dt
-E.
RT
CA
(1)
^Generated =
R'
= V m-AH R -A-e
RT
-C
(2)
The rate of heat production is mainly governed by the exponential term: exp(-E,/RTj.
Semenov further supposed an uniform temperature of the surrounding T e , which is smaller than
the temperature of the reacting mass. Assuming that there is a cooling jacket, the heat transfer to it in accordance with Newton's rule- is given by
25
^Removed
U-S (T m -T e )
(3)
Hence, the heat dissipated is a linear function of the temperature of the reacting mass.
It is possible to represent the equations in the so-called thermal diagram and obtain three
different states, which are graphically shown in figure 4. They are subcritical, critical and
hypercritical states. In the straight line 1, there are two stationary states in which the heat
generation equals the heat removal rate, the points A and B. The operating point in the low
temperature region is called stable and the other at higher temperature metastable. Any point
beyond the point B intersection represents a runaway condition. At any point below point B the
condition is stable.
There are two ways to affect the equilibrium:
a/ The coolant temperature can increase while the heat removed slope remains practically parallel.
If this occurs, the heat removed line moves parallel to itself until it becomes tangential to heat
generated line in the critical state and until there are no intersection points between both lines
(hypercritical states).
Therma
flow
cr
cr
Temperature
^
Figure 4. Semenov plot traces relationship for heat generation and removal. Influence of the
cooling temperature.
b/ The loss of heat transfer capacity, such as by the loss of cooling, or loss of mixing, will lower
the slope of the heat removed line, although the stationary cooling temperature remains the same
(see figure 5 ).
Thermal
flow
(UA)>(UA) c r
(UA)< (UA) c r
Temperature
26
Figure 5. Semenov plot traces relationship for heat generation and removal. Influence of the heat
transfer parameter.
The situation for the critical state can be described mathematically by equations (4) and (5):
qG= %
(4)
h _ dqR
dT m dT m
(5)
d(
At Tm,.,. the heat generation rate equals to heat removal rate, eq. (6), and also, the slope of the
heat generated line equals to the slope of the heat removal rate, eq. (7).
VmAHRAe
RTm b
" CA =U-S(Tmcr-Te)
^^V^-6
RT
(6)
=US
mCr
(7)
- a
mcr-2R
_ 4RT e
(8)
Eq. (8) shows that an exothermic reaction can lead to a runaway condition only if E a >4RT e , and
also, that ignition phenomena may occur only as long as T e lies in the range 0<T e S Ea/4R. Thus
the maximum value of Tmc,. occurs when T e =E a /4R and is given by Tmcr=Ea/2R which implies at
very high temperatures (i.e. 103 to 104 K). Hence, it is necessary consider only the lower of the
two critical temperatures given by equation (8). Expanding the square root of eq. (8) in a binomial
TA
with b=2,2,4,10,28,..
(9)
27
f 2 3\
' RT
RT?
+
5
T+-^ + 2 E
F
v >;
V a J
"cr
(10)
AH A E . e R T t C ,
a
US
A.
(12)
RT?
It is possible to define <J)cr by substituting the critical condition, eq. (6), into eq. (12) and
rearranging as,
E a (Tn^ Te)
*
J_
Tnw'T,
(13)
RT?
Developing E a /RT m in Taylor series for T m =T e ,
RT
^._^(T
RT
RT?
Te)
^ i . ( T m T e ) . ...
(14)
RT:
truncating the series after the second term (only valid if E a >>RT e ) and introducing the
dimensionless temperature rise:
28
Q
_ E a (T m -T e )
(15)
RT?
it is possible to define
(16)
but substituting Tm,.r by the aproximate equation (11):
(17)
that means, a steady state exists if <t><<t>cr and a thermal explosion will occur if <|>><t>cr. The
conditions at criticality can be encapsulated into the values of two dimensionless groups: the
dimensionless Semenov number which equals to e 1 and the dimensionless temperature excess 0
which is unity.
The advantage of the Semenov model is its simplicity. It gives a clear picture of the occurrence
of a critical state and of the influence of several parameters and variables such as activation energy,
heat of reaction, heat transfer coefficient and temperature. However, in order to describe other
important cases, i.e. unstirred self-heated reactants (for instance, storage tanks), the assumptions
of the Semenov model are not valid and other approaches must be employed.
4.2.2. Frank-Kamenestkii Model. The Frank-Kamenetskii model assumes a non-uniform
temperature (T m ) distribution as a function of an spatial variable [15] and it is based upon the
Fourier heat transfer equation for the conduction of heat in an isotropic medium:
f 2
5T
5t
5T,m
G "Tm
5x
5x
CPn
nr
(18)
5>
+ ^AH,r:
i=l
where the geometric symetry is defined by o as, a= 0,1,2 for infinite slab, infinite cylinder and
sphere, respectively.
Then, for a zero-order reaction:
8T
5t
CPn
at x=xn
52Tm
8x
+ c5Tm^
x
5x
-AHDAeRTmCA
R
(19)
29
and
'm = 0
dx
at
x=0
If dimensionless quantities and approximation given by equation (14) are introduced, eq. (19)
may K transformed into:
,2,
59=j5_e+Jl5e+5el
5x 6z2 z 5z
'
1+Ee
(20)
at
z=l
and
4^-=0
at z=0
dz
where e is the dimensionless ambient temperature, x is the dimensionless time, and z is the
dimensionless distance from the centre :
e=^
(21)
Xmt
1=
(22)
Pm C p m XQ
z=f
x
o
(23)
Frank-Kamenetskii was the first to solve a simplified form of equation (20) for stationary state
(d9/dt = 0) and assuming that e<<l and 0 is not large:
d!e+jrjdi=_5e9
dz 2
(24)
dz
where 5 is the dimensionless heat flow, called Frank-Kamenestkii number, that can be formulated
as,
30
5=
T e )
where Tm,, refers to the temperature in the centre. For a zeroorder reaction this number can be
defined as,
Ea
2
AHREax 0Ae
_
_
5 =
RT
'CA
\
(25)
^.mRTe
As opposed to the Semenov model where the heat resistance only occurs within the substance,
which means that this model results in a temperature gradient in the substance without a
temperature difference across the surface of the reactant. For a symetrically heated system the
temperature gradient in the centre of reacting mass is assumed to be zero, which also means that
the maximum temperature occurs in the centre.
4.2.3. Thomas Model. The FrankKamenetskii model assumes that the reactant temperature equals
to the surroundings at the wall. Apart from the thermal resistance to heat flow due to finite thermal
conductivity, Thomas (16) considered the case where there is heat exchange with the surroundings
following Newton's law. In fact, the Thomas model is the more general and it is possible to derive
the Semenov and FrankKamenetskii models as extreme cases.
The Thomas model is described by eq. (18) with boundary conditions:
^ml
'dTm
dx
^51=0
dx
= U(T
T e )
X=X
at
x=0
where Tm,, refers to the temperature of the reacting mass at the surface.
Introducing the dimensionless numbers, it is possible to obtain eq. (20) with boundary
conditions:
_d0=Bie
dz
^i=0
dz
atz=l whena = 0
at z= 1 when a = 1 or 2
whenz=0
31
where 90 is the dimensionless temperature rise at the surface and Bi is the Biot number, that can be
formulated as,
Bi =
(26)
The relationship between the theories of Semenov, Frank-Kamenetskii and Thomas can be shown
by function of the Biot number. The Semenov and Frank-Kamenetskii models are two extreme
cases of Thomas theory [5].
a) Bi
b) Bi => 1
c) Bi 1
cr
(o+l^Bi
e
32
0.8S
2.0
3.3
{1.19
1.39
and 0
for
1.61
pr
o= 0
0= 1
G= 2
(28)
Bi -> o
5^3.32
2.0
s
2.5
9 C ->1.61 sphere
sphere
Bi - > ~
^ I 7 r , 3 9 cylinder
9 ^ 1 . 1 9 slab
1.0
s ^
cylinder
Bi ->
6 -> 0.88
0.5
e0
slab
1
sphere
slab
15
Bi
Bi
a) The critical Frank-Kamenetskii number b) The critical dimensionless temperatures 9 c (centre) and
5 C as function of the Biot number
e 0 (surface) as functions of the Biot number.
33
Obviously, the ideal situations would be those in which the process is inherently safe [3], that
means, where no disturbances, whatsoever, can cause an incident.
Hugo, and Steinbach have determined some safety rules for assessing conditions under which
batch or semibatch processes cannot be carried out, and for selecting safe operating conditions,
that have been accepted as practical guidelines by Westerterp and co-workers [26]. In the next
chapter some of these criteria will be reviewed, but a more detailed treatment can be found
elsewhere [17,19-23,26] .
4.3.1. Batch and Semi-Batch Processes. Two different discontinuous operating modes are usually
carried out: batch and semibatch. In the first case, the reactor is charged with all reagents, solvents
and catalyst. In the second case, only a part of the reagents is present from the beginning, and the
other part (a reagent or a catalyst) is added.
Reaction vessels for batch operation are normally provided with standard cooling/heating jackets
and/or internal coils, a stirrer, a condenser and, for semibatch operation, a feeding system. The
temperature development is controlled by means of removing the heat generated by exothermic
chemical reactions through a cooled heat transfer fluid that circulates in the jacket or in the case of
working at the boiling point through the removal of heat by evaporation by means of a condenser.
For strongly exothermic reactions, the heat removed capacity may be too small to keep the reacting
mass at the required reaction temperature, in the case of batch operation, and then the heat
generated by chemical reaction has to be reduced. This can be accomplished by adding one of the
reagents during a certain period of time (semibatch). That means, some additional control of the
reacting mass behaviour can be obtained regulating the metering rate.
Different phases [17] can be differentiated in a batch process:
- Phase of heating: The reactor is charged at the starting temperature Tm and the reaction starts. If
the reaction rate is suficiently high, this phase is performed in nearly adiabatic conditions (with the
heating/cooling system switched off),while if the reaction rate is not considerable at the starting
temperature, the content of the reactor is heating following a temperature ramp.
Temperature
Temperature
Phase of
cooling
T i m e
Time
/
u /
a/
b/
Figure 8. Example of different tactics for cooled batch or semibatch processes, a/ Isothermal
operating mode (T^ constant), b/ Isoperibolic operating mode (Tc ^constant).
- Phase of cooling: When the reacting mass reaches a pre-selected temperature, T cooling is
started. After a certain period of time the maximum temperature , Tm is attained. There are
mainly two different operating modes: isothermal or isoperibolic. In the first case, the temperature
34
of the reaction medium is kept constant by regulating the jacket temperature; in the second case, the
cooling temperature is constant (see figure 8).
It is evident that the moment to start the cooling is critical. If it is started too soon (small T,), the
reaction time will be long and the performance will be reduced; moreover, low temperature can
produce accumulation of one of the reagents and consequently the formation of undesirable byproducts (low yield or unstable compounds that can trigger off decomposition reactions). If the
starting is produced too late (high T,) the high temperature obtained can reduce safe operability
margins by triggering off secondary unwanted reactions with the risk of a thermal excursion.
Assuming the uniformity of concentration and temperature in the entire reactor and ideal
disolution [18], the mass and energy balance equations for batch and semibatch processes can be
written as
dt
Q B
2,
S S
k=l
n
dC.
r^i
F.
i=l
dT m _
dt
(29)
j dVm
V m dt
; j = i..,n c
(30)
r nr
i
r
(31)
^ i-1
k=l
An important parameter for safety analysis is the adiabatic temperature rise, ATad, which
indicates how much the temperature of the reaction mixture would rise if the reaction were to
proceed adiabatically (q=0) to completion. Replacing eq. (30) into eq. (31), rearranging and
integrating for the specie A with the initial conditions Tn^Tnjg and XA=0 a t t=0> where %A *S ^
relative conversion of A , defined by XA= 1" CA</CA> it is possible to find that for batch
processes:
C
ad
AQAHR
(32)
pCp
and then, the relation between the temperature and the conversion in an adiabatic reactor is:
Tm = Tm
A
+ A Tad. Y
A
(33)
35
4.3.2. Criteria for safe operation of a Batch process . Hugo [17] has determined some rules in
order to decide if a process could be carried out in batch, or semibatch should be chosen because
the heat removal capacity was too small to keep the reaction mixture at the required temperature.
The analysis was applied to a batch process in which the reaction starts adiabatically at the
temperature TmQ and when the reaction mixture has heated itself to the preselected temperature T
the cooling is started. After a certain period of time the maximum allowable temperature Tm, is
reachpj (.see figure 9).
From the safety point of view, it is only necessary to limit the study at the point at which the
system reaches the maximum temperature, Tmm,and suppose that at this point the rate of heat
generation and removal are equal, then
US(Tm
-Tc):
ffl
RVD
(34)
r(Tm_ )
Temperature
Temperature
o
Time
^ a * ^nax
Conversion
\nax
AT
(35)
ad
This value of conversion represents the fraction of the chemical heat production required to heat
the reaction mixture from T m o to Tmm. The advantage of this formulation is that XA m a x is
independent of the switch-on point, but at the same time is a too conservative value of XAmaxConsidering the following formal kinetic formulation:
r
A = -kCA1.CB(1-XA)n
(36)
36
it is possible to develop the energy and mass balances corresponding to the different operating
phases of the batch process.
For the adiabatic period (Tm<T,):
dt
=-AT ad
(37)
CA
and
T
XA '
-T
AT
(38)
ad
m
dt
-U S ( T - T ) - A T
mm
cc
r
^ C
*A
-fA
dt
CA
(39)
(40)
dividing eq. (39) by eq. (40), it is possible to find the temperature-conversion trajectory:
dI
m
d%
U S C A (T m -T e )
^o
+ A T
rmrA
(41)
dXA
A Tad,
rA^m^)
(T m -T e )
(^nv^'^e)
(43)
37
dT m
T -T
T
-T
ATad
d%A
1-%rtA
rr
i-x A
'Ea(Tnw-Tm)^
exp
(44)
RTmm T,X a i
Hugo and co-workers [19] integrated numerically eq. (44) satisfying the condition given by eq.
(33;, for many different initial conditions of T m and %A and for many different combinations of
T m -T e and %A . and they found the following empiric correlation:
K =
r i-x A . >
n
a-V*)
l
-xA
(45)
-T
(!-XA)AT
(46)
ad
(47)
and
K(3 = P + 0.114
E a (T
";" T l s ) ( p 2 - l . 2 5 2 )
RTi
(48)
where p is the dimensionless temperature difference between reaction mass and heat transfer fluid:
Ea
(49)
Inserting this approximation in eq. (34), it is possible to obtain the following expression,
x)/ max Kp = ( l - x A ) n ( l - ^ ) m
with the two dimensionless variables P and y m ax given by,
(50)
38
RT^us
ma
E a ( - A H R ) k m a x C A C B Vm
0
(51)
By applying eq. (50), the temperature set-point in order to ensure the control of the reaction can be
set. Once, the data set T e , T m 0 , and T s has been established for a specific process, there are some
rules and criteria to be observed. The most important parameter is the temperature sensitivity of the
reactor, S, defined [17] as,
S = - ^
dT
(52)
S takes into account the variation of the maximum temperature respect to any change in the heat
transfer fluid temperature as the critical parameter. This parameter gives a value of the influence in
the reactor temperature of changes in the operating conditions. For example, if S=3 a change of 4
C in the heat transfer fluid temperature would produce 12 C of increase in the maximum reactor
temperature that in some circumstances can be dangerous (i.e. triggering off secondary reactions).
Replacing eq. (36) in eq. (34), derivating in function of T e , and using the approximative
solution, eq. (50), to calculate the derivative of the conversion,
S
(53)
=WHM)
with,
n
2B(l-xAi)V* (1-V*)
(54)
where
B =
ad
(55)
RT
B is the thermal reaction number that gives information about the exothermicity and dynamic of the
reaction. Hugo [17] defined two possible cases: if fx> 1, S is always smaller than 1, and it
decreases when the temperature difference p increases, then, this is the stable cooling range; if
however |0.<1, then S will increase with (J, which means it is possible to go to critical regions
(S>1). Since generally [3>1, the only requirement for maintaining control of the reaction is |J.>1.
From eq. (54) it is possible to define a B c r ; t when u=l and then using the unfavorable case of
^'Amax pl o t the divisory border line between critical and non-critical region (see figure 10).
39
0.0
0.2
0.4
0.6
0.8
1.0
V add.
U-S-t
(56)
V m -Cp-p
40
The system investigated was a 1:1 molar mixture of A and B.The objective is to assess if such a
process can be carried out batchwise with CA=C B() = 4.56 KmoL'm3.
The heat of reaction (AHR) was obtained from adiabatic tests and is equal to -64.9 kj/mol, the
specific heat capacity of the reaction mixture (Cp) is 2512. J/Kg-K, and the density is 937.
Kg/m 3 , hence pCp=2354 kJ/m 3 K. From the same set of experiments a simple reaction rate was
fitted:
'A-*cAo-cBo.(i-3tA)(i-cAo^.)
B
H=
ad
- 5 6 ; 6 4 - 9 - l o 3 = 125.7 C
2354.
From DSC studies it was shown that an unwanted reaction (C > products) can take place, and
, m a x should be established 100 C.
With this data Bcrit, eq. (54), can be calculated,
R
8348.7-125.7
<-A
From the diagram of figure 4.7 [17] it is possible to read % A m a x that should be at least 0.68.
Hence, using eq. (34), the Tm,, can be obtained as Tm,, =Tmmax-%'Amax-ATad=14.0 C.
Another important aspect is the ratio between the surface area and the volume of the reactor at
the critical conditions, eq. (33).
S_ =
V =
P-cp.ATad-k-cB(i-x;)d-c | ^ - )
_2l
U
(57)
( nw- e)
Assuming U=350 W/m 2 K, S/V= 21.4 m2/m3 in this example, For an small reaction calorimeter
(11) this ratio is higher, approximately S/V=31.4 , but for an standard installation it is too high
(S/V between 3-5 m2/m3). This indicates that this reaction may be carried out in the laboratory
with a small reactor (2 1) but if it is transferred to an standard installation a runaway will occur.
In the case of semibatch processes, the control of the heat generated can be done by means of
the cooling circuit or by regulating the metering rate. If the metering rate is too strong, the
maximum allowable temperature can be overshot, but if it is too low, the reaction rate can be low
and the total reaction time too long, with the problem of accumulation of unreacted reagents into
41
the reactor. Figure 11 shows an example of this situation. Normally, the optimum dossing time
can be calculated once the working temperature has been selected by means of,
AHCA
ds
Vm
(58)
US(Tm-Te)
393
n
1320
1
1
1
1
r
1650 1980 23t) 2540 2970 3300
Time (s)
Figure 11. Temperature-time profiles of a semibatch processes, modifying the metering rate.
4.3.4. Extension to heterogeneous systems. A great number of industrial processes involve
systems in which two partially miscible phases coexist, which are in general, organic and aqueous.
However, a considerable amount of incidents, concerning runaway reactions occur in this type of
system [2].
The analysis of reactions taking place between partially miscible liquids is further complicated
due to the mass transfer phenomena between the phases. The reaction usually takes place in one
single phase, which means one compound present in the non-reactive phase diffuses to the
interface, and then into the bulk reactive phase; while diffusing into the reactive phase, it reacts to
form the product. Examples of this type of systems are nitrations, sulphonations, polymerisations,
hydrolisis, esterifications, alkylations, etc.
The formulation described above for homogeneous systems has been extended by Steensma and
Westertep [23,24] for liquid-liquid reactions.
In order to study these type of reactions the following simple reaction was considered:
A + B > products
According to the general theory of mass transfer combined with a chemical reaction [26], it is
possible to define the following two typical situations:
42
Continuous
Dispersed
B
\
A+B products
Figure 13. B is transfered into the dipersed phase where it diffuses and reacts with A. The reaction
occurs only in the dispersed phase.
b/ B is present in the dispersed phase and A is present in the continuous phase (see figure 14).
For these cases expressions for the overall reaction rate can be developed [26], depending on
the controlling phenomena (mass transfer or chemical kinetics) and consequently, some new
variables that did not exist in homogeneous systems have an important influence making the
develoment of the criteria more complex:
- The droplet size, which will determine the interfacial area and thus affecting the overall reaction
rate in case of mass transfer control.
- The distribution coefficient between the phases, which will determine the maximum
concentration of the reactant in the reaction phase.
43
- The phase inversion phenomena which is capable of producing abrupt changes in the reaction rate
during a semi-batch process [25].
Continuous
Dispersed
B
A+B
products
Figure 14. B is transfered into the continuous phase where it diffuses and reacts with A. The
reaction occurs only in the continuous phase.
Thermal
flow
Temperature
Figure 15. Effect of emergency full cooling in stopping the loss of control. The heat removal flow
capacity is increased from 1 to 2.
44
5.1.2. Quenching, a/ Fast addition of a supressant: The method is based on the inerting in situ of
the reactive mixture through the injection of a calculated amount of inert diluent which reduces
reaction rate by cooling the mixture and also by dilution, or a chemical inhibitor which modifies
the reaction rate expression. The application of such a procedure requires that a number of
previous considerations and preparation be made:
- Choice of an appropriate liquid which does not react too exothermally with the reaction mixture.
The addition of an inhibitor requires also intimate knowledge of how the reaction rate can be
influenced
- Sufficient free volume in the reactor
- Technical installations which provide the addition of the liquid in due time.
b/ Dumping of the reactor contents: The reaction mixture is dumped into a vessel which contains
cold diluent with or without an inhibitor. This method requires again good knowledge of the
reacting medium and technical installation to perform the dumping procedure.
5.2. PRESSURE RELIEF
The venting of the reaction mass is the most used safety measure. The main effect of pressure
relief is temperature stabilisation, due to the heat removed by evaporation. However, this system
has three main problems:
- the design of the relief system (vent area sizing and starting pressure)
- the containment of the ejected material
- its ineffectiveness for low vapor pressure systems
The sizing of the vent line is one of the most important aspects in safety of chemical reactors or
storage tanks, since a faulty design could lead to a catastrophic vessel failure with very serious
consequences. The adequacy of a particular relief vent depends on three factors:
- the accuracy in the determination of the "worst case" initial conditions,
- the accuracy with which the thermo-kinetics and physical properties of the reacting mass are
known,
- the accuracy with which the effects of the emergency device can be calculated.
5.3. CONTAINMENT
The objective of a containment system is to keep toxic chemical substances away from people
and environment, to protect the surroundings from missile generation, fires and pressure wave
effects, and to avoid chain incidents (domino effect).
Final remarks
Due to the limited space available, this paper has only presented the main points. Some of the
aspects mentioned here will be covered in more detail by other contributions, however, the cited
literature is strongly recommended.
45
Notation
C
Cp
CPL
D
Ea
F
H
K
KL
k
N
n
E
n
R
n
Q
q
R
R
r
S
T
U
US
V
depends on kinetics
mol-nr 3
JKg-^K" 1
J-mol-^K"1
m7z-s -1'
m
J-mol-1
mol-s"1
J-mol-1
nvs - 1
depends on kinetics
mol
J _I
m-,3.,
-s
W
mol-m_3-s_1
m
J-mol"1-K"1
molm" 3 -s _1
m2
K
W-m-2-K-1
W-K"1
m3
m3-mol_1
Greek symbols
X
Conversion
Thermal capacity,
toichiometric coefficient, reactant(-), product(+)
Thermal conductivity,
Dynamic viscosity,
partial order of reaction
J-K-l
W-nH-K" 1
kg-irr^s" 1
46
e
p
a
T
k gm3
m
Subscripts
0
cr
e
i
L
R
s
w
initial condition
Critical
Heat transfer fluid
Reaction or Input
Liquid
Removed, reaction
Set-point
Wall or Wetted part
ad
E
h
m
r
T
z
Adiabatic
Feed of reactants
Heating
Species
Reaction mixture
radial
Total
Axial
Supersc riptS
continuous phase
dispersed phase
reaction phase
References
1. Rasmussen, B. (1988) 'Occurrence and impact of unwanted chemical reactions', J. Loss
Prev. Process Ind. 1, 92 .
2. Barton, J.A. and Nolan, P.F. (1984) 'Runaway reactions in batch reactors', The Protection
of Exhotermic Reactors and Pressurised Storage Vessels, IChemE Symposium Series, 85, 13.
3. Regenass, W. (1984) 'The control of exothermic reactors', The Protection of Exhotermic
Reactors and Pressurised Storage Vessels, IChemE Symposium Series, 85, 1.
4. Semenov, N.N. (1928), Z. Phys. Chem. 48, 571 .
5. Frank-Kamenetskii, D.A. (1969) 'Diffusion and heat transfer in chemical kinetics', 2nd ed,
Translated by J.P. Appleton, Plenum Press, New York .
6. Gray , P. and Lee, P.R. (1967), 'Thermal Explosion theory', Oxidation and Combustion
Reviews, vol. 2, Elsevier, New York .
7. Merzhanov, A.G. and Dubovitskii, F.I. (1966) 'Present state of the theory of thermal
explosion' , Russ. Chem. Revs., 35 (4), 278.
8. Semenov, N.N. (1959) 'Some problems of chemical kinetics and reactivity' vol 2,
Translated by J.E.S. Bradley, Pergamon Press, London.
9. Bowes, P.C. (1984) Self-heating: evaluating and controlling the hazards, Department of the
Environement, Building Research Establishment, London .
10. Grewer, T., Klusacek, Loffler, H.U., Rogers, R.L. .Steinbach, J. (1989) 'Determination
and assessement of the characteristic values for the evaluation of the thermal safety', J. Loss Prev.
Process Ind., 2, 215.
47
11. Riesen ,R. and Grob, B. (1985) 'Reaction Calorimetry in Chemical Process Development',
Swiss Chem., 7, 39.
12. Grewer.T. (1975), Chem.-Ing.-Tech. ,51,231.
13. Gordon ,M.D., O'Brien, G.J., Hensler , C.J., Marcali , K. (1982), 'Mathematical Modeling
in Thermal Hazards Evaluation', Plant/Operation Progress, 1, 27.
14. Gray, P. , Harper, M.J. (1960) Reactivity of solids, Proc. of the 4th Symp. on the
ReacH', iiy or Solids, Elsevier, Amsterdam, 283.
15. Frank-Kamenetskii, D.A. (1939), Zuhr. Fiz. Khim., 13, 738.
16. Thomas, P.H. (1958), Trans. Faraday Soc.,54, 60.
17. Hugo, P. (1980) 'Start-Up and Operation of Exothermic Batch Processes', Chem. Ing.
Tech., 52, 712.
18. Froment, G.F. and Bischoff, K.B. (1990) Chemical Reactor Analysis and Design, 2nd ed.,
J. Wiley & Sons, Singapore.
19. Hugo, P., Konczella , M. and Mauser, H. (1980) 'Approximation solutions for the design
of exothermic batch processes with indirect cooling', Chem. Ing. Tech., 52, 761 .
20. Hugo, P. and Steinbach , J. (1986) 'A Comparison of The Limits of Safe Operation of a
SBR and CSTR', Chem. Eng. Sci. 41,4,1081-1087.
21. Hugo, P., Steinbach , J. and Stoessel, F. (1988) 'Calculation of the Maximum Temperature
in Stirred Tank Reactors in Case of a Breakdown of Cooling', Chem. Eng. Sci. 43, 8, 21472152.
22. Steinbach, J. (1989) 'Fundamentals/Theory of Runaway Chemical Reactions', Conference
on Techniques for Assessment of Chemical Reaction Hazards, London 5/6 December.
23. Steensma, M. and Westerterp, K.R. (1988) 'Thermally safe Operation of a cooled semibatch
reactor. Slow liquid-liquid reactions', Chem. Eng. Sci. 43, 8, 2125-2132.
24. Steensma M. (1990) ' Runaway and Thermally safe Operation of batch and semi-batch
reactors', Ph.D. Thesis, Deventer.
25. Al-Khudhairy, D., Barcons, C , Hernandez, H, and Zaldivar, J.M. (1989), 'Modelling of
Two Liquid Phases with Chemical Reactions Applied to Toluene Mononitration', Technical Note
N 1.89.57, Commission of The European Communities, Joint Reasearch Centre, Ispra (Italy).
26. Westerterp, K.R., Van Swaaij, W.P.M. and Beenackers, A.A.C.M. (1987), Chemical
Reactor Design and Operation, Student ed., J. Wiley & Sons, Manchester.
Dr. G. DROGARIS
Institute
for Systems Engineering
CEC - JRC - Ispra Site
and
Informatics
1. Introduction
The EEC Council Directive 82/501/EEC [1], or the so-called "SEVESODirective" provides an effective tool for an enhanced accident and loss
prevention policy. This Directive was the response to a series of major
accidents occurred in the last few decades and which attracted the
interest of the public to the hazards presented by industrial installations and highlighted the need for an improved European policy to control potentially dangerous activities.
The accident of 1976 in Seveso, North Italy is mainly considered
as the initiating event for this directive. However it should be
reminded that the disaster of 1974 in Flixborough, U.K. had already
initiated reactions to this same end. Actually the legal frame for the
implementation of this directive in U.K. is based on acts and actions
49
A. Benuzzi and J. M. Zaldivar (eds.). Safely of Chemical Batch Reactors and Storage Tanks, 49-77.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
50
taken according to the recommendations of the Flixborough accident
investigations.
Up to now this SEVESO-Directive has been amended twice. The first
amendment was adopted in March 1987 as a response to the 1984 disasters
in Bhopal and Mexico City and mainly aimed to tightening up the provisions for certain very dangerous chemicals, including chlorine, MIC,
phosgene and sulphur trioxide.
In 1986 the accident in Basel demonstrated the need to strengthen
the requirements of the Directive for isolated storage installations.
The second amendment that was adopted in November 1988 extended substantially the scope of the Directive to storage of dangerous substances and/or preparations at any place. This amendment also attempted
to clarify requirements for risk communication to the public.
The publication cited here [1] contains the consolidated text of
the SEVESO-Directive after the first two amendments.
A fundamental revision of this Directive has been already initiated. It will be based on the experience gained after over five years
of application of the SEVESO-Directive, which came in force in July
1984. Requirements for controls on land-use planning both for new
installations and new urban developments around existing installations
will be included following a council resolution [2].
Here below the objectives and the requirements of the SEVESODirective are presented. Further the two main activities of the Institute for Systems Engineering and Informatics of the CEC Joint Research
Centre at Ispra as part of its support to the Commission for the implementation of this Directive, namely the:
- Major Accident Reporting System (M.A.R.S.), and
- Community Documentation Centre on Industrial Risk (C.D.C.I.R.),
will be discussed.
There are also other EEC Directives related to safety issues as
for example the Directive on health and safety at work and the classification, packaging and labelling of dangerous substances. They will be
presented together with other similar international regulations as well
as rules and guides on the same subject recently developed in the USA.
The role of standards on industrial safety will be also discussed.
SEVESO-Directive as all Council Directives are addressed to the
Member States. This means that they are not automatically imposed to
the national legislation; on the contrary each Member State has to
develop a legislative framework in which the requirements of the Directive will be implemented. A general overview on the national approaches
for the implementation of this Directive with emphasis on the safety
reports of installations is therefore necessary for getting an insight
in the practices for controlling major industrial hazards. Methods of
tackling run away reaction hazards in the safety reports will be dis-
51
cussed separately.
Up to the end of 1990 almost 100 accidents have been registered in
M.A.R.S. Run-away or other unexpected reactions have been identified
among the causes in 17 incidents or accidents. Their characteristics
and the lessons learned from these 17 accidents as well as from 7 accidents involving electrostatic loads conclude this presentation.
2. The SEVESO-Directive
2.1.
Objectives
2.2.
Content of the
SEVESO-Directive
2.2.1. Definitions-Application
of this Directive.
This Directive covers
industrial
activities
where dangerous substances
are processed or
stored. Both terms are defined in Article 1 of the Directive. In general toxic, flammable, explosive and oxidizing substances are considered as dangerous substances. Annex IV the Directive contains indicative criteria for characterizing substances as toxic, flammable, explosive or oxidizing according to their physicochemical properties. In
addition 180 dangerous substances are listed in Annex III with inventory threshold values. Annex II refers to storage of dangerous substances (either isolated or associated to a process plant) and contains
a list of 28 dangerous substances (26 out of these are listed also in
Annex III) with respective inventory threshold values as well as inventory threshold values for the various categories of dangerous substances .
Industrial activities for the purpose of this Directive is considered any industrial installation involving, or possibly involving, one
or more dangerous substances. This implies that manufacturers
(defined
as any person in charge of an industrial activity) should also consider
52
the potential generation of a dangerous substance (e.g. generation of
nitrogen oxides as a result of uncontrolled and/or unintended decomposition of an even not dangerous substance).
Annex I contains also an indicative list of processes and unit
operations that are covered by this Directive (practically all widely
applied processes of organic and inorganic chemicals). This Directive
is not applied to nuclear and military installations, manufacturing and
separate storage of explosives, gunpowder and munition and mines.
Installations for the disposal of toxic and dangerous waste are also
excluded provided that are covered by other Community Acts. Since for
the time being no such Acts exist this exclusion is considered inactive.
The concept of major accident
is rather fuzzy despite its definition in the Directive as: "an occurrence such as a major emission, fire
or explosion resulting from uncontrolled developments in the course of
an industrial activity, leading to a serious danger to man, immediate
or delayed, inside or outside the establishment, and/or to the environment, and involving one or more dangerous substances".
Therefore a scale of "gravity indices" has been adopted for a twoyear trial period [2]. Three gravity indices, ranging in a scale from 1
(worth noting) up to 5 (catastrophic), describe:
- the actual or potential danger (based on the amount of dangerous substance(s) involved);
- the extent of the consequences of the accident;
- the extent of intervention or safety measures external to the industrial activity.
But the scale has not yet been used to define a "threshold" of
gravity for notification purposes.
2.2.2.
Notifications-Safety
Reports.
The manufacturers of installations
are obliged to take all measures necessary to prevent major accidents
and to limit their consequences for man and environment. The manufactures have to prove to the Competent Authorities (see point 2.2.3.
below) that they have identified the major hazards and have taken all
proper safety measures and precautions. These obligations (defined in
Articles 3 and 4 of the Directive) are valid if the inventory of dangerous substances in an installation exceed the lower inventory threshold values specified in Annex II and III. Should the inventories exceed
the higher threshold values then manufacturers have to submit a notification containing information on dangerous substances, the installation, major hazards and safety precautions (safety report). Annex V of
the Directive contains the minimum content requirements for this safety
report. These installations covered by the requirements of the Article 5 of the Directive are the so-called top tear sites.
53
These requirements apply to both new and existing facilities.
Major modifications of existing installations are treated as new
installations.
2.2.3. Competent
Authorities.
The Member States have to set up or
appoint the Competent Authority (or Authorities) responsible for the
implementation of the Directive. The complete list of all Competent
Authorities has been given in [4]. Almost half of the Member Countries
have appointed more than one Competent Authority. Environmental Ministries are the most common Competent Authorities (9 Member Countries),
while the Ministry of Labour has been also appointed as one of the Competent Authorities in four Member Countries. In Ireland and U.K.
autonomous bodies (Health and Safety Executive in U.K., Occupational
Safety and Health in Ireland) are the Competent Authorities.
The Committee of the Competent Authorities (CCA) chaired by the
Commission (DG XI/A/2) and composed of the national Competent Authorities meets regularly every four months and provides a wide forum for
exchange of experience and review of the progress in the implementation
of the Directive.
2.2.4.
Information
to the Public.
The Directive required that "persons
liable to be affected by a major accident originating in a notified
industrial activity within the meaning of Article 5 are informed in an
appropriate manner of the safety measures and of the correct behaviour
to adopt in the event of an accident". Further all this information
disseminated to the public should be made available to the other Member
States concerned.
The second amendment of the Directive underlined the importance of
this information flow to the public requiring that such information has
to be communicated to all persons liable to be affected in case of an
accident without
these
persons
to request
it and adding that this
information has to be repeated and updated at appropriate intervals and
shall also be made publicly available. Annex VII states the minimum
requirements for the information to be communicated to the public.
2.2.5. Major Accidents.
Experience gained from accidents is precious
for an active prevention policy. Hence the Directive requires that as
soon as a major accident occurs the manufacturer notifies the Competent
Authority providing adequate information for assessing the effects of
the accident on man and the environment and measures to prevent recurrence of such an accident.
The Competent Authority in turn has to notify the accident to the
Commission. Minimum requirements to be supplied to the Commission for
major accidents are given in Annex VI.
54
This notification procedure is the basis of the M.A.R.S.
point 2.3. herebelow).
(see
2.2.6. Emergency
Planning.
The Competent Authorities examining the
notification of installations submitted pursuant to Article 5 shall
ensure that internal emergency plans sufficient to cope with potential
major accidents have been drawn by the manufacturer.
Further they shall ensure that an emergency plan is drawn up for
action outside the establishment in respect of the information collected from safety reports.
It shall be clarified that the Competent Authority shall not necessarily draw up emergency plans but shall mainly ensure adequate
information flow to the emergency planning authority(ies).
2.2.7.
Exchange of Information
and Experience.
Article 12 requires that
"the Commission shall set up and keep at the disposal of the Member
States a register containing a summary of the major accidents which
have occurred within the territory of the Member States, including an
analysis of the causes of such accidents, experience gained and measures taken, to enable the Member States to use this information for
prevention purposes".
Information obtained by the Competent Authorities and/or the Commission within the framework of this Directive are considered confidential. This confidentiality requirement, however, "shall not preclude
the publication by the Commission of general statistical data or information on matters of safety containing no specific details regarding
particular undertakings or groups of undertakings and not jeopardizing
industrial secrecy" (Article 13).
2.2.8. Application
Schedule.
The Council Directive 82/501/EEC required
Member States to take the measures to comply with it at the latest on
January 8 t h , 1984. Submission of safety reports of existing installations was to be due by latest July 8 t h , 1989. In the mean time installations covered by the requirements of Article 5 should submit a declaration (name, address, location, type of activity, dangerous substances) latest by January 8 t h , 1985.
Amendment 1 came in force 18 months after its publication (i.e.
September 19 t h , 1988) but safety reports for existing installation subjected to the provisions of this Directive for the first time following
adoption of this amendment were to be submitted not later than
March 19 t h , 1989.
The second amendment applies to new installations latest from
June l"c, 1990 and will become effective for existing installations by
'latest June 1 s t , 1991.
55
Existing installations covered by the requirements of Article 5
for the first time following adoption of this second amendment have to
submit safety reports by latest June 1 s t , 1994.
This directive can be amended by a majority of 45 votes, the votes
of the Member States being weighted as provided for in Article 148(2)
of the Treaty.
2.3.
Major Accident
Reporting
System
(M.A.R.S.)
step
each
56
preted and that all important lessons to be learned from each notified
accident are identified.
Up to the end of 1990, 98 accident notifications had been received
and loaded into M.A.R.S. Though the Directive requires notification of
major accidents only, in practice the tendency is, whenever important
lessons are to be learned from such events, to report also near misses
and incidents that had the potential to lead to serious consequences.
M.A.R.S. enables the Commission to fulfil the requirements of article 12 of the Directive by the preparation of the following documents
at regular intervals:
a) At each CCA meeting, a summary of major accidents notified, indicating causes, consequences and measures to prevent recurrence of similar accidents and/or mitigate the consequences. This report as well
as the content of M.A.R.S. are confidential (see also point 2.2.7.
above).
b) A report on lessons learned from notified accidents containing also
some general statistical data. This report is purged from all confidential information so that it can be made periodically available to
the public. The first version of it has been recently published as
an EUR-report [3].
2.4.
Community
Documentation
Centre
on Industrial
Risk
(CDCIR)
57
tional organizations and industrial or professional associations. Documentation on major accidents in the form of reports and videotapes are
also collected and reviewed".
The documents collected in the CDCIR are classified as follows:
i)
the SEVESO Directive requirements and related issues;
ii) technical guidelines or safety assessments specific to non process/facilities;
iii) technical guidelines or safety assessments specific to a process
or a facility;
accident documentation;
other relevant issues.
This list contains only the five major categories. Details on subcategories and on the CDCIR organization are available elsewhere [7].
iv)
v)
3.1.
Directives
58
tioned above.
Another Council Directive related to prevention of major hazards
and amelioration of their consequences is the 89/391/EEC Directive [9].
Though the objective of this Directive is limited to occupational
safety and health issues there are quif.e a few common issues with the
SEVESO-Directive (accident prevention, hazard identification, personnel
training, etc.); risk communication and consultation issues are however
limited to the labour force of the establishment concerned. Within the
framework of the Directive 89/391/EEC, another Directive [10] containing the minimum safety and health requirements for workplaces has been
already issued and further actions are expected.
Different, minimum requirements are set for new and existing
installations; a 3 years period is given to existing installations for
conforming with the respective minimum requirements.
3.2. Regulations
in USA
The OSHA is also preparing a new rule with reference to safety management of highly hazardous chemicals [11]. This proposed rule, which
makes reference to the SEVESO-Directive, mainly refers to a process
hazard analysis and gives emphasis to managerial issues (procedures,
personnel training, safety reviews, safety audits, incident investigations, emergency planning and response).
The primary requirement for on-site emergency response in USA is
the OSHA regulation 29 CFR 1910.120 "Hazardous waste operations and
emergency response", while the framework for off-site emergency planning is the Title III SARA (the Superfund Amendments and Reauthorization Act) also known as the Emergency Planning and Community Right-toKnow Act.
On an international level there are a lot of initiatives on Major
Accident prevention or related issues.
3.3.
International
Guidelines,
Conventions,
etc.
The OECD is very active on this area and work is carried out from an ad
hoc group of experts and in a series of workshops. Recent relevant OECD
publications include decision-recommendations on provision of information to the public and public participation in decision-making processes, exchange of information on accidents [12], good practices for
accident prevention [13], risk communication to the public and the role
of operators in accident prevention and emergency response [14] and
land-use planning under consideration of major accident hazards [15].
The World Bank has also developed guidelines for identifying, analyzing and controlling major hazard installations in developing coun-
59
tries [16] and has developed a hazard assessment manual which provides
measures to control major hazard accidents affecting people and the
environment. A list of dangerous substances is included [17].
United Nations Environment Programme has also prepared a handbook
on the APELL Process [18] (Awareness and Preparedness for Emergencies
at Local Level). This process is designed to assist decision makers and
technical personnel in developing countries to identify hazards and
prepare emergency response taking into account local conditions.
International conventions for major accident prevention are under
preparation in the following forum:
- United Nations Economic Commission for Europe;
- Council of Europe (compensation for damage resulting from dangerous
activities);
- Nordic Council;
- International Labour Office (Code of Practice on the Prevention of
Major Industrial Accidents).
Codes, guidelines or technical documentation under preparation at
international level include:
- code of conduct to promote economic developments focus on accident
prevention and on emergency response (United Nations Centre for
Transnational Corporations);
- guidelines for disaster prevention including land use and preparedness (United Nations Disaster Relief Coordination);
- technical documentation on chemical spills at sea (International Maritime Organisation);
- a PC-Safety Audit System for identifying, analysing and controlling
major hazard installations (World Bank);
- guidelines on external hazard, human error and common cause failure
(International Atomic Energy Authority);
Other relevant activities at international level worth mentioned:
- study on circumstances and causes of accidents associated with the
release of chemicals (World Health Organisation);
- international programme for the promotion of working conditions and
the Environment-PIACT (International Labour Conference);
- preparation of Environmental Health criteria documents (International
Programme on Chemical Safety);
- studies on lessons learned from emergencies after accidents involving
dangerous substances (EEC-CDCIR);
- studies on environmental accidents (EEC-CDCIR);
- studies on effective provision of public information on major industrial accident hazards (EEC-JRC).
60
3.4. Literature
on Major Accident
Hazards
Consequently there is also a very extensive literature on major accidents hazards. The CDCIR inventory publications [19] give a good
overview of the available material. With reference to guidelines on
hazard and risk evaluations and on safe handling of dangerous substances very active are the H.S.E. [20]. The Institution of Chemical
Engineers, the American Institute of Chemical Engineers, which has also
created a "Center for Process Safety", U.S. EPA and OSHA. It is also
worth noted that manufacturers are giving an increased attention to
safety issues [3,21-25].
3.5.
Safety
Related
Standards
61
the European level at least an harmonization of standards is needed to
remove trade barriers in view of an open European market. For this purpose the European Committee for Standardization (CEN - Comite Europeen
de Normalisation) has been created. CEN is an association of the
national standards organizations of 18 countries of the European Economic Communities (EEC) and the European Free Trade Association (EFTA).
CENELEC (European Committee for Electrotechnical Standardization)
deals specifically with electrotechnical standards.
CEN and CENELEC member countries are Belgium, Denmark, F.R.G.,
France, Greece, Ireland, Italy, Luxemburg, Netherlands, Portugal, Spain
and U.K. (EEC), Austria, Finland, Island, Norway, Sweden, Switzerland
(EFTA).
Their official languages are English, French and German. Membership is open to the national standards organization of any European
country which is, or is capable of becoming, a member of EEC or EFTA.
The principle task of CEN and CENELEC is to prepare European Standards (EN) . Other documents published may be Harmonization Documents
(HD) or European prestandard (ENV).
- An European Standard (EN) is a set of technical specifications established, in collaboration with and with the approval of the parties
concerned in the various member countries. It is established on the
principle of consensus and adopted by the votes of a weighted majority. Adopted standards must be implemented in their entirety as
national standards, regardless of the way in which the member voted
and any conflicting standards must be withdrawn latest by the date
fixed by the Technical Board.
- A Harmonization Document (HD) is drawn up and adopted in the same way
as an European Standard but its application is more flexible so that
the technical, historical or legal circumstances pertaining to each
country can be taken into account.
- An European Prestandard (ENV) can be prepared as a prospective standard for provisional application in areas of technology where there
is a high level of innovation or where an urgent need for guidance is
felt, and where the safety of persons or goods is not involved. The
time required for its preparation is therefore reduced; once adopted,
ENVs are subjected to an experimental period of up to three years.
Members have to announce their existence at national level in the
same way as for EN/HD. However, any conflicting national standards
may be kept in force.
CEN and CENELEC have also a certification task for either the
issuing of an European mark of conformity to standards, or the mutual
recognition of test results and inspection. The CEN framework European
Certification scheme is known as CENCER.
CENELEC has a Marks Committee dealing mainly with the certifica-
62
tion scheme within the framework of the Low Voltage Directive (LVD),
within the CENELEC Certification Agreement (CCA) on the mutual recognition of test results for approval of electrical equipment and household
appliances and the scheme for the marking of harmonized low voltage
cables and cords (HAR agreement).
CEN and CENELEC work on a contractual basis (General Guidelines,
Framework Contract) signed with the Commission of the European Communities; specific mandates for new standards may be also received in the
form of "Order Vouchers" for areas of concern where the Commission
wishes to resolve trade barrier situations which have been notified to
it by the Member States.
A review of the 1989 CEN catalogue [26] showed that only following
specifically safety related European Standards exist:
- EN2
: classification of fires,
- EN3
: portable fire extinguishers (5 standards),
- EN54
: components of automatic fire detection systems
(10 standards),
- EN-132 up to EN-149 : respiratory protective devices and Personnel
eye protection.
63
audits, standards, etc.
The main differences observed in this comparison [27] are related
to the control/acceptance approaches applied.
The Netherlands has made probably the most extensive use of quantitative risk studies of any of the EC countries. There are quantitative ri <;k acceptance criteria approved by the parliament and define
maximum permissible risk (action required of situation unacceptable)
and for negligible risk (no action required) . However, it should be
noted that these acceptability criteria are related to the "external"
safety report which is submitted to the Environmental Ministry and is
available to the public while there is also an "internal" confidential
safety report examined by the Ministry of Labour.
In Ireland and U.K. there is neither a specific requirement for
Quantitative Risk Assessment (Q.R.A.) nor a set of acceptability criteria. Use of Q.R.A. is however not discouraged and is accepted as a support evidence of qualitative conclusions and decisions taken by the
manufacturer. Further H.S.E. in U.K. itself uses Q.R.A. for land use
planning decisions.
In France the laws and regulations themselves are prescriptive and
do not give any quantitative acceptability criteria. No Q.R.A. is
required and the safety analysis has to follow the deterministic
approach. Results of probabilistic analysis, if submitted, are also
judged for the final decision.
F.R.G. represents the other extreme case in comparison to Netherlands. Safety analysis has to use a deterministic approach. The safety
report has to show that safety measures taken correspond to the "Stateof-the-art of safety technology" and has to prove that there is no danger for men outside the plant. Even operators of nearby process
units/facilities of the same establishment are considered as "men outside the plant". However, Q.R.A. is accepted as a support for selecting
among various alternative safety solutions but is never used for a
whole plant or installations.
Contrary to these differences there are also some important points
of convergence such as:
- Though the limitations due to lack of data of a Q.R.A. analysis, this
method is useful for selecting among various alternatives or for
stating priorities.
- Compilation of the safety report by the operator of the plant is very
useful, since substantially contributes to the awareness of the
hazards inherent in the process.
- The safety report is not a stand alone object that is once compiled
and examined. It should be a part of a dynamic process of continuous
inspections from the authorities in order an ever improving level of
safety to be achieved.
64
- Evaluation of safety reports is a multidisciplinary task. This
requires either the assistance of various specialized experts or at
least as an aid very detailed checklist containing the concentrated
experience of various experts.
- Human factors are very important and have to be addressed appropriately in the safety report, evaluation of human factor issues is not
an easy task [33].
4.1.
Run-away/'Side/Decomposition
Reaction
Hazards
Generally batch processes are considered as more problematic in comparison with continuous operation processes since they involve much often
transient operations (start-up, shut-down, loading-unloading, transient
conditions, etc.). It should be noted that in all countries (with the
exception of Spain) hazard identification is required to be performed
for every plant state [27].
With reference to run-away/side/decomposition, etc. reaction hazards the answers of the Competent Authorities can be grouped together
in three categories:
a) No specific requirements because either such a hazard is not present
in any of the installations (Luxembourg) or the Competent Authorities have not yet been confronted with this problem.
b) general requirements such as that the manufacturer has to identify
conditions that may lead to a hazardous incident and hence identify
any risk of run-away reactions. If data are not available in the
literature experimental campaigns should be performed.
c) More specific requirements such as:
- Should a major accident hazard can be perceived to result from an
unwanted or uncovenanted reaction the implications of that scenario are to be investigated and the relevant safeguards described.
This also requires experienced inspectors who can be assisted by
referring to either general rules (e.g. attention to exothermic
steps such as polymerisations, nitrations, Grignards, etc.) or to
data from sources like:
. Brethick (1975) in Lees' Loss Prevention, p. 1080 [32],
. data bank DIMDI in FRG,
. NFPA guide,
. International Process Safety Analysis guidelines (if any).
- Data to be supplied in this case may include:
. Process chemistry, thermochemistry rates of reaction, laboratory
techniques to determine reaction rates DSC, DTA, Dewer Calorimetry, ARC tests,
. details of research work performed (if any),
65
impurity controls,
vessel design,
measures to prevent releases (justification of venting or not,
DIERS systems, venting and scrubbing/blowdown/flaves systems),
other precautions to reduce such risks,
==sessment of consequences of a reactor explosion including risk
to operator/other persons in nearest building/public.
69
20
9
Total
98
66
tions;
d) substances very commonly used (e.g. flammable gases and liquids,
chlorine, hydrogen) are most often involved in the accidents;
e) the vast majority of the accidents notified could have been prevented by proper application of existing experience and diffused
knowledge;
f) managerial/organizational omissions could be identified among the
causative factors in about 90% of the accidents of which the causes
are known;
g) design modifications/improvements were suggested after the accident
in almost 70% of the accidents of which the causes are known.
6.1.
General
Characteristics
11
3
3
Total
17
2
1
1
2
3
1
7
17
67
explosion occurred while frequency of explosion occurrence is less than
45% in the total 98 accidents.
The breakdown of the accidents according to the type of industrial
installation is as follows:
Type of Activity
Other industries
Refinery, petrochemical
Pharmaceutical
Halogen, alkali, etc.
Isolated storage
Accidents
Run-away
All
9
39
2
4
40
15
1
1
7
13
17
98
68
Consequences
None or negligible
Fatalities
on-site
off-premises
Injuries
on-site
off-premises
Material damage
within the establishment
off-premises
Environmental damage
Traffic interruption
Public evacuation
Plant evacuation
Public annoyance
Public deprived from potable water
Accidents
involving
reactions
2 (11.8%)
4 (23.6%)
4 (23.6%)
12
12
3
15
15
6
1
2
2
(70.6%)
(70.6%)
(17.7%)
(88.2%)
(88.2%)
(35.4%)
(5.9%)
1
1
(5.9%)
(5.9%)
(11.8%)
(11.8%)
Tot al
ace idents
21
20
19
2
43
40
8
64
61
13
11
10
7
5
3
2
(21.4%)
(20.4%)
(19.0%)
(2.0%)
(43.9%)
(40.8%)
(8.2%)
(65.3%)
(62.2%)
(13.3%)
(11.2%)
(10.2%)
(7.1%)
(5.1%)
(3.1%)
(2.0%)
69
shore installations to be covered by the requirements of the SEVESODirective since a formal safety assessment is expected to enhance the
safety of such installations [30].
Involving
Unexpected
Reactions
The lessons learned extracted from the accidents in M.A.R.S. are presented in [3] grouped together in the following categories:
1) design/construction-related,
2) operation/maintenance-related,
3) emergency handling,
4) mobilization after and emergency,
5) substance specific,
The lessons extracted from the 17 accidents involving unexpected
reactions are given herebelow; lessons of categories 3 and 4 are presented together.
6.2.1.
Design/construction-related.
a) Control rooms in plants which handle flammable and/or explosive substances should be able to withstand expected blast (blast-proof
design) so that control room operators can take the proper actions
in the case of an emergency.
b) To a maximum possible extent an inherently safe design must be
adopted whenever possible. The application of this rule is illustrated in the case where for vinyl chloride polymerization to PVC:
- a plasticizing agent was used requiring heating by steam at 165"C175C, while latex starts to decompose at 140C;
- there was no means for the detection of the failure of the agitator and, consequently, it was not possible to automatically cut
off the steam supply when the agitator failed in order to avoid
the development of hot spots.
The same principle is also highlighted in another accident where a
by-pass on a control valve was installed without any possibility to
safely interlock this by-pass manually so that no dangerous situation would arise in case of maloperation.
Accident investigation in a third case indicated that process modification in the direction of an inherently safe design could prevent
similar accidents to occur (substitution of the waste water treatment method employed).
c) The creation of explosive mixtures when handling flammable materials
must be avoided using inert substances (e.g. nitrogen). This very
well-known practice could have prevented a series of the reported
accidents (10 out the 98 reported accidents among which one involving an unexpected reaction) . If for a certain case inerting is not
70
considered feasible, technical measures have to be taken to avoid
the presence of ignition sources
(e.g. installation of flame
arrestors in piping connecting a system with a potential explosive
mixture to an environment where ignition sources can not be eliminated) .
d) Proper thermochemical
data and
associated
conditions must
be
obtained for possible run-away reactions before any substance, suspected to cause or undergo such a run-away reaction, is used in a
substantial quantity (e.g. even in pilot-plant scale trials).
e) Provisions for safe disposal or containment of the maximum firewater flowrate, that could be used in the case of emergency, are
necessary to avoid damage to the environment (this has been demonstrated in at least 3 of the 98 accidents one of which involving
unexpected reaction; this later case caused the cut-off drinking
water to 200,000 persons). Attempts to provisionally contain the
fire water led to flooding of the establishment and damage to
machinery and electrical equipment in another case. On the other
hand, there are indications that this particular lesson has been
learned, since in five other recent accidents, the containment and
the safe disposal of used fire-water has been reported.
f) Fire water supply facilities must be designed in such a way that
adequate fire water supply is ensured in the case that even a single
failure occurs in the system.
6.2.2.
Operation/Maintenance
related
a) Extreme care must be taken and procedures must be accurately followed in order to avoid mixing of incompatible chemicals or inverting the order of introduction of chemicals in a reactor. There are 3
among the 17 accidents initiated in such a way; their causalities
were personnel injuries and material damage and in one case one
fatality.
b) Procedures for performing various analytical tests must always be
strictly followed; in one case deviation from the specified procedures masked the identification of a run-away reaction in operating
conditions.
c) Inadequate personnel training has been identified among the causes
of 9 out of 98 accidents notified (2 of them
involving also runaway reactions). However, even experienced operators may make errors
under conditions of excessive work load or due to unclear procedures
as is indicated in other 3 of the notified accidents.
6.2.3.
Emergencies
71
hydrochloric acid concentrations, measured at the plant's fence,
were very low, but the location of emission of these gases to the
atmosphere was close to the air intake of the ventilation system of
an adjacent building. Due to the low levels measured, it was not
considered necessary to alert the population, but concentrated
fumes, sucked in by the ventilation system, affected 29 persons, 11
of whom needed to be taken into the hospital for observation.
b) The effective action of spray/deludge systems, automatically or manually operated, has been experienced in several cases both for
assisting in dispersing the emission of toxic substances (two accidents both involving run-away exothermic reactions) or fighting
fires (three cases one of which involving unexpected reaction) .
There is also a reported case where the activation of a sprinkler
system by the release of hot liquid hexane was not adequate to prevent the explosion of its vapours, since the ignition point was
obviously quite far from the release source. The installation of a
water drench system was decided upon in at least one case after
accident investigation.
c) The application of cooling water, at an adequate rate, to storage
tanks adjacent to a fire can effectively prevent the spreading of
the fire to the content of the tank as has been experienced in two
accidents one of which involving an unexpected reaction.
d) Following an explosion and fire in a chemical factory situated close
to the town (about 2 km away), a blast and smoke caused public concern. The manufacturer soothed people's fear by stating on national
radio and TV that the smoke was not toxic.
e) Delays in getting skilled decision-makers and necessary fire fighting equipment and material at the scene of the accident may contribute to an unnecessary escalation of the accident. Therefore,
they must be accounted for in the emergency planning.
f) Fire fighting intervention teams must be in the position to activate
quickly alternative external fire water sources since the good functioning of the fire water supply systems of the establishment cannot
always be guaranteed.
6.2.4.
Substance
Specific
Lessons
Learned
72
than 7.5. Hence, the use of hydrogen peroxide for waste-water treatment in the presence of volatile flammable substances may be dangerous. Reference is made to an accident, where 1 person was killed and
large material damage was caused by an explosion and fire.
c) Hydrogen peroxide and pyridine: a rapid exothermic reaction may be
caused by excessive hydrogen peroxide addition rate to pyridine.
d) Nitro-organic compounds: this type of compounds is well known as
unstable and subject to run-away reactions. In four of the reported
accidents, nitro-organic compounds were involved.
Inversion of the order of introduction of sulphuric and nitric acid
in the synthesis of 3-methylthioaniline caused the formation of
methylnitrate which, in turn, initiated a run-away reaction leading
to an explosion, which caused injuries to one person and large material damage.
Still unknown catalytic effects (inorganic salts) caused an unexpected run-away reaction in a nitroanthraquinone production plant
leading to an explosion, which caused the death of one person,
injuries to 5 persons and large material damage.
The distillation of a crude product containing also orthonitrobenzaldehyde (which had been formed by the oxidation of this product by
nitric acid) led to a run-away explosion which caused injuries to 2
persons and large material damage.
The run-away explosion during distillation of l-methyl-2-formyl-lnitro-imidazole caused injuries to 2 persons and large material damage.
e) PVC: A run-away explosion in a polymerization reaction which caused
large material damage, was attributed to the combination of the following parameters:
- inadequate ammonia addition, which could not compensate the hydrogen chloride produced;
- latex coagulation due to an excess of hydrogen chloride;
- mixer failure due to latex coagulation;
- no device had been installed to indicate mixer failure;
- local overheating due to no mixing;
- decomposition of latex due to local overheating (steam at 165C175C was used due to the plasticizing agent, while latex decomposition starts at 140C);
- attack of reactor material by an excess of hydrogen chloride.
The reactor burst although the steam supply had been interrupted and
external cooling started due to overpressure and the reduction of
wall thickness from 9.8 to 2 mm.
After this accident, the following was decided:
- the substitution of the plasticizing agent so that steam at maximum 127C can be used;
73
- the installation of warning signals and a steam supply cut-out
system;
- batch analytical controls of the latex pH and its persulphate content are required
f) Raney nickel: pyrophoric nickel material may ignite vapours from
flammable material. Reference is made to an accident where ignition
causing a flash-fire was attributed to this reason.
7.1. General
Characteristics
There are 7 out of 98 accidents for which electrostatic loads have been
identified among their primary causes. Only 1 out of these 7 is a major
accident according to the interpretation of the relevant notifying Competent Authority.
In all seven accidents an explosion occurred; in four of them a
fire was also broken thereafter.
Five out of the seven occurred in pharmaceutical industries, the
other two in installations specified as general chemical industry. All
seven occurred in batch processes.
Review of these accident characteristics suggests similarities
with the conclusion drawn for accidents involving unexpected reactions
(see point 6.1. above).
In 3 out of these 7 cases persons were injured, in 2 of them the
plant was damaged and in one case plant evacuation was necessary. In 3
out of these 7 accidents consequences were negligible. However 6 out of
these 7 cases were incidents that have not been considered as major
accidents.
Only in three cases another primary cause has been identified, too
(component failure in all 3).
With reference to underlying causes following was observed:
- Managerial/organisational omissions
: 7 (100%)
- Lack of safety culture
: 6 (85.7%)
- Insufficient/unclear operating procedures : 6 (85.7%)
- Design inadequacy
: 1 (14.3%)
It should be noted that design inadequacy has been identified for
the only one major accident falling in this category.
Hence these 7 accidents suggest also the same conclusions with
reference to an active accident prevention policy as the ones drawn
from the whole of notified accidents (see point 6.1. above).
74
Caused by Electrostatic
Loads
Design/Construction-related
7.2.2.
Operation/Maintenance-related
7.2.3.
a)
Substance
Specific
Lessons
Learned
75
- exclusion of other ignition sources (e.g. sparks).
Reference is made to five incidents where explosions and fires were
caused by not respecting these rules,
b) Styrene:
- all metal components in contact with styrene must be earthed;
- it is recommended for vessels containing gelcoat (polyester in 40%
styrene) to locate safety valves on the pressurized air supply
pipe and not on the pressure vessel itself (reference is made to
an explosion and fire accident that caused large material damage).
5. References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
76
[10] Commission of the European Communities (1989) "Council Directive
concerning the minimum safety and health requirements for the
workplace - 89/654/EEC", Official Journal of the European Communities L 393/1, 30.12.89.
[11] U.S. Department of Labour - OSHA (1990) "29 CFR Part 1910 - Process Safety Management of Highly Hazardous Chemicals" Notice of
Proposed Rulemaking, Federal Register 5_5_, No. 137, July 17, 1990.
[12] OECD (1989) "Accidents Involving Hazardous Substances" Environmental Monographs No. 24.
[13] OECD (1990) "Atelier sur la prevention des accidents lies aux substances dangereuses. Bonnes pratiques de gestion" Environmental
Monographs No. 28.
[14] OECD (1990) "Atelier sur la communication d'informations au public
et le role des travailleurs dans la prevention des accidents et
1'intervention", Environmental Monographs No. 29.
[15] OECD (1990) "Workshop on the Role of Public Authorities in Preventing Major Accidents and in Major Accidents Land-Use Planning",
Environmental Monographs No. 30.
[16] Batstone, R.J. and Lepkowski, W. (1986) "World Bank Acts to Prevent Chemical Disaster", Technology Review.
[17] The World Bank (1985) "Manual of Industrial Hazard Assessment
Techniques", London.
[18] United Nations (1988) "Awareness and Preparedness for Emergencies
at Local Level. The APEL Process" United Nations Environment Programme. Industry and Environment Office.
[19] CEC-JRC-ISEI (1990) "Community Documentation Centre on Industrial
Risk", ISEI/SER-1899, S.P./I.90.18.
[20] H.S.E. (1990) "Library Information Service - Publication in Series
List" H.S.E. Library and Information Services.
[21] CONCAWE (1989) "Methodologies for Hazard Analysis and Risk Assessment in the Petroleum Refining Industry".
[22] CEFIC (1987) "A Guide to Safe Warehousing for the European Chemical Industry".
[23] LPGITA (1988) "Guide to the Writing of L?G Safety Reports".
[24] Chemical Manufacturers Association "Evaluating Process Safety in
the Chemical Industry".
[25] Chemical Manufacturers Association "Safe Warehousing of Chemicals".
[26] CEN (1989) "Catalogue" Ed. 2.
[27] Amendola, A. and Contini, S. (1990) "National Approaches to the
Safety Report. A Comparison" CEC-JRC-ISEI/SER, S.P. ISEI/SER 1761.
[28] H.S.E. (1987) "Dangerous Maintenance" ISBN 011 8839578, London.
[29] Shortreed, J. (1990) "Recent Advances in Research and Future
Requirement", Plant/Operation Progress, 9_ (3), 198.
77
[30] Boniface, A. (1990) "Piper report set to reform", The Chemical
Engineer, No. 485 (15.11.90), 15.
[31] Smokas, J. (1988) "The Role of Safety Analysis in Accident Prevention", Accid. Anal. & Prev., 1Q (1), 67.
[32] Lees, F.P., (1983) "Loss Prevention in the Process Industries",
Vol. I and II, Butterwort, London.
[33J Amendola, A. (1990) "Human Reliability Models" CEC-JRC-ISEI, Technical Note No. 1.90.23, PER 1840/90.
Paolo CARDILLO
Combustibili
ABSTRACT. In order to avoid conditions for thermic hazards it is necessary to have knowledge of the chemistry and associated thermochemistry
(kinetic and thermodynamic data) of the desired reaction and potential
side reactions and also of the thermal stability and physical properties
of reactants, intermediates and products. The various instruments and
testing methods currently used provide a means of assessing thermal hazard. Over recent years great progress has been made. New methods have
been developed and many new instruments have been offered to the market.
In addition, many institutions have codified the procedural aspects of
safety investigation.
1.Introduction
In any chemical process there is always the danger that the rate of heat
generation will be greater than the rate of heat removal so that the materials will undergo an undesired temperature increase. At this higher
temperature, it will produce even more heat, and the temperature will
increase exponentially in the direction of a thermal explosion.
Essentially chemical reaction hazards are associated with loss of
control of exothermic reactions, gas evolution and/or decomposition phenomena.
Hazardous runaway reactions may occur in all operations in which
chemicals are involved including t 1 ) :
- conversion of chemicals (desired reaction)
- unit operations (drying, grinding, distillation, etc.)
- storage and transportation of bulk chemicals.
Potentially dangerous heat formation can occur in the desired process as well as in undesired consecutive and decomposition reactions
[2,3] .
80
with the reactant accumulation can cause the batch temperature to rise
to a critical level where secondary, unwanted reactions are triggered.
In order to avoid conditions for thermic hazards it is necessary
to have knowledge of the chemistry and associated thermochemistry (kinetic and thermodynamic data) of the desired reaction and potential side
reactions and also of the thermal stability and physical properties of
reactants, intermediates and products. Only a detailed analysis can provide quantitative, reliable data on the probability of an incident and
its severity. The key to success here is initiating a comprehensive analysis at an early stage in project development.
It is not enough that the process steps have been carried out on
laboratory scale without incident, for plant conditions are different.
For example, metals are used in place of glass for the construction of
equipment, pressures and temperatures may be higher, residence times may
be longer, by-products may accumulate in recycle streams, or impurities
may be introduced by substituting commercial for reagent chemicals.
The scale of operation may also be an important factor, particularly in
batch operations.
The assessment strategy is to evaluate two types of hazard. In the
first case, one or more of the reactants, intermediates or products is
inherently unstable and limits have to be determined for the safe operation of the process so the lowest possible temperature at which the onset of decomposition can be detected is not reached. In the second case,
we must consider that any reaction which is exothermic is potentially
hazardous unless the plant used is capable of handling the heat load.
2. Testing procedures
A testing sequence is required that screens processes for potential hazard, defines process
conditions under which uncontrolled reactions
could be initiated, quantifies the consequences of such reactions, and
monitors the margin of safety between the normal operating conditions
and the onset of dangerous exothermic activity.
The questions that should be answered by a test program are:
. which process data do we have to know?
. by which methods can we determine the necessary data?
. what can we conclude from the test results?
In order to specify the safe operating conditions, the primary data required are:
1. the rate of heat evolution
2. the cooling capacity of the plant
In order to determine the consequences of a runaway reaction, it
is necessary to determine:
1. the total heat of reaction
2. the specific heat
3. the adiabatic A T
4. the boiling point of the mass
5. the temperature range in which secondary or decomposition reactions
can be expected and the heat of decomposition of such reactions
6. the amount and rate of gas or vapor evolution (pressure developed and
'rate of pressure increase)
81
7. effects of mischarging, impurities and errors.
If the intended reaction is exothermic, heat production is unavoidable and the corresponding cooling has to be provided.
Heats of reaction can be roughly assessed by calculation using
heats of formation, and for well-known reactions published data are sometimes available f 4 -9]. For the majority of the processes under developme^1", the reaction heat and the time profile of the heat evolution
rale has to be measured experimentally.
The instantaneous heat evolution rate of chemical process proves
to be an ideal property to measure. Not only it is indicative of whether
and how fast reactions are actually occurring but it is also directly
the risk-related quantity which can be easily transformed into a temperature increase rate for the case of emergency loss of cooling.
An essential quantity for the estimation of the severity of a given chemical process will be the maximum temperature increase to be expected under adiabatic conditions (adiabatic /\T). It allows us to calculate the maximum final temperature in case of a runaway situation.
Comparing this final temperature to the temperature range of secondary reactions and to the physical properties of the reaction mixture
under investigation such as melting point, vapor pressure and boiling
point, yield direct information about the consequences of a runaway
(Fig. 1).
Until a few years ago experimental determinations were all aimed
at establishing the temperature range in which decomposition occurs I 10 15] .
decomp. reaction
A H 800 J/g
desired reaction
A H - 170 J/g
50 C
90 C
ATad
- 45 C
Temperature
Figure 1. Simplified scenario of a thermal runaway.
82
In reality, only very few substances have a clearly defined decomposition temperature. Exothermic decomposition starts at the temperature
at which heat generated exceeds the lower detection limit of the measuring instrument. Thus instruments with high sensitivity will indicate a
lower starting temperature for the same decomposition reaction than instruments with a lower sensitivity [16-19]#
When the decomposition temperature is given, it is essential to
describe the instrument used and the experimental conditions employed
(sample size, heating rate, sample holder, e c c ) .
The temperature at which the initiation of exothermic decomposition is detected during thermal stability tests depends on several factors. These do not only concern the materials properties and the specific decomposition reaction but can include the experimental systems'
characteristics and experimental parameters.
A good example of this is the thermal stability study of the "Seveso mixture". The Seveso accident (1976) has met with great interest in
the scientific world; it has probably never happened in the past that
such an effort was made to understand, from a physico-chemical point of
view, an accident occurring in a chemical plant.
Several authors studied the thermal stability of the mixture and
its components supposed present in the reactor. All available methods of
thermal analysis (1976-1983) were used: simultaneous thermogravimetry
and differential scanning calorimetry (TG/DSC) [20,2i]j differential
thermal analysis (DTA) [22], miniautoclave, Sikarex, Dewar flask [23],
accelerating rate calorimeter (ARC) [24] under different experimental
conditions (dynamic, isothermal, adiabatic, isochoric, and isobaric); in
open platinum sample holder, in air or nitrogen; in closed crucibles of
nimonic, glass or gold.
The most significant results of these studies can be summarized as
follows:
. the atmosphere surrounding the sample and the material of the container does not affect the observed exothermic behaviour;
. the Seveso mixture kept at 160 C does not show appreciable exothermic reactions;
. the experiments in miniautoclave do not show exothermic reactions below 200 C; self-heating of the mixture proceeds at an appreciable rate
only above this temperature t 23 l;
. DSC measurements with an instrument made in 1970 did not show any exothermic behaviour below 230 C t 2 ];
. DSC/DTA measurements with more recent instruments show - both under
dynamic and isothermal conditions - slow and week (about 105 - 125 J/g)
exothermic behaviour starting a 180 C; these phenomena were previously
unknown. These measurements also show other, known and more intense, exothermic reactions above 230 C [20-22].
. in the Sikarex calorimeter, under isothermal conditions and with open
sample holder, the self-heating proceeds at an appreciable rate only above 200 C; under adiabatic conditions, with open sample holder at 160
C, the temperature of the mixture increases by about 10 C in four
days; at 180 C only a small temperature gradient can be observed, which
gets exhausted after 24 hours t 2 3 ) ;
'. under heat-accumulating conditions (Dewar flask) at 180 C the tempe-
83
rature increases by 1-2 C only, and the rate of pressure rise is 10
mbar/h I23l ;
. the adiabatic calorimeter ARC confirmed existence of a first exothermic effect at 180 C (/\H 122 J/g) , strong enough to increase the temperature by about 60 C (under strictly adiabatic conditions) and to
start a second, more violent, runaway reaction. The ARC test also showed
that. ^ove 180 C, the mixture gets self-heated, always in adiabatic
conditions, causing a pressure increase of 4 bar (the safety disk at ICMESA was set at this value) in 8 hours I 24 l.
The ability to detect very low rates of self heating is of great
importance. This is illustrated in Fig. 2.
I
I
O)
c
CO
CD
CO
Temperature
T1 T2 T3
84
Fig. 3 shows a typical system of two overlapping reactions. The
main thermal hazard is caused by the major peak. However, the hazard
from this major exotherm is triggered by a very small reaction at lower
temperature. Recognition of the small reaction is of absolutely crucial
importance in understanding and hence controlling the thermal hazard.
However, low sensitivity techniques will fail to recognize its existence.
c
CO
0)
CO
B
r^
I
Temperature
T1
T2
T3
85
c
'+*
0
.cI
"53
CO
Temperature
T1 T2 T3
86
safety consciousness and partly because of the aforementioned automation, which is cutting the cost of such studies.
The test methods included here do not represent a complete "stateof-the-art" but are those of which the author has practical experience.
Self-heating
Pressure
_C
'+*
CO
CD
sz
-P"
I
CD
!
i
4-
CO
CD
W
CO
CD
11 ill IIII
Temperature
Figure 5. ARC self-heating and pressure curves of a liquid monomer.
3.1. Screening tests
Where a large number of samples has to be tested, a fast screening method is necessary. Generally, screening tests determine only the rough initial temperatures for the exothermic decomposition reaction. In order
to fix a safe working temperature from these initial temperatures, safety margins are necessary.
The first step in the identification of reactive chemicals hazards
is the evaluation of the thermodynamic potential of the system. This
will tell first, whether a reaction is thermodynamically possible and
second, how much thermal energy can be released by the reaction. The potential quantity of thermal energy released by the system can then be
related to increases in temperature and pressure within the system being
considered.
One can write balanced chemical equations for the reactions to give "maximum energy release" for which the heats of decomposition can be
estimated [ 3 7 ]. The heat of reaction/decomposition can be estimated by
computer programs and used to predict the possibility of explosion or
decomposition.
Yoshida [ 3 8 ], for example, has studied different combinations of
87
typical chemicals by carrying out computations with the program REITP2
(Revised Program for Evaluation of Incompatibility).
The use of computers as a prediction tool in chemical process hazard evaluation began in 1974 with the introduction of the CHETAH (Chemical Thermodynamic and Energy Release) program ' 39 1 . Since then, the
CHETAH program has been widely used in the chemical industry for hazard
evaluation.
Although its chief aim is to predict deflagration/detonation potential from molecular structure, it can be used to estimate heats of
reaction, heat capacities, and entropies of a pure chemical, a chemical
mixture, or a chemical reaction.
Four criteria of energy hazard potential, based on classical thermodynamics, have been developed and are included in the program.
For the first criterion of energy hazard potential, CHETAH uses
the amount of each element present and thermodynamic data in conjunction
with a linear programming technique to define those products which could
be formed from the reaction mixture and which would release the maximum
amount of energy. The program obeys the laws of thermodynamics and the
principles of stechiometry. For this criterion, the program lists the energy hazard potential as low if the maximum heat of reaction is more
positive than -0.3 kcal/g, as medium if the maximum heat of reaction is
between -0.3 and -0.7 kcal/g, and as high if it is equal to or more negative than -0.7 kcal/g.
The second criterion compares the difference between the heat of
combustion of the compounds in an excess of oxygen and the maximum heat
of decomposition to the maximum heat of decomposition.
The third criterion is based on the "oxygen balance" concept of
Lathrop and Handrix l*i .
If the oxygen balance is more positive than -240 or more negative
than -160, the energy hazard potential is rated as low. If the oxygen
balance is between +240 and +120 or -160 and -80, the energy hazard potential is rated as medium. If the oxygen balance is between -80 and
+120, the energy hazard potential is rated as high.
The fourth criterion is represented by the following equation:
y = 10 A H 2 n a x W/n
where /\Hmax maximum energy of decomposition
W = weight of compound in grams
n = number of moles
If y is greater than 110, the energy hazard potential is rated as
high. If y is between 30 and 110, the energy hazard potential is rated
as medium. If y is below 30, the energy hazard potential is rated as
low.
The program uses pattern recognition to classify the compound as
sensitive or insensitive. The program gives a sensitivity rating of
high, medium or low each of four criteria, and then it gives one overall
rating (Energy Release Potential, ERP) for the compound or mixtures.
The exact details of operation of CHETAH can be found in the literature [39,41-43] .
Because of its ability to predict the potential hazards of a material or mixture solely from a knowledge of chemical structure, CHETAH is
ideal for preliminary hazard evaluation (Table 1).
(kcal/g)
acetonitrile
-0.88 H
HIGH
acrylonitrile
-1.09 H
HIGH
p-benzoquinone
-0.81 H
HIGH
diazomethane
-1.90 H
HIGH
ethylene oxide
-1.23 H
HIGH
nitrobenzene
-1.11 H
styrene
-0.68 M
HIGH
89
For a more detailed examination, these advantages often became disadvantages:
- the small sample size employed make accurate mixing difficult and the
composition of a sample of a few mg taken from a grossly heterogeneous
sospension, is not necessarily representative of the whole mass;
- the observed beginning of an exothermic reaction, i.e. the first deflection from the zero line, is a function of the heating rate, shifting
tov'j.rd lower temperatures at lower heating rates;
- no additions can be made during a run;
- no agitation is possible;
- no measure of pressures generated in sealed pans is possible;
- only a limited range of sealable pans are available.
Recently, a special designed capillary tube and tube holder have
been developed for thermal hazard evaluation using DSC I 47 J. This capillary tube container has several important advantages compared to other
common sample encapsulation approaches. The glass container is inert to
most materials and is capable of withstanding pressures in excess of 200
bar and temperatures up to 500 C. The vaporization effect, often present in DSC of liquids to high temperatures, is negligible because high
sample volume to total volume ratios are possible.
In spite of the difficulties in transfering the DSC results to
plant condition, we think DSC to be a very useful method to in many cases get a quick view of the thermal stability of a substance I 4 8 " 5 0 !.
The rate at which the energy is released is very important: thus a
sharp rise in the rate indicated by a steep slope of the exothermic
shows that the reaction may be hazardous; a broad exotherm peak is indicative of a slower reaction. The area of the peak is proportional to the
energy of the exotherm.
In the dynamic mode (temperature programmed experiment) f 51 l, one
obtains a quick overview of the entire temperature range of interest.
The observed initial temperature of the exothermic response permits a
first estimation of the temperature region in which the undesired reaction must be taken into consideration.
Isothermal measurements t52l in the region of the onset temperature of the exothermic peak usually clarify, in a short time, wheter or
not a decomposition occurs by an autocatalytic mechanism. Key results
from isothermal experiments are heat evolution rates. Since heat evolution rates vary with time, one may simple refer to their maxima, which
occur either at the beginning or, in the quite frequent cases with formally autocatalytic mechanism, after a time delay. Plotting the log of
the maximal heat evolution rates of a number of isothermal experiments
in an Arrhenius plot provides information on the temperature dependence.
The slope of the line is referred to as activation energy (Ea) t 50 !.
3.3.Isoperibolic (quasi-isotheraal) Methods
The sample temperature is increased in pre-defined steps and is allowed
to equilibrate before being monitored for exothermic sel-heating.
The isoperibolic methods (Dewar flask, Sikarex, etc.) [23,32,33]
are generally more sensitive than micromethods and, therefore, make it
possible to determine if exotherms occur at correspondingly lower tempe-
90
ratures. Also, they make it easy to recognize autocatalytic processes.
Large sample sizes are sometimes necessary, which is most disadvantageous when dealing with expensive or toxic substances.
3.4.Adiabatic calorinetry
The Accelerating Rate Calorimeter (ARC) is a commercially available
thermoanalytical instrument in which it is possible to examine the thermal characteristics of a material in a closed environment and in near
perfect adiabatic conditions. The ARC was initially designed by members
of the Dow Chemical Company and was first described by Townsend t28J in
1977. A considerable number of publications have appeared in the scientific literature: the design concept and thermochemical performance have
been described and a number of papers have outlined the basic ARC system
and its operation [24,53-70]_
The adiabatic calorimetry is almost the ideal method for thermal
hazard investigations because the adiabatic course of a reaction presents the thermal behavior under the most unfavorable conditions as far
as the safety is concerned.
There are two main modes of operation of the ARC. The most common
is known as the heat/wait/search mode. A sample is heated to a preset
start temperature. A wait period follows and, once equilibrium has been
established, a search period is initiated. During this time the sample
is held adiabatically. If the increase in the system temperature is below the preset value (0.02 C/min) then the system is heated stepwise to
the next set point. The heat, wait, search cycle is repeated until the
temperature rise during the search period exceeds the preset value.
The ARC can also be used in isothermal mode and this method of operation should be employed where unstable materials are likely to be
held for long periods of time at elevated temperatures.
In the recorded temperature/time and pressure/temperature curves,
all the thermal, kinetic, and physical property data of the reaction
mass are implicitly contained.
The risk of processes with normal residence time of the substances
can easily be evaluated with the helps of experimentally determined adiabatic time to maximum rate of the thermal decomposition.
The list below gives the data which can be obtained from an ARC
experiment:
. adiabatic rate of selfheating vs. temperature
. adiabatic time to maximum rate vs. temperature
. pressure rise or rate of pressure rise vs. temperature
. maximum rate of reaction and heat of reaction (or decomposition)
. activation energy
. pseudo rate constant
If the logP vs. 1/T graph is a straight line, this is likely to be
vapor pressure. If the graph is curved or there are variations then other reactions can be implied. If a pressure is still left in the bomb at
the end of the experiment non condensable gas must have been formed. After suitable correction for the fill ratio (ratio of reaction volume to
freeboard volume), solubility, compressibility and vapor pressures are
made, the quantity of gas generated can be established.
91
3.5. Reaction calorinetry
At all stages in the development of a chemical manufacturing process,
thermodynamic and kinetic data pertaining to each of the process steps
are required. Early in the development process, before a synthetic pathway has been chosen, a reaction step that is obviously highly exothermic may be encountered. A determination of the total amount of heat dissipated in the step would enable a judgement to be made on the thermal
hazard posed by the step. Development work on the step could more easily
be halted at this early stage, before a significant investment is made
in the reaction. A pathway having been chosen, conditions must be identified. Profiles of the instantaneous rates at which heat is dissipated
or absorbed throughout the course of each process step would indicate
the temperature range in which reaction runs to completion fastest, or
in which a runaway is a problem.
The desired reaction can generally be carried out only in a very
definite temperature range. At too low a temperature, the reaction is
either too slow or doesn't occur at all. At too high a temperature, the
formation of undesired by-products often occurs or a region of undesired
consecutive or decomposition reactions is eventually reached.
Many processes providing a high exothermicity of the reaction are
hence designed with a large safety margin (e.g. low reaction temperature
to be far away from the start point of dangerous decomposition). And again very often these preventive measures work against yield.
At the optimal reaction conditions, the maximum rates of heat dissipation and absorption for each step are required to determine the refrigeration and/or heating capacity needed for plant reactors.
Reaction calorimetry allows to simulate plant conditions and to
simultaneously determine actual heat generation data.
A prerequisite for determination of the heat effects is equipment
suitable for dealing with the many varied problems which arise. In particular the reaction enthalpies of chemical reactions should be measured
as stated in the operating conditions. Enthalpies of solution, dilution
and mixing also pose various safety related problems which must be solved.
Reaction calorimetry is not a simple task, since it has to fulfill
a number of sometimes demanding requirements:
- for measuring of the heat of reaction, the investigated chemical synthesis has to be carried-out. Modern organic chemical processes demand
often precise maintanence of the reaction conditions, such as temperatures, dosing rates, pressures, reaction times, weights of added or distilled products, etc. Also the handling of the reaction mixtures is often specifically prescribed and basic unit operations are required,
i.g. stirring, distillation, boiling under reflux, etc.
- as a result of the measurement not only the net reaction enthalpy is
required, but sometimes also various other heat sources have to be considered, i.g. dissipative energy of the stirrer, endothermic process in
the condenser, etc.
- besides heat of the reaction measurement, other information about important engineering parameters are often requested, especially if their
changes influence the interpretation of the calorimetric measurements,
92
i.g. values and changes of the heat transfer coefficient through the
wall of the reactor.
- during a reaction, a number of physical properties of the reaction
mixture can drastically change. A typical example is the change of viscosity due to forming and precipitation of unsoluble product and the
resultanting deterioration of the heat transfer on the inner wall.
Many examples of applications of reaction calorimeters can be
found in the literature [29,30,71-81].
Now we can see some of the most frequent problems in routine experimentation.
4.Selection of the substances or processes to be tested
It is not possible to test every substance for every possible type of
hazard. Obviously, there are intrinsic factors controlling the systematic procedure for thermic safety testing, such as:
- the type of problem/process to be evaluated, i.g. is it a matter of
assessing the hazards of a complete process; is it a matter of determining the permissible storage conditions for a specific product or is it
a simple distillation to be evaluated, etc.
- the structure of the compounds involved in a process
- the instruments available.
A decision as to whether a material should be submitted for test
can be made by consideration of its chemical constitution, the oxygen
balance of the molecule 40J and its behavior in small scale heat tests.
For a molecule containing x carbon atoms, y hydrogen atoms and z oxygen
atoms the oxygen balance Bo is:
Bo = -1600 (2x + y/2 -z)/mol.wt
Oxygen balance for some compounds are reported in Table 2. DSC/ARC
tests have confirmed their instability.
Table 2. Examples of oxygen balance
o.nitro benzyl bromide
o.nitro benzyl chloride
o.nitro benzyl alcohol
o.nitro benzaldehyde
isoxazole
3-amino-5-methyl isoxazole
dimethyl sulfoxide
5-nitro imidazole
1,3-dibromo-5,5-dimethylhydantoin
N-bromosuccinimide
-111.09
-139.87
-151.50
-142.93
-150.72
-163.26
-122.87
-53.92
-61.55
-71.91
It is recommended that materials with an oxygen balance more positive than (-200) should be tested for explosibility. This is an arbitrary choice but is not unreasonable; it is recognized that it errs on
the side of safety (82,83].
93
The presence of certain chemical groups may indicate thermal instability. For example, groups such as nitrate ester, aromatic nitro and
nitramine are closely linked with explosibility; azo, azide, nitroso,
peroxide and acetylene are groups that can form part of explosive structures t1J .
5.Selpf.tion of test method
If the nature of the hazard is not know, it is impossible to decide in
advance which test method wuold be most appropriate. The situation becomes critical when a chosen method would not show up the specific hazard
at all. The choice of method may often be governed by the availability
of the apparatus, by tradition or habit rather than by it's suitability.
The most powerful use of the instruments can be achieved, when
their limitations are known and when the plant conditions to be judged
are borne in mind during interpretation of results. This knowledge allows to apply simple procedures where these are adequate and to use more
extended tests where critical conditions in the plant are relatively
close to the expected real conditions.
Generally, results of safety measurements reflect only the behavior of the material under conditions of the experimental setup and testing procedure used in the instrument. Any other interpretations of the
results, or conclusions made for the manufacturing process, is an extrapolation. Unfortunatly, conditions in production plants are often significantly different from those used in testing instruments. Consequently,
there is a danger from thermal hazards which can remain undetected l 84 l.
Typical examples are: exothermicities masked through endothermic evaporation of solvent remainders, if tested in an open container; or oxidative reactions which cannot be properly detected by tests in closed vessels. Other parameters can effect the detection of the onset temperature:
1. sample size (which affects the extent of heat accumulation within the
sample)
2. thermal inertia (which will cause heat to be absorbed limiting self
heating. The level of thermal inertia is dependent on sample size relative to sample container size and material)
3. vessel material (which can either catalyse or inhibit the reaction)
4. heating rate (which can have an effect on the detection sensitivity,
with higher heating rates raising the detectable onset temperature).
An often-used rule in thermal hazard evaluation which has been
perpetuated throughout the chemical industry is the "100 Degree Rule".
This rule states that if the operating temperature of a process is 100
C away from the nearest detectable exotherm observed in a DSC experiment, the operation will not "experience" this thermal event, and it is
not necessary to obtain more detailed information via a technique such a
ARC. The "100 Degree Rule" failed for a large number of reactions which
commonly occur in the chemical industry [85,86] _
6.Selection of samples for testing
Most safety testing is carried out prior to the start of large-scale
94
production. The problem is to find a sample that will be representative
of the ultimate production material. This often proves difficult, because:
- the material produced in a laboratory may have different impurities
than the production material. And even after production has been started, apparently trivial changes, i.g. in a raw material, can produce a
significant modification of the thermal behaviour.
- some instruments are designed to handle such small amounts that the
sample cannot be considered representative.
- substance properties depend on the process itself. In other words, the
thermal behaviour of the substance changes depending on the treatment
undergone. Substantial differences in properties may be found if the
process conditions vary.
For example, recently in a fine chemicals factory a drum containing about 100 kg of 3-amino-5-methyl isoxazole exploded I 87 !. Samples
from the batch involved in the accident (A) showed a DSC purity of 96 X,
and samples from previous batches (B) gave a purity of 98 %. TLC analysis showed the presence of the isomer 5-amino-3-methyl isoxazole as main
impurity. From DSC and ARC tests, sample A appeared to be much more unstable than the others (ToA = 62 C, ToB = 106 C).
7.Process and plant operation variables
The following indicate some of the variables which must be considered in
a testing sequence l 83 3:
. effects of adding incorret quantities of starting materials, solvents
or catalysts;
. recycle and recovery streams;
. nature of by-products, still residues, waste and effluent streams;
. quality of raw materials, catalysts, solvents and reaction intermediates;
. effect of scale of operation;
. catalytic action by materials of construction of the plant or products
of corrosion;
. ingress of heat transfer fluids in the reactor;
. extended heating times of reactants due to unforeseen circumstances.
95
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
96
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
The thermochemical
Estimation
using
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
ASTM E 487
Townsend, D.I. and Tou, J.C. (1980) Thermochim. Acta 37, 1
Tou, J.C. and Waiting, L.F. (1981) Thermochim. Acta 48, 21
Cardillo, P. and Girelli, A. (1982) Chimica e Industria
64, 231
Waiting, L.F. and Tou, J.C. (1982) J. Thermal Analisis
24, 111
Duch, M.W., Marcali, K., Gordon, C.J., Hensler, C.J. and O'Brien,
G.J. (1982) Plant/Operat.Progress
1, 19
Waiting, L.F..Raykovitz, H.F.andTou, J.C. (1983) Thermochim.
Acta
62, 65
Cardillo, P. and Girelli, A. (1982) Chimica e Industria
65, 611
Huff, J.E. (1982) Plant/Operat.
Progress
1, 211
De Haven, E. (1983) Plant/Operat.
Progress 2, 21
Cardillo, P. and Girelli, A. (1984) Annali di Chimica 74, 129
97
[63]
[64]
[65]
[66]
[67]
[68]
[691
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
ry, Kings'Lynn,
Her Majesty's Stationery Office, London
Cardillo, P.(1988) J. Loss Prevention
1, 46
EQUIPMENT CHARACTERISATION
C. BARCONS I RIBES
Chemical engineering department
Institut Quimic de Sarrid, Barcelona, Spain
Present address: JRCIspra Site
Safety Technology Institute
1-21020 Ispra(VA)
ABSTRACT. The manufacture of chemical products by means of batch processes has recendy become
very important due to the flexibility offered by this type of equipment in which they are carried out. The
proper characterisation of this equipment will allow die performance of a wide variety of chemical
reactions with a minimum risk. The response of the system depends basically on the heat capacity of the
vessel, heat losses, heat transfer availability and other sources of input/output heat. An approach for
determining them either theoretically or experimentally is described in this paper. The procedure to
determine the safe operating conditions is basically the same for every different process. A basic method
for scaling-up a chemical process is described as a result of the proper equipment characterisation. This
paper has been written with the purpose of helping engineers on the characterisation of their batch
chemical plants, in order to avoid accidents when there is a need to scale-up a laboratory process.
1.- Introduction.
The chemical Industry has become important in recent years due to the large number and variety of
chemical processes which have been developed. Human needs and the important evolution of
technology have encouraged the development many chemical processes without a basic knowledge
of the potential risk involved, thus leading to a large number of accidents.
Accidents in the chemical industry are often due to unexpected and/or undesired reactions,
resulting from an over-heating of the reacting mixture and leading to a thermal-runaway. This
happens when the heat generated by the reaction exceeds the heat removal capabilities of the
equipment in which the process is carried out. According to Rasmmusen [1], the largest number
of accidents within the chemical industry occur in batch and semi-batch processes, involving 57%
of cases against 11% in continuous processes.
A significant number of chemical processes, particularly where large scale production is not a
major requirement, are carried out in batch chemical reactors because of their versatility. This type
of equipment often allows the performance of many different processes with minor modifications.
For this reason, batch reactors have become very common in the Fine Chemicals sector.
The standard equipment for this type of processes consists of a vessel, which can be heated or
cooled by means of an external jacket and/or internal coils. An agitator is frequently axially
centered in the vessel and its paddles are placed close to the bottom of the vessel to provide good
agitation of the reacting mixture. External condensers are often available for processes involving
reflux or distillation. The temperature evolution of the reactor contents depends on the heat transfer
fluid temperature, circulation speed and properties. The heat transfer fluid can be temperature
99
A. Benuzzi and J'. M. Zaldivar (eds.). Safety of Chemical Batch Reactors and Storage Tanks, 99-123.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
100
controlled for isothermal processes or may have a constant inlet temperature for isoperibolic ones.
In semi-batch processes, the temperature control can be aided by the rate of addition of reagents
(and their temperature) in order to avoid a large quantity of heat accumulated in the reaction mass.
This equipment is normally designed to work under small pressure. Many different probes and
devices for control and measurement purposes can easily be installed (temperature, pH, redox and
so on) according to the particular process requirements.
Barton and Nolan [2], analysed the incidents which occured in batch reactors due to
overheating of the reacting mass and concluded that these were due to:
- basic lack of knowledge of the process chemistry and thermochemistry
- inadequate engineering for heat transfer
- inadequate control systems and safety back-up systems
- inadequate operational procedures
From this study, it is possible to conclude that a proper equipment characterisation and scale-up
procedure are important aspects in the design of these installations. This paper will cover these
subjects and special emphasis will be given to experimental and theoretical determination of the
heat removal capabilities of a process vessel. Further, the power input due to stirring, the heat
losses to the surroundings and the influence of the addition of reactants will be considered, as well
as the heat capacity of the vessel and internal devices, in order to know how the characteristics of
the vessel can affect the process under dynamic conditions.
2.- Evaluation of thermal capacities and heat losses.
An important parameter relating to a vessel in which a chemical process will be carried out is its
thermal capacity. Normally, thermal capacities of laboratory vessels have an important thermal
effect when a reaction is studied. In process vessels this thermal effect is less important because
the ratio between the thermal capacity of the vessel and the mixture is proportionally much small.
It is relatively simple to determine thermal capacities of vessels. Theoretical prediction is
accurate as they are built-up with well known materials and geometry. There are also easy
methods to perform experiments to determine thermal capacities.
It is more difficult to use available data to find out the value of heat losses as a function of
ambient temperature. It is also problematic to evaluate them experimentally, because of their small
value, however, they can be neglected if the reactor is properly insulated.
2.1. THEORETICAL PREDICTION OF THERMAL CAPACITIES AND HEAT LOSSES.
Chemical reactors are usually made of well known materials. Data describing the physical
properties of such materials is easy to find elsewhere [3].
Theoretical calculation of the thermal capacity of the vessel is made through adding together all
the thermal capacities of the different parts of the vessel in contact with the reaction mixture, which
means the reactor wall and all the probes and devices, such as stirrer, temperature, pH and so on.
n
(2-1)
101
The prediction of the heat losses to the surroundings is more difficult to calculate than the
thermal capacity of the vessel. It will mainly depend on reaction mixture properties (vaporisation
and condensation of the mixture on the reactor walls), air circulation around the vessel (natural
convection) and on solid parts in contact with it (conduction). The importance of heat losses
should be checked and the possibility of a thermal insulation considered. If the reactor is insulated
and working conditions are not drastically far from ambient temperature, heat losses can be
ne^cctea, otherwise complex methods to evaluate them are described by Kumana/Kothari [4] and
different approaches can be found in many heat transfer handbooks [5], [6].
2.2. EXPERIMENTAL EVALUATION OF THERMAL CAPACITIES AND HEAT LOSSES.
Experimental evaluation of the whole thermal capacity and heat losses of a process vessel can be
done in several different ways using the energy balance for a closed system.
2.2.1.- Electrical Heating + free cooling. Heat a known amount of pure liquid which has a very
constant (or very well known) specific heat capacity in the temperature range in which is going to
be used, i.e. water, with a constant power input and no cooling services.
Allowing a free cooling after the heating operation, time constant for heat losses can be
determined. Heat losses can easily be influenced if the mixture is stirred. For an approximative
calculation, free cooling without stirring can be allowed.
A combination of both effects using a non steady state temperature evolution analysis allows to
calculate both values accurately at the same time if accurate data of temperature evolution is
available. Using the energy balance of the system and assuming that the whole system is in
thermal equilibrium, the equation involved for this calculations is the following one:
dTm
m
d t
q"i
r, +
+ r
r
r.
t
+T
m
j _ ( T - T
T -1
x.'
I
(2.2)
102
measured experimentally, is due to the thermal capacity of the whole equipment in direct contact
with the liquid. The first temperature inflexion is reached in a short time depending on the time
constant for the response of the system. At this moment, the vessel wall and the fittings in contact
with the liquid have reached the thermal equilibrium. The value of this temperature difference
allows the calculation of the thermal capacity of the reactor wall and inserts, which can be called
heat transfer barrier. This evaluation needs also no heat transfer fluid present inside the reactor
jacket.
n
rrVC-|Ti-T) =
P: ^ i
) ~
(2.3)
i=1
where: i is refered to each amount of liquid.
T is the theoretical final temperature.
The amount of liquid used should be enough to be in contact with the whole heat transfer area at
the end, and the temperature difference between both amounts before the injection should be as
large as possible. This method is reliable if heat losses are neglectible compared with the amount
of heat involved in the thermal changes of the system, and if thermal equilibrium is reached in a
short period.
2.3. APPLICATION EXAMPLE [8].
For a 100 L Pfaudler standard glass lined batch reactor [7], insulated from the surroundings, the
theoretical thermal capacity using material properties available data can be calculated as follows:
glass =>
metal =>
n y C p g = 3.6-800(Kg-J/Kg/K) = 2.9(KJ/K)
metal =>
m m - C p m = 72-500(Kg-J/Kg/K) = 36.(KJ/K)
103
Comparing these values, it is possible to conclude that the theoretical prediction of the thermal
capacities is rather accurate in accordance with the values experimentally found, taking into
account that the thermal capacity of the inserts was not calculated.
3.- Calculation of heat transfer coefficients
3.I.LNIKODUCTION
Almost all the operations carried out in chemical reactors involve either the production or the
absorption of heat. The knowledge of the equipment capacity for heat transfer is very important
for the characterisation of the dynamic behaviour of the system and is, therefore, of great
importance for the safe and economic design of the process.
Heat flow is defined as the speed at which heat is exchanged from a hot source to a cold
receiver. Both systems are usually considered indepently because their heat transfer dynamics are
different There are three main distinct mechanisms in which heat can be exchanged: conduction,
convection and radiatioa
Many different types of vessels and cooling/heating systems are used in chemical processes.
For batch chemical reactors, external jackets and internal coils are the most common methods to
control the reaction mixture temperature. This chapter will deal with jacketed vessels, but all the
items described hereafter are also applicable with slight modifications for coils and for other
similar types of equipment.
Heat transfer in a stirred chemical reactor with an external jacket can be represented in a
generalized form as is shown in figure 3.1.
dt
L
m
ff
i= l
'
removed
(3.1)
104
Newton's law defines the heat removed through a wall as a proportional part of the temperature
difference between both sides, which is called driving force (3.2). This proportionality involves
the overall heat transfer coefficient, U, and the heat transfer area, S):
^removed = U-S-(Tm - T e )
(3.2)
The magnitude of the driving force depends on the operating mode. There are three main
modes: adiabatic, isoperibolic and isothermal. In the adiabatic mode, no heat exchange occurs
between reaction mass and surroundings, that means qremoved = 0- The isoperibolic mode is
characterised by a constant inlet temperature in the jacket, whereas in the isothermal mode the
reactor temperature is controlled by means of adjusting the jacket temperature. In the latter case,
the dynamics of the heating/cooling circuit becomes important.
For jacketed chemical reactors, the main heat transfer phenomena is forced convection, which is
produced either by stirring effects inside the vessel or by the circulation flow of the heat transfer
fluid through the jacket. Heat conduction through the wall may be important, depending on the
material of which the reactor is made.
The exchanging of heat by means of forced convection can be described simply with the
following equation:
dq = h-SdT
(3.3)
The proportional constant h is referred to as the partial heat transfer coefficient. For a jacketed
reactor, h is applied to the internal and external sides of the reactor wall. The internal heat transfer
coefficient depends on the reaction mixture properties and on the stirring characteristics while the
external depends on the heat transfer fluid properties and its fluid-dynamics regime inside the
jacket. These coefficients must be evaluated experimentally [9].
The final parameter is the heat exchange area. This is practically constant for batch processes
because its changes are due only to density variations in the reaction mixture. However, for
semibatch processes the heat transfer area will increase as a function of the feed rate. Further,
there is also an increase of the heat transfer area due to the vortex generation in reactors that do not
have baffles inside, and this will depend on the stirrer characteristics and physical properties of the
fluid [10].
The main objective of this chapter is to show how to predict from dimensional analysis the
overall heat transfer coefficient, and to describe the procedure with which evaluate it
experimentally. After the heat transfer coefficients are modelled or experimentally evaluated for a
small laboratory chemical reactor, a procedure to scale-up the process to a pilot plant and/or to a
larger production vessel is presented.
3.2. HEAT TRANSFER IN AGITATED VESSELS
For a tank within which a chemical process is carried out, the heat exchanged with the heat
transfer fluid is calculated by taking into account all the thermal resistances between the reaction
mixture and the heat transfer fluid. The overall heat transfer coefficient, U, introduced before, is
determined by two different partial or individual heat transfer coefficients and the wall resistance,
or in other terms, the overall resistance to remove heat can be described as the addition of three or
more resistances. These are attributable to:
105
- Internal film from reaction mixture.
- Wall thermal conductivity.
- External film from circulating heat transfer fluid.
The general expression for overall heat transfer resistance is:
+ H w +
U" h
o
(3.4)
D h
1 1
Where: hg and hj are the partial heat transfer coefficients related to the internal and external
films respectively, and R w is the wall resistance. The term DQ/DJ is the ratio of areas to take into
account the diameter difference between the inside and the outside of the reactor wall.
It should be noticed that every liquid can produce a layer or deposit of extraneous materials on
the heat transfer surface, which will provoke a time-modification of both partial heat transfer
coefficients. In these cases, the heat transfer resistance increases considerably due to the fact that
this type of materials normally have a lower thermal conductivity. This effect is referred to as the
fouling or dirt factor, and should be avoided or, if this is not possible, taken into account as a
further resistance to heat transfer.
3.3. THE DETERMINATION OF THE INTERNAL HEAT TRANSFER COEFFICIENT
From dimensional analysis, using the Nusselt equation (3.5), it is possible to correlate empirically
the internal partial heat transfer coefficient as a function of operating conditions and mixture
properties by means of equation (3.6).
Nu = f (Re, Pr, Vi)
(3.5)
h 0 = a 0 N a 92
(3.6)
- o - ^ o ^ P ^ V l '
K
are
4 0
Da-Pn
(3.7)
106
When the power generated in the reactor mixture is small, the viscosity number (Vi) can be
neglected, if temperature difference between the mixture and the heat transfer fluid is not so large.
Assuming that OQ has a constant value under the conditions previously mentioned, it is possible
to determine experimentally the value of hg, using the Wilson plots [11]. For every vessel, the
values of hg and (XQ must be experimentally determined in order to derive the characteristic values
for a specific system. The correct values of IIQ and <XQ can be simply obtained using eq. (3.4).
With the same operating conditions in the jacket, R w and hj will not change, and hg can be
modified varying only the stirrer speed. A straight line can be plotted in a graph using eq. (3.8):
- ^
1
^LN"920
a
(3.8)
The experimental procedure to evaluate the overall heat transfer coefficient, U, is described in
section 3.7.
If the stirrer speed powered at -620 is plotted against the reciprocal of U, a graph such as that
shown in figure 3.2 is produced. The slope of the straight line is the reciprocal of <XQ. When the
stirrer speed has no effects on the value of ho (N), 1/U gives the constant value for the wall
and partial external heat transfer resistance due to the heat transfer fluid film.
-T
_ R
"
D l
h.
Na
20
107
Table 3.1- Exponents and standard deviation for different impellers and Re regimes.
impeller
Re range
630
620
640
OlO
0.694
0.42
8<Re<46000
0
1/3
turbine
0.42
1/3
2/3
turbine
8<Re<46000
0
0.14
0.38
1.0
anchor
5<Re<70
1/3
0.29
ar-ch^i
70<Re<600000
0.14
1/3
0.678
0.14
2/3
0.35
anchor
70<Re<600000
1/3
1/3
0.7
0.27
pfaudler
9<Re<55000
0
pfaudler
9<Re<55000
0
1/3
2/3
0.33
0.14
0.65
0.55
gate
12<Re<300000
1/3
0.14
2/3
0.47
gate
12<Re<300000
1/3
s%
3.2
8.4
11.3
4.2
6.9
4.5
9.8
5.6
6.9
Rw =
R0-Lnh/D0)
(3.9)
h1 = a 1 Q e
(3.10)
where: a\ depends on the heat transfer fluid properties and the shape of the jacket
108
From the equation (3.10), and assuming that a\ is kept in such a conditions that there are no
temperature changes, the variation of Q e is the only parameter that can modify the value of the
external partial heat transfer coefficient, due to the film of the heat transfer fluid. The value of a j
is a function of the heat transfer fluid properties and is defined by dimensional analysis with
equation (3.11):
e
a. = 0 11 -^-Pr
D
31
-Vi
41
P.-D,
21
S f f- H
*e
(3.11)
1_
rS
Q,
21
(3.12)
The experimental procedure to evaluate the overall heat transfer coefficient, U, is described in
section 3.7.
A graph similar to the one proposed by Wilson can be produced, with the heat transfer fluid
flow powered to -621 against the reciprocal of U. This also produces a straight line whose slope
is proportional to the reciprocal of ctj, and it gives a constant value when the heat transfer fluid
flow has no influence on U (high Re numbers), compared with the partial internal heat transfer
coefficient and the wall resistance. This plot is shown in figure 3.3.
109
Many different authors [12, 13, 14, 15] have reported studies on Nusselt's dimensionless
equation for different systems through a large range of Reynolds numbers (Re), and have often
reported similar values for the exponents 21 31 an( ^ 4\- Th e typical values for 02 \ 631 and
041 are respectively 2/3, 1/3 and 0.14. The constant 6 J J which is used to multiply the whole
equation have been reported to range from 0.2 to 0.8, mainly varying due to system geometry. In
the same system, it varies with the fluid dynamics regime, increasing at about 40% from the
laminar flow (Re<400) to the fully developed turbulent flow (Re>10000). The exponent for the
Prandtl Number (631) is set to 1/3 in nearly all the main relevant works, and the viscosity number
is usually found to be 0.14, though different values have been reported ranging from 0 for small
process vessels to 0.18 for bigger ones.
3.6. THE DETERMINATION OF THE OVERALL HTC DURING A CHEMICAL REACTION
In some cases it may be useful to know the variation of the overall heat transfer coefficient during
a chemical process. It might happen during a chemical reaction that the physical properties of the
mixture change drastically and, with them, its heat transport characteristics. Under such
conditions, the internal heat transfer coefficient might decrease due to the variation of the reaction
mixture properties, and a larger quantity of heat than expected can be accumulated in the reactor,
increasing the temperature of the mixture. In such cases, undesired reactions may take place and,
in the worse case, reaction mass can lead to a thermal runaway.
In order to avoid this situation, the evaluation of the heat transfer coefficient time dependence
during a chemical process can be done. This technique was proposed for the Mettler RC1 reactor
calorimeter [20], and it has been tested for a simple chemical reaction. Its performance is now
described.
The desired reaction is carried out under normal conditions, and the temperature profiles of the
reaction mass and the heat transfer fluid (also the temperature of the reagents added if the process
is semi-batch) are stored in a computer file. After this, the reaction is performed again under the
same conditions, but adding a well known and constant ammount of heat (qh) to the mixture, by
means of an electric heater, during the whole process. This heat flow input should be small
enough in order not to affect the reaction conditions, (basically the temperature), but it must
produce a significant change between reactor and jacket temperature. Temperature data is also
recorded in this case.
Using the equation for the energy balance (3.1) for both cases, and subtracting one from the
other, it is possible to obtain equation (3.15).
dT ms
dt
dime
dt
(3.13), (3.14)
where: s and c refer to the experiment without and with qh respectively.
no
To develop this equation (3.15), it is assumed that the chemical reaction occurs exactly in the
same way and rate, the heat input by the stirrer and any other different inserts are the same in both
cases, as is the ambient temperature.
US =
dTn
dTm
dt
dt
qh- K / T ' V
AT S -
Tm
AT C
(3.15)
7.00
US (IS0THERMAL. MODE)
US (ISOPERIBOLIC MODE)
US (EXPERIMENTAL DATA)
6.50
6.00
180
360
540
720
900
Fig. 3.4.- Calculations under isoperibolic mode (light line), isothermal mode (thick line)
and from previous and posterior evaluation (straight line).
Ill
This method can be applied under either isothermal or isoperibolic modes. If it is applied under
isoperibolic mode, it should be taken into account that the additional heat input will increase the
reactor mixture temperature. In order not to modify the reaction rate, jacket temperature should be
reduced on the same amount
The value of the constant for the heat losses should be determinated for every process at the
final temperature, using the experimentally evaluated heat transfer coefficient and the energy
balance for the steady state (eq. 3.16). It will be more accurate if all the heat input sources are
experimentally measured.
_ US(Tm - Te) + q a
K
1
(316)
(Ta - Tm)
US =
AT
mr A T ,no'
ATe
AT, .=
In
(3.17)
.
eo.
where: AT[nQ and AT m j are the temperature difference between the mixture and the jacket
before and during the power input (when reactor temperature steady state is reached)
respectively.
Experimental Wilson plots with different liquids were carried out with a reaction calorimeter
using the first technique [16]. A plot for toluene at different temperatures and 5 different agitator
speeds is shown in figure 3.5.
The second technique described has been used to determine U for water in a 100 1 standard
chemical reactor [17]. The results of this experimental OQ evaluation and the Reynolds exponent
are shown in figure 3.6.
112
y . 0.0076 0.0022X
y - 0.0076 * 0.002X
R-0,99
H - 0,96
y - 0,0074 . 0.0022X
R.0.96
y - 0.0072 0.0O25X
R-1.00
y - 0,0067* 0.0031 X H - 0 . 9 9
.5
NaA-.694
1/U
NaA0.7
Figure 3.6. 1/U vs. Na" u '' for water at different stirrer speed.
4. The characteristics of mixing devices.
4.1. TYPES OF EQUIPMENT.
Many different types of stirrers are used to agitate chemical mixtures. Among them, impeller,
anchor, turbine and gate are the ones mainly used. For batch and semibatch reactors, normally
only one stirrer is used and it is generally axially centered in the reactor with the paddles close to
the bottom of it
The anchor agitator is normally operated at a few rpm. It generates tangential flow which can be
accompanied with axial flow depending on the mobility of the product. It is placed in the centre of
the reactor and has the shape of the vessel outline, ensuring a low wall clearance. The height of the
two blades usually ranges from 0.5 to 0.8 of the agitator diameter.
The impeller agitator obtain optimum results at an agitator diameter of 0.5 to 0.8 of the vessel
diameter. It usually rotates at more rpm than the anchor. It is normally arranged in the centre of the
113
reactor, and with a minimum of bottom clearance. The agitator sucks in the product axially and
ejects it radially. By installing baffles in the vessel, the efficiency is considerably improved,
especially in the lower viscosity range. It is considered to be a reliable all-purpose agitator over a
wide viscosity range.
The gate agitator produces a double ring turbulence conveying the product out of the middle of
the blades and deflecting it at the vessel wall to both sides. Suction is produced both from the
re^Cujr DOLtom and from the liquid surface. Moreover, it develops a uniform shearing head above
the entire filling height. It is applied for homogenizing services throughout the whole viscosity
range. This type of agitator is specially suitable for the high-viscous range with good results even
at small Re (Reynolds number) with a relatively low power input
The turbine agitator covers approximately 1/3 of the reactor diameter and can be fitted with a
variable number of blades. The conveying flow is normally ducted to the vessel bottom. It is
especially suitable for dispersing services. Power consumption is only slightly affected by
viscosity increase.
Often, baffles are installed in stirred tanks in order to cut the liquid flow inside the reactor,
allowing a better mixture. The baffles can be placed in many different places and in many different
ways.
The use of baffles is optional but they are efficient and advantageous in mixing throughout the
whole turbulent flow range. Depending on their size and position they more or less convert the
tangential flow into an axial flow. Moreover, they increase the turbulence within the vessel and
reduce the formation of vortices, although this is at the expense of power consumption.
The use of probes and other types of inserts in the reactor will also produce a non
homogeneous flow, and vortex generation (if no baffles are present) can be really different,
depending on the whole reactor geometry.
For homogenizing tasks, all the stirrers can be used, but there is, however, an optimum Re
range for each agitator, where shorter mixing times can be obtained with less power consumption.
The application limits of every agitator depends on the minimum Re number required by means
of the physical properties of the mixture. Baffles are recommended in the turbulent flow range
whereas they may be omitted in the laminar range.
4.2. POWER INPUT BY MIXING.
The calculation of the power input disregarding mechanical loss is calculated using the following
equation:
P = NeN3-d5-p
(4.1)
In the first instance, power calculation is reduced to defining the dimensionless Ne number
(power number).
N e = f( agitator shape, Re, d/D, b/d, H/D)
,4 2)
114
d = agitator diameter.
p = density of the mixture.
Re = Nd2p/|JD = vessel diameter,
b = height of blades.
H = height of the vessel.
Normally the course of the Ne is plotted against Re. The influence of baffles in mixing
processes is more important at large Re numbers, having no effects on power consumption when
Re has small values.
The power transmitted by an agitator to a mixture in agitated vessels has mostly been
determined mechanically by measuring the torque transmitted by the shaft. If it is not possible to
.do this, a calorimetric measurement can be carried out.
With a properly insulated vessel in which heat losses are negligible, the following equation can
be applied to calculate the power given to the mixture:
U-SAT = - ( m m - C p m
+ n
t}-^ +P
(4.3)
In a steady state (dT/dt = 0), and without any external power input, the agitator is the only
power input source. Then, if the U S value is known (previously evaluated), and the AT
generated is measured, it is possible to calculate the power input by means of agitating.
Data for predicting the power given by different stirrers can be found either from the
manufacturing company or from the literature [18].
4.3. HEAT TRANSFER AREA INCREASE DUE TO VORTEX GENERATION.
The stirring speed in chemical reactors does not only modify the internal partial heat transfer
coefficient, but can also produce an important change of heat transfer area. This is applicable for
the vessels that do not have baffles or inserts inside (see figure 4.1). Heat exchange area increases
with stirrer speed, and thus, the proportional factor that multiplies the temperature difference in
Newton's law for heat transfer also increases.
The heat transfer area is defined as the surface wetted by the reaction mixture. It can be
calculated as a function of the volume of the mass added plus the stirring effects (vortex) with the
following expression:
S=S +
B
2(V - VJ
m
- + AS
(4.4)
v
'
115
A theoretical model to predict the increase of heat transfer area due to the vortex generation was
developed [19] using the modified Navier Stokes equations. This model is valid for a steady
rotational laminar flow of a newtonian fluid around a vertical axis, and without radial and vertical
speed. The expression deduced from this equations is:
H v =-
N?RS
(4.5)
2g
= v(2JIR J H
0
(4.6)
116
Experimental results show that this increase in the heat exchange area is not independent of
liquid volume, type of agitator, and physical fluid properties. In order to correct the non-idealities
of such an assumption, the value of N a in the equation (4.5) was multiplied by a factor, f, that
depends empirically on all the characteristics mentioned above. Hence, eq. (4.6) can be write, for
a determined reactor and geometry, as follows:
AS^K^fNf
(4.7)
Po = K -(Re)
-(Fr)
ASV = C1 Re Cz Fr 3 N* g(V m )
(4.8)
(4.9)
V - V
9(V m ) =
(4.10)
where: C i , C2, C3 and C4 are constants that were calculated experimentally for four
different agitators using several different fluids and modifying their volume, the temperature and
the stirrer speed [21].
5.- The Scaling-up of batch reactors.
5.1. INTRODUCTION.
In the Chemical Engineering Industry, one of the main problems to solve is how to carry out a
chemical reaction in a production plant under the same conditions as in the laboratory. The
manufacturing of a product requires large vessels in order to be economically viable for the
company, which can lead to unforeseen problems in the laboratory, i.e. heat released, mass
transfer.
117
There is a procedure which should be followed through the laboratory experimentation to the
industrial production which can contribute substantially to safe reactor design thus avoiding
accidents.
This procedure concerns the full calorimetric study of the mentioned reaction, beginning with
small amounts of reagents (for example using DSC or adiabatic calorimetry), followed with bigger
quantities (i.e. using heat flow calorimetry), and scaling-up from this reactor to the plant one,
s o ^ n m e s also going through a pilot plant reactor stage before producing the final design.
Temperature control of the reaction vessel is essential for the safety and performance of a given
process. The proper knowledge of the heat transfer coefficients play an important role for a
satisfactory scale-up procedure, because often it is necessary to ensure that the temperature
evolution in the large vessel will be the same as that in the laboratory one. Thus, once the thermal
response of the vessel and the heat of reaction (from calorimetric studies) are known, the scale-up
procedure can be applied. For the first larger runs, a large safety margin should be conserved in
order to account for any unexpected behaviour of the reaction mixture during the process.
5.2. RULES FOR BATCH AND SEMIBATCH REACTORS SCALE-UP.
The two main phenomena in which special care has to be taken for scaling-up from laboratory to
plant reactors are heat and mass transfer. The main parameter for the former is the overall heat
transfer coefficient, while for the latter the interfacial area through which the diffusion of chemical
species takes place is of great importance.
The mass transfer phenomena will not be considered in this chapter, but the most important
thing to remember is that the interfacial area per unit volume should be constant in both reactors
for the right scale-up of a process. If the main limiting factor for a chemical reaction is the
diffusion of the species from one phase to the other, the reaction rate will change in the same
proportion that it has changed the interfacial area per unit volume.
5.2.1.- Scaling-up to a geometrically similar plant reactor. For this case it is necessary to conserve
geometrical similarity (shape factors, i.e. relations between diameters and height, blades, inserts)
and thermal conditions (reactor and jacket temperature).
It is not so difficult to design a bigger reactor geometrically similar to the laboratory one, but for
the overall heat transfer coefficient approximation, a complex experimental work should be
performed.
It has been defined previously a general procedure to determine the film heat transfer coefficient
for a chemical reactor with equations (3.5), (3.6) and (3.7).
For the plant reactor (P) as well as for that of the laboratory (L), these equations are applicable.
If Sjop is not known but the vessels are geometrically similar, the ratio of Nusselt analogies of the
plant and laboratory reactors may be applied. Since the properties of the reaction mixture remain
constant for the same process at the same temperature, this ratio can be drastically simplified.
Rearranging of the equations and using hoL determined with the laboratory reactor, hQp for the
plant reactor is calculated at the same temperature using equation (5.1):
fD.VcO
h
0P = V '
vS
4/3
V
,
2/3
'Vip>
Vi.
0.14
(5.1)
118
In the case of geometrically similar reactors and if hydrodynamic conditions remain the same
(Newtonian fluids), scale-up involves only a change of heat transfer area and the results can be
viewed with confidence.
With the procedure presented before it is enough to scale-up a process even if no data is
available for the plant reactor.
5.2.2.- Scaling-up to a non geometrically similar plant reactor. More accurate calculations should
be performed if possible to use the material properties. With this information there is no need to
have geometrical similarity between reactors. The use of Nusselt's relation for the film inside the
reactor can be expressed in the following manner
4
0L =
dNa
YI/3
V^ c D g
0/3
Na
go
2/3
(5.2)
The first therm involves the geometry of the reactor and inserts (Rn), the second refers to mixture
properties (Mp) and the third is a relation between the stirrer speed and a reference one.
The calculations involve the experimental determination of 6 J O L by means of the Wilson
method, using a low viscosity solvent with a well behaved boiling point and vapour pressure.
9 J O P is determined following the same procedure used for 0JOL- The overall heat transfer
coefficient U can be experimentally determined as it is proposed for heat flow calorimeters [ 16]
and described in section 3.7.
U
9
10P
n3Mp^[ Na/Na,
- ^
1
1P
(5.3)
For the development of this equation it has been assumed that 620 = 2/3, 630 = 1/3, and that
the viscosity number can be neglected. The last therm of this equation is called the apparatus
resistance. Now, for the plant reactor, it is possible to get the 9iop value from the slope of the
Wilson plot, because either Rn and Mp are known (Mp can be calculated from the bench scale
reactor Wilson's plot). Once it is evaluated, any unknown Mp can be calculated if a new plot is
done with the same reactor, with the goal of avoiding any possible lack of mixture properties data.
For accurate data acquisition, the vessel should be insulated from the surroundings. Viscosity
data is necessary at the reaction temperature and at the estimated wall temperature, if there is a
large temperature difference between reactor and jacket temperature.
The best results are obtained if the linearisation with the Wilson plot using a Reynolds exponent
2/3 is good and heat transfer measurements have been accurately carried out in the plant reactor.
119
5.3. DOSING CONTROLLED REACTIONS (SEMI-BATCH).
With the known resistances due to the reaction mixture (h^p) and the apparatus (h^p), the overall
heat transfer coefficient can be calculated. The maximum heating/cooling power can be estimated
with the larger reactor and jacket temperature difference. The heat transfer area can be calculated
using the reaction mixture volume inside the reactor, thus available power (q av ) is obtained by
means of.
qav = U S - ( T m - T e )
(54)
Heat of reaction and proper reaction temperature are known approximately from laboratory
experiments, by means of the different types of calorimetric analysis. From the available power
plus data on the heat of reaction, a minimum dosing time for an isothermaly controlled reaction can
be approximated by:
t
dos
A H
qav
(5.5)
This applies for an addition of a liquid at the same temperature of the reaction mixture. It is
possible sometimes to add a warmer or a colder fluid depending on the requirements of the
reaction, having a heating or cooling influence according to the needs of the process. The heat
involved in this case should be added to the available power as it is described in equation (5.6):
q a v = U S ( T m - Te ) + q ^ C p d o s - ( T m " T d o s )
(5.6)
Nu = 0 . 2 7 - R e 0 7 P r 1 / 3
(5 . 7 )
120
1 / U - 0.0018 6.074a-4"Na-0.7
R - 0.98
R - 0.99
EI
1/U(25"C )
1/U(19"
C )
1.2
NaM).7
Figure 5.1. 1/U vs. Na~07 for the 1001 reactor filled with 80 Kg of water.
T=19C
7273
7.220
5.219
1375
T=25C
8378
6.154
5.305
1463
Using data from Figure 5.1, rig evaluated for the 100 1 reactor agitating at 50, 100 and 150 rpm
at 19C and 25C are compared with the ones obtained scalingup from the 21 reactor to the 1001
one, by means of equation (5.1). The same exponent for Re used in (5.7) has been used to rise the
relation between stirrer speed for both agitators. The equation used in this case has been 5.1. This
comparison is shown in table 5.2.
Table 5.2. comparison between scaledup/evaluated hg values.
RPM
100
50
150
hO scaleup (19)
1762
3665
2797
3426
hO evaluated (19)
1588
2580
3900
hO scaleup (25)
1875
2976
hO evaluated (25)
2354
3127
1449
These calculations have been carried out considering for the calorimetric reactor d=0.067 and
D=0.114, and for the 100 1 reactor d=0.350 and D=0.508, where d is the stirrer diameter and D is
the reactor diameter.
NOTATION
c : parameters for correlations.
Cp : Specific heat capacity, JKg" l. K l
121
d,D :
f
g
h
H
K
L
m
Mp:
N
Na
Q
q
p
R
Rn
S
T
t
U
US
V
diameters
empirical factor for vortex increase
gravitational acceleration, nvs"2
partial heat transfer coefficient
Molar enthalpy (liquid), J-mol"^ orHeight.m
dimensionless constant or Coefficient of heat losses, W-K"1
laboratory
mass,Kg
mixture properties
Number of species, stirrer speed
Stirrer speed s~*
Volumetric flow, m^-s"!
Thermal flow, W
power, plant
Radius, or resistance
reaction number
Surface, m^
Temperature, K
time, s
Heat transfer coefficient, W-nT^-K'l
Effective heat transfer coefficient, W-K"l
Volume, rrP
Gre ek symbols
a
Time constant, s
Sub s c r i p t s
0
1
a
B
dos
e
g
i
J
m
P
R
r
T
t
v
w
x
122
Liquid, laboratory
Axial
Dimensionless groups
Fr
Ne
Nu
Pr
Re
:
:
:
:
:
Po :
Vi :
Froude number,
power number
Nusselt number,
Prandtl number,
Reynolds number,
for stirred tank,
Power number,
Viscosity ratio
N a 2 -D a -g-l
u-D-r1
ti-Cp-r1
p m -Q-D-S-l-Li-l
Pm-Na'Da2-^"1
qa-Pm^-Na"3^-5
Rjulk^atwall
6.- References.
( 1) Rasmussen, B. (1988) 'Occurrence and impact of unwanted chemical reactions', J. Loss
Prev. Process Ind., 1, 92.
( 2) Barton, J.A. and Nolan, P.F. (1989) 'Incidents in Chemical Industry due to ThermalRunaway Chemical Reactions', Chemical Reaction Hazards, London Press Centre, London.
( 3) Dubbel (1981) Taschenbuch fur den Maschinenbau, Auflage, Springer.
( 4) Kumana, J.D., Kothari, S.P. (1982) 'Predict storage-tank heat transfer precisely', Chemical
Engineering, March, 127-132.
(5) R.H. Perry and C.H. Chilton (1973) Chemical Engineers Handbook, 5th ed., Mc GrawHill, New York.
( 6) McCabe, W.L. and Smith, J.C. (1956) Unit Operations of Chemical Engineering,
Mc Graw-Hill, New York.
(7) Pfaudler reactor Type: AE-100
( 8) Private communication from Dr. Hernandez, H. FIRES Project, J.R.C. Ispra (VA), Italy.
( 9) Kern, D.Q. (1950) Process Heat Transfer, Mc Graw-Hill, New York.
(10) Zaldivar, J.M., Hernandez, H. and Barcons, C. (1990) 'Development of a Mathematical
Model and Numerical Simulator for a Reaction Calorimeter FISIM, RC1 Version', technical
note n 1.90.109, J.R.C. Ispra (VA), Italy.
(11) Wilson, E.E. (1915) Trans A.S.M.E. ,37-47.
(12) Chapman, F.S. and Holland, F.A. (1965) 'Heat Transfer Correlations for Agitated Liquids
in Process Vessels', Chemical Engineering, 1, 153-158.
(13) Chapman, F.S. and Holland, F.A. (1965) 'Heat Transfer Correlations in Agitated Vessels',
Chemical Engineering, 2, 175-182.
(14) Bourne, J.R., Buerli, M. and Regenass, W. (1981) 'Heat transfer and power measurements
in stirred tanks using heat flow calorimetry', Chemical Engineering Science, 36, 347-354.
(15) Brooks, G. and Su, G.J. (1959) Chem. Eng. Prog.,Oct., 54.
(16) RC1 Operating Instructions (1986) Mettler Instrumente AG, CH-8606 Greifensee.
(17) Private comunication from Zaldivar, J.M., FIRES Project, J.R.C. Ispra (VA), Italy.
(18) Uhl, V.W. and Gray, J.B. (1966) Mixing: Theory and Practice, vol 1&2, Academic Press,
New York.
(19) F.A. Holland (1973) Fluid flow for Chemical Engineers, Edward Arnold, London.
123
(20) Hoffman W. (1989) 'Qdos Qflow Qaccu and all these things', RC1 3rd International User's
forum.
(21) Zaldivar, J.M., Hernandez, H. Barcons, C , Nomen, R. and Sempere, J. (1990), 'Heat
Transport for Immiscible Liquids in Agitated Vessels', Proceedings of the 5th Mediterranean
Congres on Chemical Engineering.Barcelona, 1,176-177.
H.J. HERNANDEZ
Joint Research Centre,
Institute for Safety Technology
Process Engineering Division,
21020 Ispra (Va), Italy
ABSTRACT. The application of mathematical modelling and numerical
simulation to hazard analysis is discussed. The principles of
mathematical modelling are described, with particular reference to
batch reactors and storage vessels where runaway reactions can take
place.
Two representative cases are presented in order to show the
applicability and limits of numerical simulation in assessing the
safety of a batch process by calculating critical operating conditions.
1.
Introduction
126
a p p r o a c h , and t o f i n d o u t t h e a p p l i c a b i l i t y and l i m i t s of t h i s
technique
t o t h e h a z a r d a n a l y s i s of b a t c h p r o c e s s e s . F i r s t l y , a
g e n e r a l model i s p r e s e n t e d i n o r d e r t o d e m o n s t r a t e t h e complex
i n t e r a c t i o n i n v o l v e d , and t h e l a r g e q u a n t i t y of d a t a t h a t i s r e q u i r e d .
S e c o n d l y , t h e f o r m u l a t i o n and a p p l i c a t i o n of s p e c i f i c models f o r a
b a t c h r e a c t o r and a s t o r a g e v e s s e l i n which runaway r e a c t i o n s can t a k e
place are described.
2 . Mathematical
modelling
2 . 1 GENERAL MODEL
Consider a system consisting of a mixture of (Nc) chemical species
confined in a volume (Vm) of fixed coordinates.
The state of the system, represented by the temperature(Tm), the
pressure(p), and the concentration of different species (Cj), changes in
time and space according to the chemical reactions taking place, and to
the transfer of mass, energy and momentum. These phenomena are
interdependant, and so must be considered simultaneously. However, for
the systems studied here, the transfer of momentum is not strongly
coupled, and may be determined separately
(e.g. by means of
dimensional analysis).
Let consider first, the characteristics of these phenomena, that
will be introduced into the fundamental equations.
2 .1.1. Chem
i cal
react
i on.
A chemical reaction is characterised by the
kinetic function that expresses the rate of matter transformation,
which depends on the state of the system (see complete description of
symbols in section 4) :
r. = r.(Tm,C.,p)
(1)
Nr
Rj =
;/ , VV H. r.
r(
X ^
*!
; j=l,...,N c
(2)
i=l
127
N . = C. v
J.
; j = l, . . .,N C
(3)
: 3-1/ /Nc
(4)
e = pm (U + l/2v2)v + N w + n v
(5)
3c.
-^
= R. - V N.
; j = l, ..,NC
(6)
Energy Balance
" = -
. V U
- U V . Nw + p V v]
(7)
Equations 6 and 7 are strongly coupled, and their solution needs the
following information:
128
2.2.
PARTICULAR MODELS
2.2.1.Storage
Vessel.
Consider a vessel of symetric geometry specified
by CT, (where <T=0, 1 or 2 for a slab, cylinder, or sphere), containing
an isotropic medium where physico-chemical processes may take place.
The characteristics of such a system permit to assume one-dimensional
transfer of mass and energy.
The mass balance equation is obtained directly from equation 6,
introducing the characteristics of this system: no velocity gradients
and mass flow by diffusion expressed by:
a2c
V N.
3c.
\_
x 3x
3x
(8)
3c.
32c.
3t
3x
3c,
- ^
3x.
;j=l,...,N r
(9)
= o
; j = l, . . . ,NC
(10)
and
pVv/pm ,
3u _ 3H
3t
3t
dp
V-dt
(11)
and neglecting the term Vdp/dt, which is small compared with the
enthalpy term,
introducing the enthalpy change of the mixture and substituting the
concentration variation by the mass equation 9,
introducing the Fourier law for heat conduction,
N.,
A.m V T
(12)
129
AH . =
V. ,H.
;j=l,...,N
(13)
j=i
The t i m e - p r o f i l e of t e m p e r a t u r e
reads:
r
3T
dt
3x 2
Cp
(14)
3x_
i=l
3x
(15)
5Tm
3TP
_ u
" x?~d^
~d^
Xp
(Te
"
(16)
2 . 2 . 2.Stirred
Batch
Reactor.
In a typical batch reactor, the
diffusional phenomena are negligeable, due to the elimination of
composition gradients by the stirring.
The integration of the continuity equation 6 over the reactor
volume Vm, including the mass input from the surroundings, gives the
time-profile of composition for each species j:
dC,
3
"dt
R.
3
^ (F,
dVm
Fc ) - C i dt
;j=l,...,N C
(17)
dt
rm+r
(T
j=l
- T
) - V
r.AH.
(18)
130
3.
Application
cases
3.1.
products
H2C=CH-COOC2H5
CH3-CH2OH
CH3CN
========
H2O2
CO3K2
And e t h a n o l as s o l v e n t
CH2-CH-COOC2H5
CH3-C0-NH2
o2
H 2 02
>
2) CH3-COOH=NH + H202
CH 3 -COOH=NH
CH3-CO-NH2 + H 2 0 + 0 2
3) CH3-COOH=NH + H2C=CH-COOC2H5
>
CH 2 -CH-COOC 2 H 5 + CH3-CO-NH2
4) CH3-COOH=NH
>
decomposition product
5) H 2 C = C H - C O O C 2 H 5
decomposition product
The main problems of this process are the high exothermicity and
instability of the peroxycarboximidic acid, which decomposes (reaction
4), and reacts with the hydrogen peroxide at low temperatures (side
reaction 2) . The objective is to determine the optimal operating
conditions, (temperature and initial composition), that minimize the
reaction 2 and 4 and the decomposition of ethyl acrylate (reaction 5).
The thermo-kinetic parameters were obtained by "thermal flux"
calorimetry[4] . These results are presented in table 3.1.
This process is implemented in a standard 500 1 glass lined reactor
in semi-batch operation mode. The main characteristics of such reactor
are described in table 3.2.
131
TABLE 3.1. Thermokinetic parameters. Ethyl glycidate synthesis
Nreact.
1
2
3
4
5
AHi
(kJraol"1)
54,43
255,39
9,63
102,96
67,83
TABLE 3.2.
(kJinol1)
Ei
Ai
fi (cone.function)
1,90108
45,02
fi
6,26102
1,211014
6,861020
117,65
83,32
124,35
122,67
f2
f3
f4
= cR0/55
3,0610!6
= cAcB
f5 = C
B
C
C R
C R
Glass lin ed
dimensions:
ceramics film:
steel wall:
heat capacity,
reactor
hight, HR = 0,8 m
thick., Ei = 0,002 m
thick., E2 = 0,013 m
T R = 51,7
kj K~l
diameter, D R = 1,0 m
conduct., X-\ = 1.2 W mlRl
conduct., X.2 = 4 6.5 W miK 1
Brine.
Water
kg"IK1
heat capacity,
Cp c
density,
Pc
kg m"'S
viscosity,
4,19
1.103
kg irf "I
s1
1,25
103
11
^c
thermal conductivity, Xc
Wm
temperature range,
Tc
Tmin=2 8 3
flow,
Qc
m 3 s"1
2,08
Temperature
_1
2,78103
103
K~1
p r o p . = 7,
int.=
120
s,
103
0,5
0, 6
Tmin=2 5
10~3
controller
P.I.D.constants:
103
diff.=
60 s
132
3.1.2.
Typical
operating
procedure.
for this
TaavwaturB
3.1.a
( C)
64.0-
SHNEHET
Concentration
5.00.
REMOVED
Concentration
2.00r
(nol/lt)
(raol/lt)
4.0i
1.20-
0
25
CH3CN
50
H202
75
100
ACflYLATE
125
150
ACIDE-fl
175
200
225
250
Time (rnn)
0.00
AC-AMIOE
SLYCIDTE
FW
C*
FIGURE 3.1. Simulation results of ethyl glycidate synthesis with "normal operating conditions".
Tannarature ( C)
3.2.a
32.0-
lB.O-\
-25. VMIXTURE
THEBMOVE
Concentration
5.00-
Concantratlon (rnol/lt)
(ml/It)
1.60-
1.20-
2.00 i
0.40-
1.00-
29
CH3CN -
SO
H202
75
100
ACHYLATE
"*>
125
ACDJE-H
1
190
i
179
i
200
JBBU.VSI
229
290
Tlaa tan)
29
AC^MUDE
90
6LYCUJTE
79
BM
100
129
C""
175
200
229
250
T I M tun)
FIGURE 3.2. Simulation results of ethyl glycidate synthesis with Tset=303 K, Feed rate H202=35 1/hr
rapsr*tura
Co
9 3.1
S.O
B4.0
1.0-
4S.0
-a.o-
32.0
\
A
16.0
0.0
29
MIXTUFE
SO
75
1~
100
IBS
ISO
,
173
.mum
200
THEHHOVE
Concentration
9.00]
223
TlaB 0m)
*23
I
u50
79
SHOVED
Concsntnatlon
3.3.C
tool/It)
15.0"
0
2.001
lOH
IB
150
173
200
100
125
190
178
BOO
'~
C* '
Tlao (an)
taol/ltl
1.20-
0.80-
0.40-
O.Ol
25
CHSCS-"
80
H202~~
73
100
12S
ACHYLATE ACIBM
ISO
170
200
228
"*
250
""
25
ACHWIOE
90
6LYCHJTE
73
5*
223 250
Tlae (on)
Tmparatura (*C)
3. A.
3.4.6
64.0-
16.0 \
0.0^
0
25
SO
- - - - -
MIXTURE
5.0&
73
100
125
150
173
225
Tin*
TWHMOVE
Concntrtlon
BOO
(aol/lt)
250
(on)
Concentration
2.00r
3.4.c
3.4.d
taol/lt)
1.20-
3.00-
2.00 i
0.4O-
1.00-
28
CH3CN
50
H202
75
100
ACHYUkTE
123150
ACIOE-R
173
JSaMLyoJ
2 0 0 2 2 5 2 5 0
Tlaa
(on)
O.OO
25
50
73
AC^AMWe BLYKOTE SS
100
123
CK
150
175
200
225
FIGURE 3.4. Simulation results of ethyl glycidate synthesis with Tset=298 K, Feed rate H202=35 1/hr
2fi
""
Taaporatura
SO.Of-
S.S.a
( CI
64.0-
-7.0-
-11.0-
-1S.0 1
0
24
GENERET
48
72
REMOVED
Concentration (Ml/It)
a.OOr
Concentration (asl/lt)
i.ao-
0.40-
0.00'
0
24
EHSCN"
48
SaniT"
72
96
120
ACHYLATE A C I D M
144
1M
192
216
""
240
0.00*0
24
48
72
AC^AMIDE 6LYCH>TE BM
96
120
ci
144
168
192
216 240
T i m (an)
FIGURE 3.5. Simulation results of ethyl glycidate synthesis with T set =293 K, Feed rate H 2 O 2 =20 1/hr
Tsaparaturs ( C)
3.B.0
-83.01
GENBST
Concentration
REMOVED
Concentration
(Hol/lt)
tol/lt)
S.B.d
1.60-
l.ao0.80-
o.ooi
CH3CN
H202
ACHYLATE
ACIDE-fl
AC-AMIDE
BLYCIDTE R*
FIGURE 3.6. Simulation results of ethyl glycidate synthesis with Tset=293 K, Feed rate H202=35 1/hr
139
3.2. STORAGE OF CYCLOPENTADIENE
The cyclopentadiene, in liquid state, is an intermediate reagent in
many organic syntheses. However, this product is very unstable even at
low temperatures, when it dimerizes exothermically to dicylopentadiene,
which is the stable form until the boiling point, 170C (see ref. [5]) .
The objective of this study is to determine whether is worth to
store the cyclopentadiene, by predicting its stability under storage
conditions.
3.2.1.
General
description
and thermo-kinetic
reaction of cyclopentadiene is:
2C 5 H 6
====>
C10H12
(2) cyclopentadiene
data.
The dimerization
AH
dicyclopentadiene
-3,869-104
cyclopentadiene
1,71
A
E
=
=
1,34-106
6,47-10^
s-l
,
J-mol-l ,
The t h e r m a l c o n d u c t i v i t y was
c o r r e l a t i o n found i n r e f e r e n c e [ 6 ] :
=
dicyclopentadiene
1,64 J-g-l-K-l
A-exp(-E/RT)-Cs
With:
J-mol"1
calculated
using
the
following
3 , 5 9 - 1 0 - 3 - C p - (p4/3) / (Ml/3)
Which g i v e s :
Cyclopentadiene,
X = 0,113
D i c y c l o p e n t a d i e n e , X = 0,110
W-m-l-K - 1
The d i f f u s i v i t y
coefficient
was
calculated
s e l f - d i f f u s i o n phenomenon under i d e a l c o n d i t i o n s [ 6] :
KT
== n
Obtaining,
vv*y
Ds =
4,33-10"9
mZ-s"1
assuming,
140
3.2.2.
Calculation
of storage
conditions.
This case was simulated, by
solving numerically equations 9 and 14. The critical conditions are
determined
by means
of succesive
simulations,
fixing the
characteristics of the vessel and following the behaviour of the
mixture for different storage temperatures.
The following case was used as a reference:
steel spherical vessel:
storage conditions:
for
Xo = 0,10 m
The t e m p e r a t u r e p r o f i l e s f o r t h e s e c o n d i t i o n s a r e shown i n
f i g u r e s 3 . 2 . 3 and 3 . 2 . 4 .
I t i s i n t e r e s t i n g t o compare t h e s i m u l a t i o n r e s u l t s
with the
a n a l y t i c a l c r i t e r i o n of F r a n k - K a m e n e s t k i i [ 7 ] , which e x p r e s s e s t h e
c r i t i c a l s i z e of a v e s s e l a s :
8
X
cr
cr
A. RT^r
m
cr
' AH E A exp(-E/RTcr)
This criterion yields the following critical storage conditions:
TCr = 275.8 K,
for
Xo = 0,10 m
141
350
Temperature (K)
TIME
(hr)
TIME
(hr)
334.0-
31B.0
302.0-
2B6.0
270.0
0.5
0.5
Radius
15.00
Concentration
(mol/lt)
3.7.0
R
(R 0 .10 m)
12.80
10.60-
B.40
6.20 -
4.00
0.5
0.5
FIGURE 3 . 7 . S i m u l a t i o n r e s u l t s of c y c l o p e n t a d i e n e s t o r a g e T s t o r a g e = 2 8 3 K
142
Temperature (K)
320.0
TIME (hr)
3.8.a
0.92
2.93
4.90
6.85
8.83
10.82
12.74
14.47
16.29
18.28
20.27
22.21
24.17
26.12
28.02
30.01
32.00
33.93
310.0
300.0-
290.0-
280.0
270.0
0.5
Temperature (K)
320.0
0.5
R
Radius (R 0.10 m)
TIME
3.a.b
310.0
300.0
290.0-
2B0.0
270.0
FIGURE 3 . 8 .
(hr)
33.93
35.95
37.71
39.65
41.58
43.5B
45.5B
47.55
49.52
51.41
53.38
55.32
57.34
59.28
61.23
63.02
0.5
Simulation r e s u l t s
0.5
Radius
(R 0.10 m)
of c y c l o p e n t a d i e n e s t o r a g e T s t o r a g e = 2 7 9 K
143
FIGURE 3 . 9 . C r i t i c a l
conditions
of
storage
of
cyclopentadiene.
C o m p a r i s o n of s i m u l a t e d a n d F r a n k - K a m e n e s t k i i r e s u l t s .
300.0
X)'
10'
3.2.3 Concluding
remarks.
Under adiabatic conditions, dimerization of
cyclopentadiene will produce a temperature rise of approximatively 350
K. This means that secondary reactions that become significant at
temperature greater than 550 K, may be activated by the main reaction,
leading to a thermal explosion.
The above discussion confirm the high potentail hazard of
storing
important amounts of cyclopentadiene, which requires a
relatively low storage temperature.
As far as possible, the processes where the cyclopentadiene is
an intermediate, should be integrated, in order to avoid the storage of
this compound.
Concerning the difference between the two calculation methods
presented here, it is due to the fact that the Frank-Kamenestkii model
does not take into account the influence of extent of the reaction
neither a resistance of heat exchange with the sourroundings, while,
both were included in the numerical simulation.
144
4 .
Conclusion
5 . Notation
A
C
Cp
Cp L
D
E
e
F
H
J
K
N
Nw
N
Na
n
P
q
R
r
S
T
U
U
V
v
v
X
Pre-exponential factor
Molar concentration,
Specific heat capacity,
Molar heat capacity of a chemical species,
Diffusivity,
Activation energy,
Energy flux,
Molar flow,
Molar enthaply (liquid),
Molar flux by conduction,
Chemical equilibrium constant,
Molar flux by convection and conduction,
Heat flux,
Number
Stirrer speed
Molar Hold-up,
Pressure
Thermal flow,
Rate of chemical production,
or Gas constant,
Rate of reaction,
Surface,
Temperature,
Internal energy,
Heat transfer coefficient,
Volume,
Average velocity,
Local velocity,
Radius,
depends on kinetics
mol-m" 3
J-Kg-l-K-l
J-mol-i-K-l
m 2 s-1
J -mol -1
W-m-2
mols _ 1
J-mol-1
mol-m - 2s _ 1
depends on k
mol-m - 2s" 1
W-m-2
mol
"2
Kg-m-1
W
m o l m" 3 S" 1
J - r n o l " 1 K" 1
m o l m--3 s - 1
m2
K
J
W-m"-2 . K" 1
m3
m- s"-1
m- s" l
m
145
Greek
r
X
n
v
n
P
symbols
Thermal capacity,
J-K~l
Thermal conductivity,
W-m _1 -K~
Dynamic viscosity,
Kg-rrrl-s
Stoichiometric coefficient, reactant(-), product(+)
Partial order of reaction
Momentum flux,
Kg-irrl-s
Density,
Kg-m 3
Vessel geometry, 0=slab, l=cylinder, 2=sphere
Subscripts
c
e
i
m
R
S
E
h
P
s
X
Input
Heating
Species
Wall
Set-point
Radial
6.References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
A. Benuzzi and J. M. Zaldivar (eds.). Safety of Chemical Batch Reactors and Storage Tanks, 147-159.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
148
events which
could transform
this potential
into an
accident.
This stage in the assessment of an installation is called
hazard identification. It is essentially a qualitative
process, although aspects may be revealed which require
calculation,
for
example,
analysis
of
transients
to
establish the boundaries of safe operation.
2. Hazard Identification Techniques
Many hazard identification methodologies are available to be
applied at differrent steps of the project. Most of them are
listed in Table 1.
149
3. H a z a r d a n d O p e r a b i l i t y
Analysis
150
plant
or
operation,
irrespective
of
the
degree
of
complexity, provided that only a suitable list of guide
words is selected. There are several variations of these
lists of guide words. Some are intended for specific
situations and the guide words will reflect this. Other
lists are designed to be far more wide ranging
in
application.
Table 2 : A comprehensive list of guide words
Process variables
FLOW
PRESSURE
TEMPERATURE
LEVEL
Low / High
PHASE CHANGE
COMPOSITION
Contaminants / Composition
Change
NO SUPPLY
DISCONTINUUS
OPERATION
Start-up / Shut-down /
Emergency Shut-down /
Maintenance Activities
LEAKS
EXTERNAL EVENTS
151
A rather comprehensive list of guide words and their
possible "states" is presented in Table 2 for most important
process variables and other events of interest.
Each guide word is applied to the "integral part". How each
deviation might be caused and what the possible consequences
of it may be, are then considered. The kind of questions
that ought to be asked at this stage of the procedures are :
- Could there be no flow ?
- What are the consequences of no flow ?
- Are the consequences hazardous or do they prevent
efficient operation ?
- If so, how could it arise ?
- How will the operator know that there is no flow ?
- If so, can we prevent no flow (or protect against
the consequences) by changing the design or method of
operation ?
- If so, does the size of the hazard or problem
justify the extra expense ?
Having completely examined one part of the design and
recorded any potential hazards associated with it, or
remitted for detailed separate consideration, the study
progresses to focus attention on the next part of the
design. The examination is repeated in the same fashion
until
the whole
section
of
the
design,
etc.
under
consideration has been reviewed.
3.2 FAULT PROPAGATION IN HAZOP STUDIES
The objective in HAZOP studies is to identify deviations
from design intent which may give rise to safety or
operability
problems.
This
necessarily
involves
some
exploration of the possible causes and consequences of the
deviations. It is not, however, the purpose of the study to
make a detailed analysis of these causes and consequences.
If this is considered necessary, it is carried out as a
separate exercise using such techniques as fault trees.
Nevertheless, since a HAZOP study does involve in some
degree the detection of fault paths, the method may be
regarded as one of a member of techniques developed in
recent years for the exploration of fault propagation in
process plants. The fault tree aspects of HAZOP studies are
illustrated by work described by Lihou
(1) , who has
developed a method of encoding the written results of a
hazop studiy to produce a set of fault trees for the plant.
The completeness of the fault trees so produced necessarily
152
depends on the completeness of the fault propagation
information produced by the HAZOP study.
In
normal
HAZOP
studies
the
exploration
of
fault
propagation
by
the
study
team
is
unsystematic
and
fragmentary. This is to be expected, since the systematic
determination of the fault propagation pathways is not the
object of the HAZOP study. These considerations indicate
that there are likely to be difficulties in constructing
fault trees from HAZOP study records.
3.3 OTHER FEATURES
The main advantages of the procedure are :
- Several minds with different engineering
backgrounds address the problem.
- Very detailed results
and the drawbacks :
- Considerable time and manpower required.
- Not all causes are investigated.
Many
interesting
comments
regarding
other
important
features of HAZOP studies as well as some application
examples can be seen in references (2) and (3).
4. Fault Tree Analysis
4.1 GENERAL DESCRIPTION
It is a deductive technique that focuses on one particular
accident event and provides a method for determining causes
of that accident event. The fault tree itself is a graphic
model
that
displays
the
various
combinations
of
equipment/component failures and human/operator errors that
can result in the accident event of interest. These diagrams
must start from an event which has been identified by some
other method.
The technique is essentially binary, i.e., the events or
states in a fault tree are generally assumed to be those
which can be identified as existing or not existing. In
reality there is a whole spectrum or multiplicity of failure
possibilities, some will constitute a state of partial
failure others may be total. Decisions must be taken about
what degree of partial failures constitutes a "failure" in a
fault tree.
The
result
obtained
is a set of
combinations
of
failures/conditions
(cut-sets)
that
being
present
simultaneously produce the accident event. These cut-sets
153
154
simultaneously. If, for example, a protective system is
failed on demand, the time at which the protective system
became unable to react to demand is irrelevant.
There are more complex gates which are sometimes used to
show conditions or sequences. The use of certain gates can
complicate quantification of a fault tree. Examples of these
complex gates are the "EXCLUSIVE OR" and the "SEQUENTIAL
AND" gates. Take for instace the "EXCLUSIVE OR" gate, if an
input A or an inut B exists then the output will be
propagated but should both inputs A and B exist, then the
output will not be propagated. A "SEQUENTIAL AND" gate,
sometimes known as a "PRIORITY" gate, can be used in
situations where the output, for example an explosion, will
only propagate if one of the inputs, an inflammable gas
release, is closely followed by a second input, an ignition
source (note the flammable gas must be in the combustible
range).
4.4 USE OF FAULT TREES IN HAZARD IDENTIFICATION
The main use of fault trees in the overall context of hazard
identification is to structure the failure logic although
they are also used for identification of the causes of top
events. Fault tree construction provides a powerful prompt
to the analyst to consider ways in which
situations
appearing in a fault tree can arise thereby identifying
causes. This prompting action is not structured and so
cannot be guaranteed to be exhaustive. For this reason it
may be wise to apply "bottom-up" techniques, such as FMEA or
HAZOP to parts of the system to improve the likelihood that
all important causes are included.
Because Fault Trees are labour intensive, i.e., require
specialist expertise to perform the analysis and can involve
very detailed analysis of components and operations, their
use in the process industry has mainly been limited to the
analysis of critical areas. Hazard and operability studies
can provide an exhaustive way of going through deviations
but it does not provide any structure for combinations of
events. The techniques are therefore complementary and
should not be seen as alternatives.
Fault trees tend to be equipment oriented therefore the
analyst must be aware of this and ensure that sufficient
attention is given in other areas for instance human errors.
Human errors can be incorporated into the technique but the
main problem here is in identifying them.
Fault tree analysis provides a powerful technique for
identifying and structuring the contributory causes to a
155
156
our example, as they are synthetically represented in Table
3.
The technique has two main parts. The first one (steps 1 to
4) consists of generating all the possible system state
configurations by the combination of the different component
states. Every one of these configurations
is finally
represented by giving appropriate parameter values to a set
of parameteric equations, which contain both logical and
physical information. The second part (steps 5 to 7) solves
numerically the corresponding equation set in order to
foresee the physical consequences of the event sequences
generated, and it verifies whether the TOP conditions are
satisfied (i.e., if some physical variable has reached
certain threshold value).
The process under analysis is described by a set of
component models, obtained by a quantitative failure mode
and
effect
analysis, resulting
in
logical
analytical
relationships characteristic of each component. Such models
should be developed for each single component, but if this
tends to be very CPU-consuming by increasing considerably
the number of sequences to be analysed, the grouping of
single components in macro-components is advisable.
DYLAM provides all the nominal sequences that produce one
or more TOP conditions and the evolution of the physical
variables of interest for those sequences. On the other
hand, the probability distributions of the top conditions
are given as a function of time, taking into consideration
both statistical and functional failure dependencies. It
must
be
stressed
that
the
functional
dependency
consideration is made possible just because of the step-bystep procedure in the evaluation of the physical variables
along the transient state and is has been implemented in the
new version (DYLAM 2) of the code (6).
In order to minimise the number of numerical solutions,
which are generally most CPU consuming, steps 2, 3 and 4 of
Table 3 are performed as soon as an event sequence has been
generated and before proceeding to the physical consequence
evaluation. In that sense, the maximum order of simultaneous
failures to be considered and the cut-off probabilistic
limits are key factors in the total CPU time of the
analysis.
157
T a b l e 3 : S t e p s of D Y L A M
procedure
158
failures. The accuracy of this code should be compatible
with the uncertainties which are present in the other steps
of risk assessment.
5.2 CASE STUDY
An application of DYLAM to chemical plants falling in the
second case indicated above is described elsewhere (6) . The
chemical plant analysed is a batch plant for sulfolane
production; it is divided into two parts : the addition unit
and the hydrogenation unit. The process can be summarised in
the following chemical reactions :
Addition reactor :
S02 + C4H6
>
Sulfolene
C4H6 : butadiene
Hydrogenation reactor :
Sulfolene * H2
>
Sulfolane
159
Different top event conditions were assumed,
regarding
maximum pressure to be obtained for all failure combinations
generated by DYLAM. Results obtained and discussion
are
detailed in reference ( 7 ) .
It was concluded that with DYLAM methodology, the complete
probabilistic and physical related information for all the
possible
and
significant
combinations
of
failures
or
component states were obtained, giving thus the parameters
References
1.
2.
3.
Roach, J.
456-461.
4.
5.
6.
Amendola,
Ispra
7.
and
A.
Lees,
and
F.
Reyna,
(1981) The
C.
(1987)
in the
Chemical
Process
Engineer,
T.N.I.87.128,
(1986)
JRC;
CONTROL TECHNIQUES
C. MOUSSAS
1.
Introduction
The introduction of digital computers to chemical process plants has been started approximately 30
years ago. Even in the first applications, improvements in product quality and process
reproducibility were immediate, since the sequencing of the process steps was being done in a
more consistent and systematic way. With the increasing development of microelectronics, digital
computers became smaller, more powerful, and less expensive. At the same time, new theoretical
results in control theory, specially suited for computer-controlled systems, appeared in the
literature. As a result, advanced controllers can now be implemented in a fast and inexpensive
manner. Also, the substitution of many process units by microprocessors led to the idea of
distributed control systems, where specific control functions are implemented by means of
microprocessor-based devices which are supervised by a general-purpose digital computer.
Because of their complexity, batch chemical processes lend themselves to the application of such
advanced control systems. Major improvements include increased production, better product
quality, cost reduction, and safer plant operation. On the other hand, there is an inevitable increase
in the complexity of the overall system which dictates the need for better personel training, and
elevates the role of process operators [1].
2.
2.1.
161
A. Benuizi and J. M. Zaldivar (eds.). Safety of Chemical Batch Reactors and Storage Tanks, 161-200.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
162
Batch chemical processes consist in the manufacturing of a product by processing raw materials in
a discontinuous way. Usually, the batch reactor includes an agitator, a cooling/heating jacket, a
condenser and a feeding system. Important batch process variables are, among others, the reactor
temperature, the jacket temperature, the reactor pressure, and the feed rate. In all cases, the reactor
temperature, as well as its variations, play an important role in both the quality of the
manufactured product and the safe operation of the plant, and they can be directly controlled by
means of the fluid temperature circulating in the jacket. In the case of highly exothermic reactions,
temperature control may also be performed by controlling the reactant feeding rate, i.e. semi-batch
operation. Other control variables can also be controlled indirectly, since the reactor temperature
definitely affects the reaction rate, as well as the pressure. However, difficulties in controlling the
above mentioned quantities arise form the following facts.
First, some of the process variables may not be easy to be measured, and thus they have to be
inferred by using other available measurements.
Second, many of the process variables vary considerably throughout a batch, and they never
achieve a steady state. This is an unavoidable situation, since it is inherent in every batch process.
Third, startups and shutdowns are normal procedures which have to be automated effectively,
since they represent one of the most difficult phases with respect to control.
Finally, a batch process plant must be able to operate in a wide range of operating conditions,
so that various desired products can be manufactured.
In spite of the above difficulties, batch chemical processes can be effectivelly controlled with a
high degree of reliability and safety, by a proper selection of a process control system [2]. This
requires a good understanding of the batch process characteristics, and a careful consideration of
the process control needs. In the following section we discuss some common control system
configurations, paying more attention to the so-called Distributed Control Systems which are
mostly used today.
2.2.
2.2.1. Classification. Commonly, the existing batch process control systems can be classified into
the following three general categories [2] :
1. Systems based on Programmable Logic Controllers (PLC).
2. Direct Digital Control (DDC) Systems.
3. Distributed Control Systems (DCS).
In the first category, the control system is build around microprocessor-based programmable
controllers which sequence the process and they can handle a limited amount of control functions.
These systems are suitable for small to medium scale batch processes, and where there do not
exist frequent changes in the manufactured product specifications. The main reason is that PLC
units provide two-state(on/off) control functions, and thus, complex controllers are more difficult
to be both programmed and modified. However, as long as no advanced control techniques are to
be implemented, they represent simple, well-functioning, and highly reliable control systems.
They can also be easily expanded by the introduction of either new PLC units or just I/O
hardware. It is important to mention that PLC's are becoming more and more powerful and they
can now be equiped with A/D and D/A modules in order to implement special analog control
functions, such as PID controllers.
In the second category, the control system uses a minicomputer for direct control. That is, a
computer handles all the required control actions, and its output control signals act directly to the
specific final control elements, or other process units. Whenever continuous time signals are
involved, the necessary transformations are performed by means of A/D and D/A converters. Note
163
that these functions can also be implemented on the computer. Futhermore, by appropriate
programming, advanced control techniques can be easily applied. DDC systems are suitable for
medium to large scale batch processes, having complex requirements. Among their drawbacks is
their poor expandability. Also, their complex harware and software usually requires highly trained
personel.
In the third category, microprocessors, designed to perform specific tasks, are responsible for
all the required functions of the control system. Through a communications link they are all
connected to a general-purpose computer which acts as a supervising unit. The supervising
computer does not act directly to the various process elements, but it rather sends appropriate
requests to the specially designed microprocessor-based units which, subsequently, transform
these requests into actions applied directly to the process. Similarly to DDC systems, distributed
control systems are appropriate for medium to large scale batch processes. Furthermore, they are
far more expandable. They also provide more standard functions and they require less
programming skills. With respect to PLC based systems, they provide more sophisticated
operaror interface, and they are able to implement advanced control algorithms as well as better
recipe handling functions. On the other hand, PLC based systems tend to be less expensive for
smaller plants. On the whole, distributed control systems are much more flexible than PLC and
DDC systems, and they are the most appropriate for processes where either the product
specifications are frequently changing, or different products are manufactured.
The main idea in a DCS is that the implementation is totally distributed, and supervision of the
various units is performed through levels of increased centralization. This leads to the notion of
different levels of functionality in a batch process control system. According to the previous
discussion, we can identify two main levels of functionality which will be referred to as the
regulatory control level and the supervisory control level, respectively [3]. Specifically, the
regulatory control level contains the microprocessor-based units as well as any existing manual
control functions, while the supervisory control level contains the supervising computer together
with its peripherical units. In the following subsection we discuss these two levels in more detail,
and we identify the specific control functions that each one of them can handle.
2.2.2. Functionality Levels. Figure 2.1 shows a distributed control system which has been
divided into three basic levels of functionality. Level 1 contains the various process elements, that
is, valves, pumps, agitator, measuring elements e.t.c, level 2 represents the regulatory control
functions, and level 3 the corresponding supervisory control functions. In order to make a clearer
distinction between levels 2 and 3, we now consider some general control functions which are
common to all batch chemical processes. A predefined recipe which determines the specific
product to be produced and also selects the appropriate operations which are to be performed, will
contain the following actions [4] :
- feeding of the raw materials
- strirring of the reactor components at certain speeds
- following a desired temperature profile
- measuring various process variables
- checking the proper function of valves, pumps, e.t.c.
- discharge of the product
- reporting, during and after the batch
The execution of the above actions can be obtained by a combination of sequence and
continuous control, including the following basic control functions :
164
- actuation
- sensing
- interlocking
- regulatory (continuous) control
- recipe handling
- sequencing
- failure detection
- data handling and reporting
LEVEL 3
^iap
>^
V*
sa i
Terminals
SUPERVISING
COMPUTER
Terminals
MICROPROCESSOR-BASED
UNITS
-
Operator Interface
Single Loop controllers
Weighting subsystem
A/D converters
mrm
Printer
J Disc J
*l
Communications
Link
LEVEL 2
Panels
-'-.LJ.
PROCESS
LEVEL 1
165
Data handling and reporting, is the function which organizes data from various devices, and
makes them available as reports, or it transmits them to other systems.
We now turn to the classification of the above control functions with respect to the
functionality levels. As we can see in figure 2, the regulatory control level, or level 2, is directly
connected to the process equipment and to other output devices, while the supervisory control
level, or level 3, acts on the process by giving specific requests which are subsequently executed
by level 2. That is, it is not directly connected to the process. Thus, actuation and sensing
definitely belong to level 2. Simple, well-defined regulatory control functions (e.g. PID), are also
included in level 2, as well as safety interlocks which should not be bypassed by a plant operator.
On the other hand, recipe handling, data handling and reporting functions usually belong to level
3, together with any advanced regulatory control algorithms and any flexible interlocking
functions. Sequencing and failure detection control functions, can be divided between the two
levels, depending mainly on the complexity of each function. The equipment used in level 2 must
be totally error-free, thus requiring low complexity. It must also be protected against changes. On
the other hand, level 3 equipment can be quite complex and it should be able to change whenever
operation conditions, product specifications, or safety measures, change.
3.
Control Techniques
3.1.
INTRODUCTION
3.1.1. General Control Considerations. Control systems are found in all sectors of industry, such
as chemical industry (quality control of desired product), space technology (flight control), power
systems (energy consumption), robotics and others. In fact, our every-day life is greatly affected
by some type of control system (car, air-conditioning, electronic devices e.t.c). The design of
166
such a system consists in determining the appropriate control signals which will be applied to a
specific process, so that its output satisfy some prescribed properties.
In the chemical industry, process means the operations necessary to put together raw materials
and cause them to react in a prescribed fashion in order to produce a desired end-product.
Generally speaking, the process can be described by an equation of the form
Y = f(X,U,t)
(3.1)
where Y represents the desired properties of the end product, and it will be referred to as the
output of the process. The letters X and U represent the variables on which the properties depend.
Specifically, let U represent the variable(s) which we can control directly, i.e. the input of the
process. The control system, or controller, can also be described by an equation of the form
(3.2)
U = g(X,Y,t)
u
Controller
y
Process
167
adjust the controller parameters. If the process parameters change, then loop 2 will take a
corrective action by changing the parameters of the controller.
Adjustment
Mechanism
Loop 2
Set Point
Controller
Process
Loop 1
Figure 3.2 - An Adaptive Control System
Two very important adaptive control schemes which lead to such control system
configurations are :
1. Self-Tuning Controllers (STC) [5],[6]
2. Model Reference Adaptive Systems (MRAS) [5],[6],[7]
For the rest of this report we shall be concerned with these two categories. For information about
other important adaptive control schemes see [5].
3.1.2. Self-Tuning and Model Reference Adaptive Control. A block diagram of a self-tuning
control system is shown in figure 3.3.
Desired
Performance
Design
Estimation
i r
Set Point
Controller
Process
_y_^
168
An assumed dynamic model is associated to the process. Based on this model and on a
controller design procedure, a specific controller can be derived. This controller, which obviously
depends on the model parameters, is then applied to the process. At each time instant the
parameters of the model are estimated from input-output data, and based on these estimates, the
parameters of the controller are computed according to the specific design procedure. We see that
a variety of self-tuning controllers can result from this scheme by considering different structures
for the controller and the estimator. Typical design procedures include the recursive least squares
and projection algorithms, for the estimator design, and the minimization of a quadratic cost
function and pole-placement techniques, for the design of the controller.
A block diagram of a model reference control system is shown in figure 3.4. In this case the
desired performance of the control system is described in terms of a reference model. We again
have to assume that the process is described by some mathematical model. The goal is to
determine the adaptive mechanism which changes the values of the controller parameters, so that
the output y of the process be close to the desired output y m .
STC systems where originally developed for discrete time systems [9],[10], while MRAC
systems were developed for continuous time systems [11],[12],[13]. Also, MRAC as shown in
fig. 3.4, represent the so-called direct schemes because the controller parameters are directly
updated from the adaptive mechanism. On the other hand, STC as in fig. 3.3, are indirect
schemes in the sense that first the model parameters are estimated, and then the new controller
parameters are computed. However, as we can see in these figures, the two configurations are
quite similar. In fact, we can design MRAC for discrete time systems and STC for continuous
ones, as well as, indirect MRAC and direct STC which, as we shall see later in this paper, in
some cases result in identical adaptive control schemes [5],[6],[7].
Desired
Performance
Reference
Model
ym
Adaptive
Mechanism
t
Set Point
u
Controller
Process
169
overall system must satisfy some stability and convergence requirements. Specifically, we require
global stability, which means that the process inputs and outputs remain bounded for all time, and
that the error between the process output and the desired one go asymptotically to zero. Also, in
the case of constant process parameters we require that the estimates converge to the real values.
Since, in principle, adaptive control systems are nonlinear systems, their analysis is quite
complicated and global stability and convergence proofs exist only for some cases and under ideal
situations. An overview of the current stability and convergence results can be found in [5],[6].
Issues associated with the application of adaptive schemes to real nonideal situations, as well as
general guidelines for resolving some practical problems are discussed in [5],[19],[20],[21].
3.1.3. Digital Control and Process Modelling. Nowadays, digital computers play an important
role in the implementation of adaptive schemes. Recent achievements in microelectronics was one
of the main reasons for the development of many different adaptive algorithms, since their testing
through simulations could be done in a very fast and inexpencive way. Since many processes are
inherently continuous-time, some kind of conversion from continuous-time values to discrete-time
values, which can be direcdy processed by the computer (and vice-versa), is needed. This is done
by using analog-to-digital (A-D) and digital-to-analog (D-A) converters. A block diagram of such
a computer controlled system is illustrated in figure 3.5.
z(t)
z(k)
h-
A-D
u(k
Digital
Controller
u(t)
D-A
measurements
y(0
Process
input
w
output
(3.3)
where u and y are the process input and output, respectively, e is a stochastic noise variable (white
noise process), and by definition
170
q'VGOsyfc-l) .
(3.4)
Any delay between the input and output of the process is expressed through the integer d, for then
d > 1. The polynomials A, B, and C are defined by,
n
1
A(q )=l
2^a.q- i
(3.5)
i=l
m
B(q" ) = X b i q " i
(36)
i=0
CtqV^C.q1
(3 7)
i=0
The coefficients a;, bj, and q, represent the parameters of the model. At first glance, one
would say that this model cannot represent a highly nonlinear process, such as a chemical
process, since 3.3 is a linear difference equation. However, by allowing the parameters a;, b;, q,
to change (time-varying parameters), we introduce a nonlinear behaviour into our model. In the
adaptive control case, this is done by the estimation mechanism which continuously updates the
model parameters. The use of ARMAX models is furthermore justified by the fact that when
combined with appropriate control and estimation schemes, they lead to adaptive algorithms for
which stability and convergence results can be established. For these reasons ARMAX models
have been extensively used in the literature [5],[6],[17],[18].
3.1.4. Applications. In a recent article, surveying the design of adaptive control systems [27], the
author begins with the phrase :
" Adaptive control is now finding its way to the market place after many years of effort"
In fact, even if the first applications of the adaptive control systems appeared in the 1950s in
the context of aircraft flight control, it was only in the 1980s where the development of both the
theory and the computer hardware, allowed for a systematic application of various adaptive
control schemes to the industry. As reported in [5], it was estimated that in May 1988 there were
at least 70000 industrial process loops in which adaptive techniques were used. A description of
various commercially available industrial adaptive controllers, announced in the 1980s from
companies in Europe, North America and Japan, can also be found in [5]. At least half of them are
based on the PID controller structure, which is the most common technique applied to industrial
processes, and provide automatic tuning of the PID parameters. They are very useful, since it
appears that even today many industrial PID controllers are poorly tuned [27].
Some recent applications of adaptive control to chemical reactors are reported in [25], [28][31]. In [25], the generalized minimum variance self-tuning controller (sec. 3.4.2.4) was applied
to a batch chemical reactor (Real Plant). The results showed a smoother temperature profile, than
that obtained by a PID controller, as well as insensitivity to reactant quality. In [28],[29],[30],
applications of various adaptive control techniques to a batch polymerization reactor are described,
171
both to simulated [28],[29],[30] and to pilot plant [30], experiments. The techniques included,
among others, minimum variance and generalized minimum variance selftuning control (sec.
3.4.2), as well as adaptive poleplacement (sec. 3.4.3). Their performance was showed to be
either equally good, or better than the performance of well tuned PID controllers. Various control
objectives were considered, such as isothermal operation, constant rate operation, constant weight
average molecular weight, and specific monomer conversion profile. Finally, in [31], a
modification of the basic minimum variance selftuning controller (sec. 3.4.2.3) was applied to a
simulated continuous stirredtank reactor, where it was demonstated that it was more effective
than the basic method. Other numerous applications of various adaptive control schemes can be
found in the survey parers [21], [27] and [32].
3.2. PID C ONTROLLERS
Even today, PID feedback controllers are the most widely used controllers in the industry. The
reason is that they perform quite well whenever the process is of reasonable complexity and the
control specifications are not very tight. Among their features are the integral action, which
eliminates steady state errors, or offsets, and the derivative action, which improves the stability of
the closed loop system. Their function can be described, in its simplest form, by the equation
(,) . KI M + Y |=W
I
ds + T
d^ T I ,
'
(3.8)
where u(t) is the control signal, e(t) is the control error which equals the difference detween the
output y(t) and the set point y r (t), and K, Tj and T j are the socalled PID parameters, namely, the
proportional gain, the integral time, and the derivative time, respectively. In Laplace transform
representation, the above equation can also be written as
U(s) = K
1 + T ^ + T .d s
T. s
E(s) .
(3.9)
In practice, the algorithm described by 3.8 is subject to various modifications which improve its
performance. For examlpe, in the case of pure proportional control we have
u(t) = Ke(t) .
(3.10)
Since the goal is the elimination of the error e(t), and since 3.10 implies that u(t)=0 whenever
e(t)=0, we realize that in most cases there will exist a nonzero steadystate error between the
output and the set point. For this reason, instead of 3.10 we frequently use the equation
u(t) = K e(t) + A
(3.11)
172
u(t) = K| Z(Q + Y
e(s)ds | .
'
(312)
Now assume that after a time instant t' we are in steady state with u(t) = u(t') = u', while e(t) =
e(t') = e' * 0, for t > t'. That is, a nonzero offset. Then, for a t > t' equation 3.12 gives
rI ,
i f.,
.
e'(t t') I
u(t) = K | e + e ( s ) d s +
|
fort>t'
(3.13)
Note, however, that the third term of the above expression is not constant, but it depends on the
time t. Consequently, u(t) is not constant. But this contradicts our hupothesis. On the other hand,
we see that for e' = 0 the control signal equals
r}
e(s) ds
which is constant.
A problem associated with the integral action of a PID controller is the socalled integral
windup, or reset windup, which appears whenever the control signal drives an actuator, or final
control element, having limited range. Then, when the actuator reaches its upper limit, it will
remain there even if the control signal is increasing. Simultaneously, the control signal will keep
increasing, since the error is not zero and the algorithm cannot detect that the nonzero error is not
due to the magnitude of the control signal, but rather to the saturation of the actuator. This
situation has the following consequences. When, at some time t, the error becomes zero, the
integral part stops integrating, but the control signal is so big that the output of the process will
still be increasing, and it will take time before its derivative changes sign. Therefore, this
behaviour leads to large overshoots and long settling times. One way to avoid this problem is to
use the integral part only when the error is sufficiendy small, thus avoiding the actuator saturation.
Another way is to stop updating the integrator when the actuator saturates, or keep its value
bounded. This can be done, for example, if we use the following modified form of the integral
part
t
= - fe(s)ds + ^
fe t (s)ds
In this case, the second term integrates the error et(t) defined as
(3.14)
173
e t (t) = v(t) - u ( t ) ,
(3.15)
where u(t) is the control signal, and v(t) is the actuator signal. In normal conditions, v(t) = u(t)
and therefore et(t) = 0. On the other hand, when the actuator saturates, et(t) becomes negative and
keeps the integrator output bounded.
As we have already mentioned, the derivative action improves the stability of the closed loop
system. Intuitively, this can be understood if we rewrite the derivative part as
deffl
dt
e(t + At)-e(t)
At
This expression implies that the derivative term provides anticipatory action, for the error at time
t+At, thus improving the performance of the controller and the stability of the overall system. If
high frequency measurement noise is present, the derivative term will result in flactuating control
signals whose amplitude will be an increasing function of the frequency. For this reason, the
derivative term is often filtered through a first order filter. Then, in Laplace transform
representation, tthe derivative term becomes
= T7TWE(S)
<3-17>
fE(s)+fE,(s)
T s
K T d s E d (s)
1+T. s/N
(3.18)
where
E p (s) = b-Y r (s) - Y(s)
(3.19)
(3.20)
(3.21)
(3.22)
N is an integer in the range 3-10, T t is a rather small constant, and K, Tj, T j are the PID
parameters. The implementation of the PID algorithm on a digital computer requires a discretetime model, which must be equivalent to that of equation 3.18 when the sampling period is small
enough. There are many methods to derive such a discrete-time model [14],[38]. In the case
174
where an Euler approximation for the integral part and a backward-difference approximation for
the derivative part are used, equation 3.18 becomes
Khe(k)
u(k) = Ke p (k)+ T (q-1)
;
he t (k)
Td(q-l)ed(k)
T t (q-1)
(3.23)
y(t)
K
1/RL
0.9/RL
1.2/RL
Td
3L
2L
0.5L
175
The second method is a frequency responce method. First, a proportional controller, as in
3.10, is applied to the process. After, the proportional gain is increased until the output of the
process oscillates periodically. At that point, let K c be the gain of the proportional controller and
T c be the period of the oscillations. Then, the PID parameters are obtained by means of table 3.2.
Table 3.2 - PfD parameters (closed loop methodt
Controller
Type
P
PI
PID
Ti
Td
0.5KC
0.4KC
0.6KC
0.8TC
0.5TC
0.12TC
The PID parameters obtained by using either of the two methods have to be considered as a rough
estimate, and based on them, more accurate values must be derived manually. A more detailed
discussion of the above methods, as well as methods for automatic tuning of PID controllers, can
be found in [38] .
3.3.
PARAMETER ESTIMATION
erxr(k) ,
where y(k) can be thought of as the output of the system, and xi (k)
(3.24)
xr(k) are combinations
e(k) = X [ y ( i ) " y + ( i ) ] 2
i=l
(3.25)
176
is minimum. The estimated parameters, at time k, should minimize this error. Equations 3.25 and
3.24 contain the two most important features of the least squares formulation, namely, the
minimization of a squared error criterion, and the linear fashion in which the parameters are
introduced into the model. These two facts allow for an analytical solution to exist. Specifically,
let
x(k) = [ X j (k) x 2 (k) - x r (k) ] T ,
(3.26)
and
e=[e1e2-er]T>
(3.27)
which will be referred to as the regression vector and the parameter vector, respectively. Then,
3.24 can be written as
y(k)=_x T (k)_9 ,
(3.28)
which will be called the regression equation. At time k, the estimate which minimizes the error
e(k) can be shown to be the solution of the equation (see [16],[22])
R(k)-8(k) = p(k)
(3.29a)
or
6(k) = K 1 ( k ) p ( k )
(3.29b)
where
m = 2^_x(i)xT(i)
(3.30)
i=l
and
p(k) = 2^y*(i)-xT(i) ,
(3.31)
i=l
a matrix of dimension (r x r) and a vector of dimension (r x 1), respectively. As far as the notation
is concerned, the underscore indicates a vector, the double-underscore a matrix, and the
superscript T the transpose of either a vector, or a matrix. Equation 3.29b implies that,
theoretically, we are able to compute the solution if the matrix R is inversible. However, the
amount of computation needed for inverting this matrix, for large r, makes this approach not
177
feasible for real-time applications where time is an important factor [22]. For this reason, in many
practical applications the vector 6(k) is computed recursively, and its value at time k depends on
the previous estimate, at time k-1, plus new information available at time k-1. The idea is the
recursive computation of the matrix R and the vector pa in 3.29b, based on their values at time k1. This leads to the so-called Recursive Least Squares (RLS) algorithm which provides the basis
for many different recursive identification algorithms. In what follows we discuss the application
of the RLS, ELS, and projection algorithms, in the parameter estimation of processes described
by ARMAX models.
3.3.3. On-line Estimation Algorithms. The ARMAX model, which was introduced in section
3.1.3, is described by the set of equations 3.3-3.7. Assume that e(k)=0 for all k's (deterministic
case), so that the process is approximated by the model
ACq 1 ) y(k) = BCq 1 ) u(k-d) ,
(3.32)
y(k) = - a 1 y ( k - l ) - - a n y ( k - n ) + b 0 u ( k - d ) + - + b m u ( k - d - m )
(3.33)
(3.34)
-Ve2=-a2'---'er=bm
(3-35)
and
Xj(k) = y(k-l) , x 2 (k) = y(k-2) , . . . , x r (k) = u(k-d-m) ,
(3.36)
3.34 yields 3.24. Therefore, the application of least squares is straightforward and the solution is
given by 3.29. In the sequel we present its recursive version, i.e. the RLS algorithm, in the case
where the error to be minimized is given by
k
(3-37)
i=l
(3.38)
178
and
u(k-l) = [ y * ( k - l ) - y * ( k - n ) u ( k - d ) - u ( k - d - m ) ] T .
(3.39)
The argument k-1 in the definition of the vector u indicates that all its components are available at
the sampling time k-1, or before. With the above notation, the recursive least squares equations
are as follows ([16],[22]):
9(k) = e(k-l)+g(k)[y*(k)-_u T (k-l)_e(k-l)] ,
(3.40)
where
P(k-l)-u(k-l)
g(k)=
UuT(k-l).P(k-l)-u(k-l)
(3 41)
and
P(k) = -L[P(k-D-g(k)-uT(k-l)-P(k-i)] _
A.
(342)
Equation 3.40 implies that the current estimate depends on the previous estimate, plus the error
y (k)-u(k-l)9(k-l) resulting by the use of the previous estimate multiplied by a time-varying gain
(k). This gain is determined by 3.41 and 3.42, where the matrix P(k) is being interpreted as the
covariance matrix of the estimation error. For the algorithm to work, initial values of the parameter
vector (0) and the covariance matrix P(0) are needed. Usually, P(0) is chosen to be the unit
matrix multiplied by a constant. A large value of this constant indicates that 6(0) is a poor
estimate, and vice-versa. This will, initially, cause rapid changes in the estimated parameters Q(k).
The constant X, taking values in the interval (0,1], is called the forgetting factor. The choise X=l
will equally weight all the measurements from k=0, while a value of \<l will discard old
measurements by weighting more heavily the recent ones. The smaller the value of X. the less
weight into the old measurements wil be assigned. This will allow for parameter tracking in the
case where the parameters are not constant but rather drift. For more details concerning the
specific choise of X, as well as other techniques dealing with time-varying parameters such as
covariance reseting or use of variable forgetting factors, see [5],[6],[19],[21]. Another important
issue associated with the RLS algorithm is parameter convergence. As we have said, the criterion
is the minimization of the error in 3.37. We notice that no requirements about the parameter
behaviour are included in this criterion. This can very well lead to the case where the error goes to
zero, while the parameters do not converge to the real values. This situation can be avoided if the
vector u(k) possess the so-called persistent excitation property (for a formal definition see
[5],[6],[16]). Practically speaking, a persistently exciting sequence should contain a sufficient
number of frequencies, in its frequency spectrum, so as to allow for parameter updating
[5], [6],[19], [21]. Lack of this property can also lead to various bursting phenomena [20].
179
3.3.3.2. Projection Algorithm. The above recursive computation of the solution to the least
squares problem considerably reduces the computational complexity, by avoiding the matrix
inversion. It is probable, however, that further reduction of the computation effort be required,
depending on the application and especially in the case where the dimension of the vector j is too
large. Then a frequently used simplified algorithm is the projection algorithm described below.
g(k) =_6(kl)
+ n
p.43)
(3.44)
If {e(k)} is white noise, then the RLS estimates, as described adove, will converge to the true
parameter values. But, if e(k) is a sequence of correlated random variables, then it will produce
biased parameter estimates [16],[18]. This correlation in the noise sequence is expressed through
the polynomial C (q~l) which leads to a process model as in equation 3.3, with c, * 0 for some i >
1, and {e(k)} being white noise. In the following we describe the extended least squares
algorithm (ELS) which, based on this model, produces unbiased parameter estimates [16],[18].
First, we rewrite 3.3 as
y ( k ) + a 1 y ( k l ) + + a n y ( k n ) = b 0 u(kd)++b m u(kdm) +
+c0e(k)+c]e(kl)++cne(kn)
. (3.45)
Then, by defining
9 = [ a, a a b b m c c n ]
L 1 2 n 0 m 0 n
(3.46)
and
(3.48)
180
Therefore, the RLS algorithm can be directly applied, where now the vector 0 contains also the
correlation coefficients of the noise. Notice, however, that the regression vector u ^ k - l ) , which is
supposed to be known, contains the variables e ( k ) , . . . , e(k-n), for which measurements are not
available. For this reason, in applying the RLS we have to replace these variables by
approximations. A possible choise is
e(k) = y * ( k ) - u T ( k - l ) - 6 ; k - l ) ; V k
(3.49)
i.e. the prediction errors. The method which results by applying the set of equations 3.40-3.42 for
the estimation of the parameters in 3.48 with the above modification in the regression vector, is
referred to as the extended least squares method [5],[6],[16]. Finally, in order to give an example
for the presence of correlated noise, consider the following situation. Let the process model be
deterministic, that is, as in 3.32. Assume that the measurements of the system output are
corrupted by additive white noise. Then, our modelled output is not y(k) but rather y'(k) where
y'(k) = y(k) + e(k) .
(3.50)
(3.51)
(3.52)
3.4.1. Introduction. The design of adaptive control systems is based on the idea of simultaneous
parameter estimation and control. Therefore, by using different parameter estimation methods and
different control design procedures, many adaptive control schemes can be produced. Since
parameter estimation was treated in the previous chapter, we shall now discuss some control
design procedures together with their adaptive versions. Two main categories are considered in
the following sections. In the first category (section 3.4.2), the desired performance of the overall
system is expressed by means of a quadratic cost function, and the design of the controller is
based on the minimization of this function, while in the second category (sections 3.4.3 and
3.4.4), the desired performance is expressed by specifying some characteristics of the closed loop
transfer function which, also, determine the structure of the controller. We shall always assume
that the process is described by an ARMAX model as in equation 3.3 which we rewrite below as
A(q-') y(k) = q - f y q 1 ) u(k) + C(q" ! ) e(k) ,
(3.53)
with A(q"l) being a stable polynomial, that is, having all its roots inside the unit circle.
181
In every case which we present below, the resulting controller will be a special case of a
general linear controller described by an equation of the form
V ^ q 1 ) u(k) = V ^ q 1 ) y r (k) - V^q" 1 ) y(k) ,
(3.54)
where u is the output of the controller, y r the set point and y the process output. V j , V2 and V3
are polynomials of the backward shift operator q~l.
3.4.2. Generalized Minimun Variance Self-Tuning Controllers.
3.4.2.1. Deterministic One-step-ahead Self-tuning Control. We begin with the deterministic case.
We assume, also, that the process delay is d=l. Then, the process model becomes
A(q-1)y(k)=B(q-1)u(k-D .
(3.55)
Let y r (k) be the set point, not necessarily constant. Also, define a polynomial P(q~') as
P(q" 1 ) = l+p 1 q- 1 +-+p I 1 q- n .
(3.56)
Then, our goal is to determine a control law u(k) which minimizes the following quadratic cost
function:
k>o.
(3.57)
P(q"l) determines the dynamic behaviour in which y(k+l) approaches y r (k+l). The way to
minimize J(k+1) is to take its first partial derivative with respect to u(k) and set it equal to zero. In
this case, however, we can simplify the calculations if we notice that the minimum value of J(k+1)
can be made equal to zero by requiring that
P(q" 1 ) [ y(k+l) - y r (k+l) ] = 0 ; k>0 ,
(3.58a)
or, equivalently
P ( q " 1 ) y ( k + l ) = P ( q - 1 ) y r ( k + l ) ; k>0 .
(3.58b)
In order to proceed with the derivation of u(k), we define A (q" 1 ), B ( q - 1 ) a n d P (q"1) by the
following equations
A(q1) = l+q-'AV')
(3-59)
B(q- 1 )=l+q" 1 BV I )
(3-60)
182
P(q_1) = l+qVfo" 1 )
(3.61)
(3.62)
(3.63)
(3.64)
(3.65)
(3.66)
We see that the controller output at time k is a function of the set points yr(k+l), y r (k),. . . , of
the process outputs y(k), y(k-l),. .. , and of the previous controller outputs u(k-l), u(k-2),. . .
Equation 3.66 can be written also as
u(k) = L - [P(q-1)yr(k+l)+[A*(q-1)-P*(q-1)]y(k)] ,
B(q 1 )
(3.67)
which implies that the roots of B(q~l), i.e. the process zeros, should lie inside the unit circle or, in
other words, the process should be minimum phase. This is necessary, in order to have a stable
control law which will remain bounded when disturbances are present. The transfer function of
the closed loop system can be derived by substituting 3.67 into 3.55, and this gives
A(q' 1 )y(k+l) = P(q-1)yr(k+l) + [AV 1 )-P*(q" 1 )]y(k) -
(3.68)
(3.69a)
183
y P (k+l) = P(q" 1 )y(k+l) ,
(3.69b)
(3.70)
we get
P ( q 1 ) y ( k + l ) = yPr(k+l) .
(3.71)
H(q
)=^- =^ V ,
y P (k+l)
(3.72)
P(q" 1 )
that is, the roots of P(q~l) are the poles of the closed loop system, and they have to be inside the
unit circle so as to result in a stable closed loop system.
If the model parameters, that is, the coefficients of A(q"l) and B(q"l), are known, then the
computation of the control law u(k), given by 3.66, is straightforward. If not, then we need first
to estimate them, using one of the methods described in section 3.3, and then compute u(k) as in
3.66 where instead of a j , a2, . . . a n , bQ, b j , . . . , b m , we put their estimates. This can be
repeated in every sampling instant and it leads to a so-called indirect adaptive control scheme. The
name is due to the fact that the controller parameters are not estimated directly, but rather
indirectly, through the controller design procedure. Another approach, which leads to a direct
scheme, is described below. We notice that, using 3.69 and 3.61, the process model in 3.55 can
be written as
y P (k+l) = B(q" 1 )u(k) - [A*(q V p V ) ] y ( k )
(3.73)
(ai-Pi)-(an-pn)
]T
(3.74)
and
_u(k) = [ u(k) - u(k-m) -y(k) -
-y(k-n+l) ] T ,
(3.75)
(3.76)
which has the form of the regression equation as in 2.28. Then, the parameter estimation methods
of section 3.3 can be directly applied for the estimation of the controller parameters. This can be
184
done in every sampling instant, and it leads to a direct adaptive control scheme. Notice that in both
methods, indirect and direct, the number of the estimated parameters is the same, and equal to
m+n+1. We see that in the indirect case the controller design step (i.e. the computation of the
controller parameters from the estimated process parameters), has been eliminated, thus resulting
in a simpler method. This, however, will not be always the case as we shall see in the sequel.
We now turn to the case where the process delay d is greater than one. The process model will
be
A ( q 1 ) y ( k ) = B(q- 1 )u(k-d) ,
(3.77)
and the control law u(k) should minimize the cost function
2
k>0
(3.78)
which corresponds to 3.57. Let us proceed as previously, that is, define 3.59 - 3.61 and try to
satisfy 3.58 where instead of k+1 we put k+d. By following exactly the same steps, the
corresponding controller equation of 3.65 will be
(3.80)
which has a unique solution for S(q _1 ) and T(q _1 ) (see [6],[14]), defined by
S(q" 1 ) = l + s ^ + ' - H - s ^ q - " * 1
(3.81)
and
Kq'1) =
T1q"1+-+Tnlq-n+1
(3.82)
185
respectively. We define, also, the polynomials BS(q~l) and BS (q"l) by the equation
(3.83)
Then, the steps corresponding to 3.62 - 3.64 are as follows. Substituting 3.80 into
PCq'1) y(k+d) = PCq'1) y r (k+d)
(3.84)
yields
[A(q" 1 )S(q" 1 )+q" d T(q" 1 )] y(k+d) = P(ql)
y r (k+d) ,
(3.85)
or
A(q" 1 )S(q" 1 )y(k+d) = P(q 1 )y r (k+d) - T(q" 1 )y(k) .
(3.86)
(3.87)
(3.88)
and
( 3 .89)
where, now, u(k) is a function of the set points y r (k+d), y r (k+d-l), . . . , of the process outputs
y(k), y(k-l), . . . , and of the previous controller outputs u(k-l), u(k-2), . . . , i.e. it represents a
causal controller. Notice that 3.89 reduces to 3.66 when d=l.
For the adaptive controller based on 3.89, it is interesting to consider the direct and indirect
schemes in more detail. Specifically, for the indirect case we have the following algorithm :
Algorithm 3.1 (Indirect Schemel
1) Estimate the model parameters (coefficients of A(q~l), B(q"'))
2) Compute T(q" ! ) and S(q" J ) by solving 3.80
3) Compute u(k) as in 3.89
4) k=k+l ; go to 1
We see that the controller design is now rather complicated, when compared with that in the case
of d=l, since the solution of the equation 3.80, at each sampling instant, is required. For the
186
direct case we proceed as follows. First, we multiply both sides of 3.80 by y(k+d). Then, by
combining the result with the model equation 3.77, we get (compare with 3.73)
(3.90)
Since we are interested in reducing the computational effort of the parameter estimation algorithm
as much as possible, we must not choose a very small value of h. On the other hand, we must
keep its value small enough, so that the discrete model adequately describe the process. More
details on the selection of the sampling period in implementing discrete-time control systems, can
befoundin[5],[14],[21].
The controller described by 3.89 is usually referred to as the one-step-ahead controller because
the criterion which is based on, takes care only of one step ahead ((k+d)th sampling instant).
Criteria including the system behaviour for a set, or horizon, of future sampling instants lead to
control techniques such as Model Predictive Control [23], or Generalized Predictive Control [24].
3.4.2.2. Weighted Control. Frequently, the method described above produces control signals
whose magnitude is quite large. This is due to the specific form of 3.78 which does not include
187
any requirements about u(k). One way to put a weight on it, is to consider a cost function of the
form
2
(3.91)
and minimize it with respect to u(k). Intuitively, this criterion implies that a situation like
y(k+d)=y r (k+d) being achieved by a lagre value of u(k) will be avoided, since even if the first
term of the right-hand side of 3.91 is zero, the second one will result in a very large value of
Jj(k+d). Therefore, it may be preferable to reduce the absolute value of u(k) in order to produce a
smaller value for Jj(k+d). For the same reason, a control law based on 3.91 can produce an offset
in the steady state responce of the closed loop system. A slight generalization of the cost function
which alleviates this problem is given by the following :
J 2 ( k + d ) = [ P(q" 1 )[y(k+d)-y r (k+d)] ] + l.[ u (k)-u(k-l)] 2 ; 1>0 . (3.92)
We see now that the weighting is placed on the increments of the control signal, rather than on the
signal itself. The term u(k)-u(k-l) can be rewritten as (l-q"')u(k) and by defining Q(q~l)=l-q~l,
3.92 becomes
2
)yr(k+d)-T(q-1)y(k)-BS*(q-1)u(k-l)]]
(3.94)
b0+l
Notice that if 1=0, then equation 3.89 results. Also, if we just remove the term lu(k-l), then the
resulting controller equation is the one which minimizes Jj(k+d). As both cases are simple
generalizations of the controller in 3.89, their adaptive versions can be obtained in exactly the
same manner as described before. Their application will produce control signals of reduced
magnitude with respect to the previous cases, and for this reason they are useful when constraints
on these signals exist. We must keep in mind, however, that specific maximum or minimum
values of u(k) are not included in 3.93.
The controller equation 3.94 can also be written as
u(k),P(q-
)y^d)-T(q^)y(k)
which implies that it is not necessary for the process to be minimum phase, as long as, the roots
of the denominator of 3.95 are inside the unit circle. Also, by substituting 3.95 into 3.55 we can
easily obtain the closed loop pulse-transfer operator, given by
H ( q ' ) - - ^ =
yP(k+d)
BCq-^PCq-Va/boXl-q-VCq-1)
(3.96)
which implies that also the roots of B(q~l)P(q~l)+(L/bo)(l-q~l)A(q"l) should lie inside the unit
circle, in order for H(q"') to represent a stable closed loop system.
3.4.2.3. Minimum Variance Self-tuning Control. So far we considered the deterministic case,
where C(q"l)=0. In the stochastic case we assume that disturbances are present, which can be
modelled as a random noise process, say {h(k)}. This leads to a process model as in 3.53. In that
model, {e(k)} is a white noise random process and C(q~l) is a polynomial representing any
correlation between the random variables of {h(k)}, whose roots should be inside the unit circle.
We can also assume that co=l[17],[18]. Furthermore, we assume that the random variables e(k)
have zero means and finite variances. Thus, the output of the process y(k) will also be a random
variable and therefore any criterion describing its desired behaviour should include some kind of
probabilistic expectation. In the simplest case, the control law u(k) must minimize the expectation
J(k+d) = E{ [y(k+d) - y r (k+d)] 2 } ,
(3.97)
given measurements available at time instant k. Notice that for y r (k+d)=0 V k (regulation
problem), the criterion becomes the minimization of the output variance. The resulting controller,
when combined with the RLS parameter estimation method, leads to the minimum variance selftuning regulator, originally proposed in [9].
In order to minimize the expectation in 3.97 we proceed as follows. First, we determine two
polynomials S(q~l) and T(q~l), defined by 3.81 and 3.82 respectively, by solving the polynomial
equation (compare with 3.80)
C(q" 1 ) = A(q- ! ) S ( q 1 ) + q d T t q 1 ) .
(3.98)
(3.99)
or, equivalently
y (k+d)
A(q 1)S (q 1)
" 1 "
1
C(q )
y(k+d) + ^ - y(k) ,
C(q1)
(3.100)
189
(3.101)
BS
^(k)3^y(k)-yr(k+d)
CCq1)
CCq1)
+EnS(q- 1 )e(k+d)]
(3.102)
because {e(k)} and {u(k)} are assumed to be independent, and furthermore, y(k) is independent
of each one of the random variables e(k+d), e(k+d-l), . . . , e(k+l). Finally, since the second
term of 3.102 does not depend on u(k), we require that
^mlm+mlyik).yik+d)=0
CCq"1)
CCq"1)
(3.103)
u(k) = J - [C(q-1)yr(k+d)-T(q-1)y(k)-BS*(q-1)u(k-l)] ,
b
(3.104)
i.e. the same as that in 3.89 if we chose P(q"')=C(q"'). The main difference is that while in the
deterministic case this control law sets the error P(q~l)[y(k+d)-yr(k+d)] to zero, in the stochastic
case the minimum value of J(k+d) (eq. 3.97) is given by
W k + d ) = E{[s(q"1)e(k+d)]2.
= Var[e(k+d)]+SjVar[e(k+d-l)]+-+s^jVar[e(k+l)]
. (3.105)
However, the various assumptions pertinent to the stability of both the control law in 3.89 and the
overall closed loop system in the deterministic case, apply also to this case for C(q~l)=P(q~l).
The adaptive version, or minimum variance self-tuning controller, will be a combination of the
controller in 3.104 with a parameter estimation technique. If C(q~l)=l then the RLS algorithm
will suffice, while if C(q"l) is a polynomial of a degree greater than zero, then the ELS method
should be used. As before, either the direct, or the indirect schemes can be applied.
3.4.2.4. Generalized Minimum Variance Self-tuning Control. We now present a generalization of
the above controller, which has been introduced in [6], based on the cost function
(3.106)
190
which generalizes 3.97. P(q"'), R(q"^) and Q(q"^) are polynomials in q~l where, generally
speaking, Q(q"*) weights the control signal, R(q~l)yr(k+d) represents a filtered version of the set
point, and P(q ) greatly affects the dynamics of the closed loop system. The expectation is
conditioned on system input-output data available at time k. If we remove the expectation, we
have a generalization of 3.93 for the deterministic case.
The control law which minimizes 3.106 can be shown ([25]) to be given by
Rfa-VOc+d) - (^(k+d/k)
u(k) = i ^
\
,
(q0/b0)Q(q1)
(3.107)
where qo=Q(0), brj=B(0), and <j)v (k+d/k) is the least-squares prediction ([22]) of P(q"^)y(k+d)
given data up to k, and it is given by
.
"
T V W + BS^Mk)
C(q"1)
where BS(q _ 1 ) has been defined in 3.83. T(q _ 1 ) and S(q" 1 ), defined by 3.81 and 3.82
respectively, are determined by solving the polynomial aquation
C(q" 1 )P(q" 1 ) = A(q 1 )S(q" 1 ) + q ^ q
(3.109)
. T ( q ' ) y ( k )
B S ( q ) + (q 0 /b 0 )Q(q- )C(q- )
and the closed loop system output is given by
[BS(q- 1 ) + (q 0 /b 0 )Q(q- 1 )C(q- 1 )]e(k)+B(q- 1 )R(q- 1 )y r (k)
y(k) =
:
i
;
;
P(q- 1 )B(q 1 ) + (q 0 /b 0 )Q(q- 1 )A(q- 1 )
(3.111)
Thus, this method can be applied also to non-minimum phase processes for which the roots of the
denominator polynomial of 3.111 are inside the unit circle. Notice that if in 3.110 we set C(q"
1)=1, then the resulting control law will be either the one which minimizes 3.106 when the
process model is given by 3.44 with e(k) being white noise, or that which minimizes 3.106 in the
deterministic case (i.e. drop the expectation) where the process model is as in 3.77.
The adaptive version is obtained by combining 3.110 with one of the parameter estimation
algorithms described in section 3.3. In the indirect case one has to estimate the coefficients of the
polynomials A, B and C at each sampling instant, and then apply 3.110 by using their estimates.
In the direct case the controller parameters, that is the coefficients of T(q"'), BS(q _1 ) and C(q _1 ),
can be estimated by means of 3.108 which can be written as
191
n
VlbS0 ^ f
bS
m+d-l
"Cl " V
"Cn ] T
(3-113)
and
u(k) = [ y ( k ) - y ( k - n + l )
u(k)-u(k-m-d-l)
<|>y(k+d-l/k-l)-<t)J(k+d-n/k-n) ] T
(3.114)
respectively, it becomes
(fyk+d/k) = u T (k)-9
(3.115)
which is a regression equation, as in 3.28, hence the RLS method can be applied. Notice however
that <t>v (k+d/k) is not known because it depends on the parameters, and for the same reason the
vector u.(k) is not completely known. Therefore, an approximation of these quantities has to be
used (compare with ELS). Specifically, we can put instead,
ty(k+d/k) = u T ( k ) 6 ( k )
(3.116)
where f)(k) is the previous parameter estimate and the symbol " A " over u implies that the <|>components in 3.114 have been calculated also by 3.116. Finally, once the estimates of T(q~l),
BS(q"l) and C(q"l) are available, the control law can be readily computed by 3.110.
The choise of the polynomials P(q"^), R(q ) and Q(q"') depends on the specific requirements
of the closed loop system behaviour. As we mentioned above, P(q"^) affects the closed loop poles
and if, for example, Q(q~l)=0 then equation 3.111 shows that the roots of P(q"^) become poles of
the closed loop response. In the case of sudden set point changes, R(q"*) can be chosen, for
example, as
R(q" 1 ) =
-
1 - rjq" 1
(3.117)
so that to filter the set point. This will be advantageous in many cases since it will reduce the
magnitude of both the control signal and the overshot, at the expense of slower tracking. In this
case r should be chosen such that P(1)=R(1) in order to avoid a steady state offset. Finally, Q(q"
1) is used in order to reduce the magnitude of the control signal. Equations 3.91 and 3.92
represent two such examples forQ(q"^)=l and Q(q"')=l(l-q"'), respectively. More details on the
choise of the above polynomials can be found in [6], [21], [25], [33] and [34].
192
3.4.3. Adaptive Controllers Based On Pole Placement. In section 3.4.2 we presented a class of
adaptive controllers which were based on control laws having the form of equation 3.54. In all
cases, these control laws were the result of a minimization of a quadratic cost function as, for
example, in 3.106. In this section we follow a different approach. We assume from the beggining
that the controller is described by an equation as in 3.2 which, by changing the notation, we
rewrite as follows
(3.118)
Our goal will be to determine appropriate polynomials P(q"^), R(q~') and T(q"'), such that the
closed loop responce possess desired prorerties. For example, we can ask that the poles of the
closed loop transfer function be in prespecified locations (pole placement). The closed loop
responce can be determined by combining 3.118 with the process model 3.53. This gives
q d B(q- 1 )R(q- 1 )y r (k) + C(q- 1 )P(q- 1 )e(k)
1
(3
u 9 )
y-
y?"',
,y,.
(3-.20)
where the process model is given by 3.77 (For extensions to the stochastic case see [5],[6]). Our
objective is to choose the P(q _1 ). T(q _1 ) and R(q _1 ) such that
y(k)=qdBm((;
yr(k) ,
(3.121)
where B m (q" 1 ) and A m (q" 1 ) represent the desired zero and pole locations, respectively. That is,
we require that
193
y(k)
q^B(q-1)R(q-1)
qXCq' 1 )
Yr(k)
A^q"1)
(3
m )
By comparing the nominators of the above equation we see that, first, some (or all) of the zeros of
the open loop system (roots of B(q~l)) can become zeros of the closed loop system, and second,
other zeros can be added by means of R(q~l). On the other hand, roots of B(q"l) which are not
desired as roots of B m (q~l), have to be cancelled by the denominator of the middle term of 3.122.
Hence, they should be included in P(q~l). Specifically, let B~(q~l) contain the open loop zeros
which appear also in B m (q~l), and let B+(q~l) contain those which are cancelled. Then according
to the above discussion we must have that
B(q']) = BV'jB-fq 1 )
(3.123)
P ( q 1 ) = P 1 (q- 1 )B + (q- 1 ) .
(3.124)
and
u(k)=^J-yr(k)-^3J.y(k) .
P(q _1 )
P(q _1 )
(3.125)
This implies that the roots ofP(q"l) should be inside the unit circle, for the control law to be
stable. Therefore, B + (q~l) must contain only well-damped zeros of the open loop transfer
function. That is, only these zeros can be cancelled. The rest must be included in the zeros of the
closed loop transfer function. For this reason B m (q~l) will be given as
BJq" 1 ) =
ffJq^BV)
(3.126)
where B m '(q"^) contains additional desired zeros. As we mentioned before, these will be
provided by the polynomial R(q~'), for then
RCq1) = R ^ J B ' J q 1 ) .
(3.127)
We didn't take just R(q" 1 )=B m '(q" 1 ), because the term Rj(q"l) will be useful afterwards. In the
light of the above definitions, the left-hand side of 3.122 becomes
194
y(k)
yr(k)
A ^ P , ( q ^ B V 1 ) V B V ^ B X q ^TCq1)
q^B^q-^R.Cq"1)
ACq^P^q 1 ) + q^BCq-^TCq1)
(3.128)
Until now we dealt with the closed loop zeros. The corresponding poles can be determined by
the denominator of 3.128. Normally, we would equate this term with A m (q"l) and solve with
respect to Pi(q"') and T(q~l). A slight generalization is to solve the equation
ACq-bP/q^ + qVCq^TCq^AJq-bRjCq" 1 ) .
(3.129)
This will assure that a solution exists independently of the order of A m (q" 1). Notice that R \ (q" *)
has already been included in the nominator of 3.128 (by means of R(q~l)), so as to be cancelled.
Hence, we have that
y(k) _ q-Xtq-Xtq-1)
Yr(k)
A m (q- 1 )R 1 (q 1 )
dr>
/-l>
q" a B m (q~')
AJq"1)
(q.e.d.) .
(3.130)
The above procedure can be applied also to non-minimum phase systems, since only welldamped zeros are cancelled. Equation 3.129 is the so-called Diophantine equation, and it is
discussed extensively in [5],[6],[14]. Note that it is a generalization of 3.80. In fact, many of the
control design procedures presented in section 3.4.2 can be shown to be special cases of this zeropole placement technique [32].
An indirect self-tuning controller based on the above procedure can be summarized in the
following algorithm :
Algorithm 3.3 (Indirect scheme)
1) Specify polynomials A m (q _ 1 ), Bjjj'tq"1) and Ri(q _1 )
2) Estimate A(q~l) and B(q~l) (model parameters)
3) Determine B+fq"1) and B-(q_1)
4) Solve3.129forP 1 (q- 1 )andT(q- 1 )
5) Calculate Ptq" 1 ) and Rfa"1) by 3.124 and 3.127
6) Compute u(k) by 3.125
7) k=k+l ; go to 2
In order to get a direct scheme we first multiply 3.129 by y(k), and then combine the result with
the process model in 3.77. This yields the equation
A ^ q ^ R / q ' ^ y C k ) = q d B-(q" 1 )[T(q- 1 )y(k)+P(q 1 )u(k)] .
(3.131)
195
The idea is to use this equation in order to directly estimate the controller polynomials T(q"l) and
P(q"l). But as we can see, 3.131 is not linear in the parameters (i.e. it doesn't have the linear
form of equation 3.28) because B~(q"l) multiplies both T(q~l) and P(q~l). In the simplest case,
we can make it linear by taking B"(q"l)=l, but now the method is not applicable to non-minimum
phase systems since all the process zeros are cancelled. Other approaches dealing with this
nonlinearity are described in [5],[27],[32].
3.4.4. Model Reference Adaptive Controllers. A block diagram of a model reference adaptive
control system (MRAS) is shown in figure 3.4 (sec. 3.1.2). The idea is to express the desired
behaviour of the closed loop system by means of a reference model, and then design the controller
in such a way so that the closed loop system "follow" the reference model as close as possible.
Originally, model reference adaptive control systems were developed in the continuous-time
domain where the process was described by state-space equations [11],[12],[13]. Later, the
interest was moved more to the transfer function representation and also to the discrete-time
domain [5],[6].
Traditionally, the design of MRAS was based on stability criteria, which means that the
adaptive control laws were being derived in such a way so as to assure the stability of the closed
loop system. In order to illustrate the early ideas we now present an example where the design of
the adaptive controller is based on the Lyapunov stability criterion [37]. Assume that the process
is described by a first-order differential equation
y(t) =-ay(t) + bu(t) ,
(3.132)
where y(t) is the process output, u(t) is the process input, and a,b are constant but unknown
parameters. The problem is to find an adaptive control law such that y(t) follow the output y m (t)
of a reference model, given by
y m W = -a m y m (t)
b m u m (t) .
(3.133)
(3.134)
(3.135)
The above equation implies that if a and b were known, then a choise of C2(t)=b m /b and
ci(t)=(a m -a)/b would suffice. But, since a and b are unkown constants, we proceed as follows.
We first define
e(t) = y m (t) - y(t) ,
to be the output error. Then, by combining 3.133 with 3.135 we get
(3.136)
196
(3.137)
(3.138)
d 2 (t) = b m - b c 2 ( t ) .
(3.139)
and
Our goal is to appropriately adjust dj(t) and d2(t) so that e(t)0, dj(t)>0 and d2(t)>0 as
t>+. Let d^(t) and d2(t) satisfy differential equations of the form
d,(t) = g L (e, y, y m )
(3.140)
d 2 (t) = g 2 (e, y, y m )
(3.141)
V(e,d1,d2) = l
b-sgn(b)-e2(t)+^-dj(t)^-d2(t)
,XxX2>0
(3.142)
is a candidate Lyapunov function for the dynamic system described by the equations 3.137, 3.140
and 3.141 [7]. This system will be stable, according to the Lyapunov stability criterion, if the
derivative of V(e,di,d2) with respect to time is negative semidefinite, and this is achieved by
choosing [7]
d 1 (t) = X 1 [bsgn(b)]e(t)y(t)
(3.143)
d 2 (t) = - y b s g n ( b ) ] e ( t ) u m ( t ) .
(3.144)
and
(3.145)
(3.146)
and
197
respectively, and they represent the adaptive laws. Notice that the sign of the parameter b is
assumed to be known. Also, in order to get a uniformly asymptotically stable system, further
properties on the reference input signal u m (t) must be imposed. Finally, since the adaptive laws
3.145 and 3.146 directly adjust the controller parameters, the above procedure corresponds to a
direct MRAS. For details, as well as extensions of the above method to more general cases, the
reader is referred to [7].
We now consider the case where the process is described by a transfer function model, in the
discrete-time domain. Specifically, we assume that the process model is given by 3.77 which we
rewrite below as
ACq 1 ) y(k) = q^BCq' 1 ) u(k) .
(3.147)
The objective is that the process output y(k) follow a reference output y m (k) determined by a
reference model given below
Am(q4)ym(k) = q'dBm(q1)um(k) .
(3.148)
In order to derive a suitable controller form, let Pj(q"l) and T(q~l) be polynomials satisfying
the algebraic equation
AmCq"1) = P / q V t q ' 1 ) + q"dT(q X) .
(3.149)
(3.150)
(3.151)
(3.152)
would cause y(k)=y m (k), in the case of known model parameters. If they are unknown, they can
be estimated using equation 3.148 together with a parameter estimation method. This procedure
results in a indirect MRAS. If, on the other hand, equation 3.151 is used instead, then the
controller parameters can be directly estimated, thus resulting in a direct scheme.
By comparing equations 3.149 and 3.152 with 3.129 and 3.118, respectively, we notice that
the above model reference adaptive control scheme can be considered as a special case of the pole
placement design technique, by taking u m (k)=y r (k), B~(q~l)=l and Ri(q~l)=l, corresponding to
the case where all process zeros are cancelled and no additional zeros are introduced.
198
4.
Conclusions
Adaptive control systems result from the combination of two basic techniques, namely, controller
design and parameter estimation. In the general case, the resulting system is a time-varying
nonlinear dynamic system which may be very difficult to analyse. For this reason, theoretical
results exist only for some classes of adaptive control systems.
In the previous chapters we gave a description of self-tuning controllers and model reference
adaptive systems which represent two of the most important classes of adaptive systems for which
stability and convergence results have been established. Also, numerous applications of these
schemes have revealed that they are quite effective in controlling a large number of industrial
processes. Both methods require the existence of a mathematical model which should adequately
describe the process. Usually, the model is given in discrete-time representation since the
implementation is done by means of digital computers. Typical examples are the so-called
ARMAX models. Based on such models, various controller design methods can be applied, as for
example, generalized minimum variance, pole placement, model following, e.t.c. Finally,
identification schemes based on the recursive least squares (RLS) are usually applied since they
appear to be the most appropriate for real time applications.
5.
References
[I]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[II]
[12]
[13]
[14]
199
[15] E. I. Jury, "Sampled-data systems revisited: Reflections, recollections and reassessments ",
Trans, of ASME, Vol. 102, pp. 208-217, Dec. 1980.
[16] L. Ljung and T. Soderstrom, "Theory and practice of recursive identification ", MIT Press,
1983.
[17] B. Friedlander, "System identification techniques for adaptive signal processing ", Circuits
Systems and Signal Processing Journal, Vol. 1, No. 1, pp. 3-41, 1982.
[18] K. J. Astrom and P. Eykhoff, "System identification-A survey ", Automatica, Vol. 7, pp.
123-162, 1971.
[19] B. Wittenmark and K. J. Astrom, "Practical issues in the implementation of self-tuning
control", Automatica, Vol. 20, No. 5, pp. 595-605, 1984.
[20] B. D. O. Anderson, "Adaptive systems, lack of persistency of exitation and bursting
phenomena ", Automatica, Vol. 21, No. 3, pp. 247-258, 1985.
[21] D. E. Seborg, T. F. Edgar and S. L. Shah, "Adaptive control strategies for process control:
A survey ", AIChE Journal, Vol. 32, No. 6, pp. 881-913, June 1986.
[22] S. T. Alexander, "Adaptive signal processing: Theory and applications ", Springer Verlag,
1986.
[23] C. E. Garcia, D. M. Prett and M. Morari, "Model predictive control: Theory and practice A survey ", Automatica, Vol. 25, No. 3, pp. 335-348, 1989.
[24] D. W. Clarke, C. Mohtadi and P. S. Tuffs, "Generalized predictive control - Parts I,II",
Automatica, Vol. 23, No. 2, pp. 137-160, 1987.
[25] D. W. Clarke and P. J. Gawthrop, "Implementation and application of microprocessorbased self-tuners ", Automatica, Vol. 17, No. 1, pp. 233-244, 1981.
[26] A. Papoulis, "Probability, random variables, and stochastic processes ", McGraw Hill,
1984.
[27] K. J. Astrom, " Adaptive feedback control ", Proceedings of IEEE, Vol. 75, No. 2, pp.
185-217, Feb. 1987.
[28] C. Kiparissides and S. L. Shah, "Self-tuning and stable adaptive control of a batch
polymerization reactor ", Automatica, Vol. 19, No. 3, pp. 225-235, 1983.
[29] W. R. Cluett, S. L. Shah, J. M. Martin-Sanchez and D. G. Fisher, "Adaptive predictive
control of a polymer reactor ", 32nd Canadian Chemical Eng. Conference, Vancouver, pp.
1313-1323, Oct. 1982.
[30] V. K. Tzouanas and S. L. Shah, "Adaptive pole-assignment control of a batch
polymerization reactor ", Chemical Engineering Science, Vol. 44, No. 5, pp. 1183-1193,
1989.
[31] A. V. Papadoulis, C. A. Tsiligiannis and S. A. Svoronos, "A cautious self-tuning
controller for chemical processes ", AIChE Journal, Vol. 33, No. 3, pp. 401-409, March
1987.
[32] K. J. Astrom, "Theory and applications of adaptive control-A survey ", Automatica, Vol.
19, No. 5, pp. 471-486, 1983.
[33] D. W. Clarke and P. J. Gawthrop, "Self-tuning control", Proceedings of IEE, Vol. 126,
No. 6, pp. 633-640, June 1979.
[34] P. J. Gawthrop, "Some interpretations of the self-tuning controller ", Proceedings of IEE,
Vol. 124, No. 10, Oct. 1977.
[35] K. J. Astrom and B. Wittenmark, "Self-tuning controllers based on pole-zero placement",
Proceedings of IEE, Vol. 127, pp. 120-130, 1980.
[36] D. W. Clarke, "Model following and pole-placement self-tuners ", Optimal Control
Applications and Methods, Vol. 3, pp. 323-335, 1982.
200
|37]
K. S. Narendra and P. Kudva, "Stable adaptive schemes for system identification and
control", IEEE Trans, on Systems, Man, and Cybernetics, Vol. SMC-4, No. 6, Nov.
1974.
[38] K. J. Astrom and T. Hagglund, "Automatic Tuning of PIDControllers ", ISA, 1988.
ABSTRACT. In the early stages of a runaway reaction, when the rate of heat generation exceeds
the rate of heat removal by a small amount, it may be possible to restabilize the control of the
reactor by taking emergency actions. The problem is to detect these potentially hazardous
situations in sufficient time to allow the necessary counter-measures to be taken. In this paper, a
general overview of the different techniques for early detection of potentially dangerous situations
will be given and the advantages and disadvantages of each technique will be discussed.
1. Introduction
Normally, the temperature of a reactor in which exothermic reactions take place is controlled by a
cooling system. If, for some reason (e.g. loss of cooling, loss of mixing, etc.), the rate of heat
generation exceeds the rate of heat removal, the temperature of the reacting mass will begin to rise.
This will cause an increase in the rate of heat generation, due to the exponential dependence of the
reaction rate on the temperature, and the process will continue to accelerate producing a large
amount of heat in a very short time with the consequent dangers for people, installations and
environment.
However, in the early stages, when the rate of heat generation exceeds the rate of heat removal
by a small amount, it may be possible to restabilize the control of the reactor by taking emergency
actions such as full cooling, fast injection of a suppressant or dumping the reactor contents. The
problem is to detect these potentially hazardous situations in sufficient time to allow the necessary
counter-measures to be taken to avoid temperature and pressure excursions associated with the loss
of control of such processes. The early detection of these potentially hazardous situations is,
consequently, of great importance in the safe and economic design and operation of a plant.
Control of potential thermal explosion hazards once had to rely on laboratory measurements of
process chemistry and direct control of process variables within fixed limits on the plants. In recent
years, new techniques have been developed allowing a better understanding of the chemical
reaction [1,2] and improving the methods to control it.
Although the so called off-line tests, performed under laboratory conditions, are necessary, and
should be carried out for each new process, there are some disadvantages which must be taken into
account when the results are evaluated [31. Off-line tests are not totally representative because the
properties of materials being used in the plant are never exactly the same as those of a laboratory
201
A. Benuzii and J. M. Zaldivar (eds.). Safety of Chemical Batch Reactors and Storage Tanks, 201-226.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
202
sample. Another disadvantage is the fact that in these tests there is an implicit assumption that the
"worst case" conditions have been identified, and consequently there is the possibility that
unexpected hazards will remain undiscovered. Moreover, in the real processes there are
unpredictable disturbances that affect the process behaviour and can not be simulated in the
laboratory tests. However this can be partially compensated by performing the tests under
conditions (temperature, pressure, initial concentrations, etc.) more severe than those forseen for
the process.
Complementary to these tests, there is another type of procedure to recognize the potential
hazard. These procedures are called on-line supervision, and are carried out in real time, with the
real reaction mixture, equipment and operating conditions. Because of the development and the use
of digital computers, the level of sophistication has increased from simple supervision of directly
measurable variables to complex signal analysis, treatment, and estimation techniques that allow
prediction of state variables or parameters, not directly measurable, to be included in the criteria for
hazard detection. The disavantage of on-line methods is that the information appears only when the
process is outside of the desired conditions and normally in a dangerous state. From all these
considerations, it can be shown that both techniques (off-line and on-line) should be
complementary.
Table 1. Differences between off-line and on line prevention measures [3].
Offline
On-line
laboratory
actual process materials
sample
process equipment
simulated conditions
real conditions
without disturbances
with disturbances
sensitive to the choice of test conditions
non sensitive
conditions more severe than real process
information obtained before process design information appears only when the
process is outside desired
conditions
There is a fundamental requirement that early detection should provide sufficient time for plant
operators to correct the deviation from safe operation. However early on-line detection of
hazardous states is difficult [4] because a chemical process is described by a large number of state
variables, such as temperatures, pressures, concentrations, etc. and only some of these can be
measured on-line with an acceptable response time to allow the information to be used for the
detection procedure. Particularly for batch chemical reactors, the difficulties are increased due to the
wide range of processes that are carried out, and their complexity, strong non-linearity and timedependences (in a batch cycle there is no steady state).
203
y =f{x, n, w, }
(1)
where n(t) and y(t) are the measurable input and output variables, w(t) represents non-measurable
disturbance signals from the process and its manipulating and measuring equipment, nonmeasurable process parameters, and x(t) non-measurable state variables. The process parameters
are constants or slow time-variable coefficients, while the state variables are time-dependent.
Disturbances
Control
variables
>< (t)|
Control
variables
Measured
variables
PROCESS
DETECTION
SYSTEM
Alarm?
Figure 2. Block diagram of a detection system.
The methods for early on-line detection can be divided into two categories depending upon the
204
y(
<yXt) < y;
(2)
This is referred to an absolute value check. If the measured variable exceeds the set limit, alarm
or automatic counter-measures must be initiated. The most common is temperature supervision [3],
but recent theoretical studies show that detection procedures based on pressure are more suitable in
certain conditions [6]. However for practical applications the sensitivity and reliability of the sensor
play an important role to determine the choice of the detection system.
A number of other variables: pH, viscosity, thermal conductivity, etc. are also easily measurable
and can be used for some processes, for example, oxidation can be a dangerous secondary reaction;
therefore the oxidation-reduction potential (Redox) of the reacting medium is a good measure in
order to detect initiation of these reactions at its earliest stage.
The limits are usually so set that a large enough distance to the non-return point is retained on the
one hand, while avoiding false alarms on the other. In general these measures are easy and cheap to
install and can have good predictive capacity, but they are completely dependent on knowledge of
the process and are unsuitable for detection of unexpected dangers.
2.2.2. Supervision of the rate of increase and/or acceleration. The limit check can also be applied to
the derivative of the signal. The hazard identification criteria in this case is:
dy=
dt
dy ;
ty
< < dt
dt
(3)
oV,
<J.2J
4<
dt 2
^
dtA
(4)
205
pressure excursions of the reacting mass due to the large amounts of heat released in a very short
time. That means that the derivatives of temperature and pressure or derivatives of the rate increase
can be used to predict the runaway excursion.
The method of supervising the rate of temperature or pressure rise is not as cheap or as simple as
temperature or pressure measurement, because amplification and filtering are necessary before
reliable derivative can be calculated. Since the "safe" temperature or pressure need not be specified,
the independence and selectivity of this method is higher, but depends also on specific kovvledge of
the system in order to define the limits.
The method of monitoring the acceleration is similar to the previous method; the predictive ability
is however higher.
2.2.3. Frequency analysis. A physical process can be described either in the time domain, by the
value of some quantity as a function of time, or in the frequency domain, where the process is
specified by giving its amplitude as a function of frequency. The method to change from one
representation to another is by means of the Fourier transform equations:
"J
y(t)
(5)
(6)
Output signals Y(t) often consist of lower frequency components with large magnitudes which
mainly determine the nominal values of the signal, and higher frequency components with small
amplitudes which give additional information on the inner state of the process. Some attempts have
been made to identify frequency signal patterns and to pinpoint process errors from the changes in
their corresponding frequency behavior. A simple approach [7] consists of representing the
frequency response by its amplitude ratio and phase angle (see fig.3) and then defining the upper
and lower limits tolerated in the case of a change in the typical pattern obtained when the process is
operating under normal conditions.
Real axis
Figure 3. Polar representation of the transfer function with the tolerance limits.
206
Simple bounds at one frequency are not very effective for fault detection. Instead, fault
dictionaries need to be prepared in which bounds are placed at several different frequencies or more
complex characteristics need to be examined [8].
2.2.4. Recipe-based supervision. The hazard identification criterion is the deviation of some
characteristic values, or of status of different parts of equipment from the "recipe". The procedure
consists of observing the time record of one or more process variables or parameters, computing
simple statistics of the variables, and carrying out elementary tests to detect faulty operation.
Normally the practical application of this method is through control charts [7], which are a
graphical means of representation and analysis (see figure 4).
When the process is subject only to variations due to random fluctuation, such as that
engendered by environmental changes, internal mixing conditions, etc. then it is possible to say that
it is "in statistical control". If some change takes place via a non-random change, such as
engendered by a deterministic component added to the process variable, the process is "out of
statistical control". Consequently, "in control" means that the same probability distribution will
continue to represent the observed variable as the process goes on. The objective of control charts
is to provide a visual observation of the measured variable and to detect the category change as
soon as possible after it occurs.
. '
Time
Figure 4. A process quality control chart.
This method is completely specific for a process, fully dependent on knowledge, data available,
and judgements. For these reasons it is better for continuous processes at steady-state rather than
for batch processes in which all the variables change with time and there is no steady-state that can
be used as a reference.
The predictive ability and selectivity are given by the quality of the evaluation and specification of
the hazard limits.
2.2.5. Detection of the progressive increase of heat evolution. The hazard identification criterion is:
'Generated
r\
(7)
dt
where qc.eneraied ' s t n e P o w e r generated by chemical reaction. The principle applied is based on a
simple heat balance:
207
dt
US (T m -T c )
(8)
Power generated = power used to increase the temperature of the reaction mixture + power
removed by the jacket.
From expression (8) it is possible to obtain:
dt
dt2
(9)
dt
.......t... >o mm
^Potentially dangerous
Jzone
J~r
dr
>0
and
d(T m -T c )
dt
(10)
208
For the evaluation of these criteria there is a comercially available system, that was developed by
Hub [3], and is called OLIWA. Figure 6 shows an idealized form of the flow diagram of the
OL1WA system.
OLIWA MONITORING
SYSTEM
VARIABLES,
PROCESS
PARAMETERS
OR
If a mathematical model of the process exists, the state of the reactor can be reconstructed from
measurable variables, which will allow predictive calculation of the future status or at least,
evaluation of new criteria based on these non-measurable quantities. That means that highest
predictive power and selectivity can be reached and hence better early detection of hazardous states
can be obtained.
Figure 7 illustrates the general structure of a model-based detection system. Firstly, all the
known information about the process is put in the form of a mathematical model that normally
consists of a set of algebraic and differential equations. This model is solved on-line by numerical
procedures in order to obtain the whole state of the batch reactor.
209
In addition to the available measured variables, the model-based detection system must be
supplied with all the control variables, and the initial and operating conditions. The simulated
variables must be compared with the measured variables; a non-zero difference will indicate an
incorrect calculation in the model that can be due to unknown disturbances, unknown initial
conditions, erroneous parameters, etc. Consequently, the model must be corrected from process
measurements. The method to correct these deviations is by minimizing the error using estimation
techniques, for instance, a state variable observer (deterministic case) [10] or state variable filter
(stochastic case) can be used [11].
Disturbances
Control
variables.
PROCESS
Measured
variables^
Error between
predicted and
observed respons
0
Mathematica
model of the
process
State variables,
parameters,
xiiaiacteristic quar tities
Estimation
techniques
Pattern
recognition
(Criteria)
Model-based
detection s y s t e m
*
Alarm ?
Figure 7. Block diagram of a model-based detection system.
Once the whole state of the system is estimated and the error between predicted and observed
responses has been minimised by modifying parameters in the model, different criteria can be
applied using data estimated.
It is possible to divide these methods, depending on the type of quantities in which the safety
criteria is based [5]:
- Non-measurable state variables
- Non-measurable process parameters
210
,'
x2
T T"
l m - lir^
Ea
T_
RTm
(11)
They demostrated that for a batch reactor in which a simple reaction, with decomposition behaviour
(nA products) , takes place and in which a proportional controller regulates the temperature of
the jacket, the phase trajectories show a positive inflection when the reaction is self-accelerating,
and that all points in the phase plane for which the conditions:
dx_
->0
dx
i
and
d x
-?->0
dx^
(12)
(13)
and hence,
dq,
^T=VmAH.|
(14)
supposing that the volume of the reaction mixture does not change considerably during the reaction
(approximately true for batch processes).
This type of criteria can be extended and applied only if the conversion can be measured or
inferred on-line. The measurement of this variable is difficult or expensive in most cases,
consequently model based techniques for estimating the conversion are necessary for the use of
such as criteria.
Another similar approach was followed by Bonvin and Saner [13]. They developeded an on-line
procedure for supervising the operation of batch reactors with the aim not to detect runaway
initiation but the entrace of a disturbance in the system. The general procedure was based on the
estimation of the rate of heat production by two different methods. The former used kinetic
reconstruction from temperature and time measurement, eq. (13), to infer the conversion and the
heat evolution of the reaction. The latter used calorimetric reconstruction based on a model of the
211
reactor to estimate the total heat which evolves in the reactor, eq. (8). The difference between the
two estimates indicated that a disturbance had entered the system. For example, if during a reaction
there is a disturbance in the heat transfer coefficient (U) due to a stirring problem, the estimation of
the rate of heat production using the calorimetric approach will not be correct because it is based on
the calculated value of U by simulation, while the kinetic approach does not depend on the reactor
model, and consequently will not be affected.
2.3.2. Non-measurable process parameters. Process parameters are constants or slow timedependent coefficients which appear in the mathematical description of the relationship between the
input and the output of the mathematical model, i.e. overall heat transfer coefficient (U), heat
transfer area (S), kinetic parameters, heat capacity (Cp), etc.
There are no applications in literature for tine use of these parameters for safety criteria for batch
or semibatch chemical process, but for instance, if the kinetic parameters are changing it is normal
to infer that an unexpected reaction has taken place and this could be used for detecting
decomposition reactions. The following method, practically the same as that for non-measurable
process variables, should be used for this purpose [5]:
- Establishment of the process equation for the measurable input and output variables by theoretical
modelling.
Y(t) = f { U ( t ) , }
(15)
- Determination of the relationship between the model parameters 6; and the physical proce:SS
coefficients pj:
= f(p)
(16)
- Estimation of the model parameters 9; as results of measurements of the signals Y(t) and U(t).
- Calculation of process coefficients:
P = f '()
(17)
212
n = g{u,Y}
(18)
CHj-CHOH-CH^CH
H+
CH -CH -COOH
3 2
CHj -CH2-COOCH-CH -CH
CH
The system investigated was a 1:1 molar mixture of 2-butanol and propionic anhydride. The
process was carried out in a RC1 reaction calorimeter [2] of 2 1 of volume. Initially the =6.8 mols
of 2-butanol containing 0.8% in weight of sulphuric acid were added to the reactor, and allowed to
reach thermal equilibrium, the same number of mols of the propionic anhydride were then rapidly
introduced (=10 s.). Different series of experiments in isoperibolic (modifying the jacket
temperature) and isothermal conditions (varying the reactor temperature set-point) were performed.
3.1.2. Isoperibolic experiments [14]. Figure 8 represents a set of isoperibolic experiments
(constant jacket temperature).
The initial drop in the temperature of the reacting mass is associated with the endothermic
mixing of the reagents. The effect of jacket temperature on the rate of reaction can be seen from the
changes in the reactor temperature-time profiles, with high temperatures leading to exothermic
runaway.
Table 2. Jacket temperature for the different isoperibolic experiments.
TeHO
Experiment
El
293.9
E2
295.7
298.2
E3
E4
300.7
E5
303.2
308.2
E6
213
1200
2400
n
1
1
1
360048006000720084009600
r
tBOO
12000
Time (s)
~\
1300
2600
3900
5200
6500 7800
Time (s)
9t
tt+00
11700 13000
214
1
150
I
300
I
450
1
600
1
750
1
900
1
C50
I
1200
050
Time (s)
Figure 10. Esterification reaction: Isothermal experiment E9. Temperature-time profiles for the
reactor (Tm) and jacket (Te) respectively.
3.2. OLIWA TESTS
The OLIWA system was tested using the esterification reaction described above. The OLIWA
system has three different alarm levels [15]: a/ Pre-alarm (flashing light with increasing on-time
and constant cycle), b/ Alarm (acoustic signal and flashing light together, with increasing on-time
and constant cycle), c/ Extreme danger alarm (continuous acoustic and optical signals).
Two independent temperature sensors were placed in the reaction calorimeter in order to
measure the reactor and jacket temperature, and connected to the OLIWA system. The signals of
these two measures, the first and second derivatives of reactor temperature, as well as the first
derivative of the temperature difference between reactor and jacket, and the alarm level were
recorded independently of the RC1 data acquisition system.
215
376
"T
752
T
1128
n
H
1
B80
1
2256
1
2632
r
3008
3384
3760
T-^
2632
1
3008
1
3384
3760
Time(s)
d2Tm/dt2W
^f-
Akrm level
T
1128
n
"604
1
B80
^
2256
Time(s)
Figure 12. Values of the two variables of the criteria used by the OLIWA system and alarm level in
experiment E5.
Figure 11 and 12 show a typical isoperibolic experiment and the OLIWA results. There are two
alarms, the first at approx. 927 s and the second at approx. 1408 s, which convert into an extreme
danger alarm, while the time for the maximum temperature is approx. 1969 s.
216
The results of the OL1WA experiments show that potentially dangerous reaction states can
be detected in the early stages if the reactions proceed slowly at the beginning. In the case of
autocatalytic or radical type behaviour the time for taking counter-measures is reduced due to the
fact that the self-heating process is enhanced by the effect of the concentration in the Arrhenius
expression for the reaction rate and in some circumstances the OLIWA is not useful.
Moreover, the existence of an alarm does not imply that the system will runaway (see figure 8,
El), because perhaps all the reactants will be consumed. Consequently, the installation of such a
device should provide a warning, but should not substitute the judgment of the people in charge of
the plant.
3.3. DEVELOPMENT OF AN IN-HOUSE WARNING SYSTEM
The OLIWA system is a dedicated equipment with its own micro-processor and can be used
simultaneously for five reactors or storage tanks. In the case of a single system, if a reliable on-line
estimate of both temperature derivatives can be made, then the criteria given by eq. (10) can be
used as a part of the warning system of the installation, and this can be carried out in a personal
computer or as part of the control system.
3.3.1. Signal treatment and noise suppression. The numerical methods for differentiation, i.e. eq.
(19) for five centered points and eq.(20) for nine centered points are affected by the disturbances
superimposed on the measurement signal. This can lead to rate estimates that amplify this effect
making the use of these values for early warning detection very difficult (see figure 13.a).
^- 1
"280h
(19)
F3f(xn+4h)+16f(xn+3h)-36f(xn+2h)+48f(xn+h)-48f(xn-h)
o
o
o
o
o
f-36f(x0-2h)-16f(x0-3h)+3f(x0-4h)]
(20)
(21)
217
(22)
a=
At
+1
where At is the sampling period and if the time constant of the filter. Limiting cases for a are: oc= 1
(no filtering), a=0 (the measurement is ignored).
Another similar filter is the double exponential or secondorder filter, which offers some
advantages for eliminating highfrequency noise (see figure 13.c).
yn = Yaxn+( 2 Ya)y n _ r (la)(ly)y n
(23)
6
4
dflm-Te)/ U2
2
0
lETm/dO'ld
-2
(CTm/dO* 1*3
-4
6
-t
-D
-1
85
370
555
740
r-
525
Tn(i)
110
1295
H80
665
370
555
740
925
fine (a)
III
1565
B50
b/
a/
C-j
dam-T)/dl'li2
^
0
(CTm/dt2*l*3
_v
*"
(OTm/iM'UJ
-4-6
IB
370
555
-1
1
740
925 110
TVre(i)
C/
T"
K80
-tB65 - B50
370
555
740
925
Trt(j)
IID
1295
1460
K5
850
d/
Figure 13. Derivatives of criteria from eq. (10) in experiment E5. a/ Single exponential filter,
Tp=20 s, b/ Double exponential filter tpi= 7 s and Tp2= 8 s, c/ Moving averagefilter,j = 20 points.
Moving average filter: This filter averages a specified number of past data points, by giving equal
weight to each data point (see figure 13.d).
218
n
yn
(24)
i=n-j+l
where j is the number of past data points that are being averaged.
Comparing the three different filters and using the same arbitrary margin for the level of
derivatives and for the time interval in which they must exceed these limits, the detection is
achieved at 1556, 1504 and 1400 s respectively. The same type of treatment was carried out for
experiment E9, the results are shown in figure 14. In this case the time for the maximum
temperature was 556 s. Using the same criteria, the detection is achieved at 364, 348 and 328 s
respectively.
Better results can be obtained applying adaptive filters similar to exponentials but using variable
averaging weights [18] or Kalman filters [11]. The disadvantage is that extra computation is
required but they provide standard error estimates that can be used in formulating decision rules,
reducing the false alarm risks.
60-
T
-
dCTm-WUtf
dCIm-T.y*'li2
4836-
36-
A
y
24E-
24-
-^y
00-
<tZrm/*2'U3
-0-
-0-
-24-
-24-
-36-
-36-
-48-
-48-
91
112
63
204
255 306
Tn(i)
357
40B
459
5D
51
IK
(BWdB'UJ
83
204
36-
403
459
dCTm-T.Vdl U2
dCIrn-W* U2
4836-
24-
24-
0-
0-
0-
<ZI>n/iK'U3
-a-
-B-
-24-
-24-
-36 J
n36-
-48-
-48-
51
60-
4t-
357
b/
a/
60-
255 306
Trab)
IQ
E3
204
255 306
Timd)
357
405
459
50
/
/
51
112
dTIm/da'l*!
B3
204
255 306
ThW
357
l
408
459
d/
d
Figure 14. Derivatives of criteria from eq. (10) in experiment E9. a/ Single exponential filter,
Tp=20 s, b/ Double exponential filter TFI= 7 s and Tp2= 8 s, c/ Moving average filter, j = 20 points.
219
Moreover, in order to avoid false alarms due to the noise, it is necessary to define a certain limit
value greater than zero. However, this will reduce the sensitivity for small values of derivatives,
that means, the slow starting self-heating processes will be detected later. In order to correct this
problem is interesting also to evaluate [15]:
dt z
dt>e.
(25)
psdt I ^ d t > ,
(26)
1
160
I
320
1
480
1
640
1
800
1
960
I
1120
I
1280
1440
1600
Time (s)
Figure 15. Experimental and simulated temperatures of the reacting mass and the heating/cooling
jacket in a runaway scenario in the RC1 Reaction Calorimeter for experiment E9.
3.4.1. Off-line application. As the simulator exists, the application of criteria given by eq. (10) and
other criteria based on conversion or another non-measurable state variable or parameter may be
carried out. The advantage of this approach is that the experimental noise can be eliminated. Figure
220
16 shows the application of the criteria given by eq. (10) to the simulated results.
dfJm-Te)/dtte2 Simulated
204
255
Time(s)
306
357
408
459
5t)
Figure 16. Derivatives of criteria from eq. (10) using simulated data.
Figure 17 shows the evolution of derivatives of criteria given by eq. (12).
60
0.02
0.03
0J04
0.04
i
0.05
i
0.06
i
0.07
r
0.08
0.08
0.09
x1
Figure 17. Derivatives of criteria from eq. (12) using simulated data.
3.4.2. On-line developments . Up to now, the model-based criteria for early detection of runaway
221
initiation have been applied off-line, but this technique has been developed to solve on-line the
mathematical model of the process and to correct it by comparison with the real measurement [1922].
Some applications for incipient fault diagnosis in chemical process have been reported [23], and
particularly, Gilles, King and Schuler [4,12, 24] who studied the application to detect hazardous
states in a batch chemical reactor.
The first step for a development of such system is the reconstruction of the state of the system
and/or the estimation of some parameters. In order to achieve it, the most common method is the
Kalman filter [11], but other methods can be used (i.e. Lainiotis filters [25], etc.).
The model of a chemical process, in this case a batch reactor, can be be represented by the
following expression:
x = F ( x r x 2 ,...,x n > QVQT...,9p)
(27)
where x; are the state variables and 0j the model parameters. Through linearization by Taylor series
around any operating point, it is possible to obtain the following stochastic state-space model:
(28)
y(t)=Mz(t)+v(t)
(29)
l=*
yk = M k X k
+ r
+ V
+ W
(30)
(3D
where, 4>= eAAl, T= A-\fy-'iyB and w is an r dimensional gaussian random noise for the inputs.
The equations of discrete Kalman filter recursive algorithm are summarized in Table 4. A more
detailed and rigorous description is available elsewhere [26,27]. The discrete Kalman filter gives
an estimate x of the state x at the time sample k+1. This estimate is based on the previous estimate
x k and the previous measurement yk. Kk is the filter gain vector that minimizes the covariance P of
the estimation error. Moreover, if some parameters 9, need to be estimated along with the original
state vector, the Kalman filter algorithm can be extended by introducing these parameters in the
state vector [26,27].
222
- Prediction:
*k+i = *k + iV r A
p
+
k+1 =iWLi Qk.
- Correction:
K
^k+1
=P
^k+l^k+ll
\+l =
Qk = K-f}
\+l^Jy^-M*^
k+U
R
. ^k
= ^V V k}
k+,=Pk+,-Kk+lMk+,Pk+l
King and Gilles [4,24], in order to take into account that different fault modes have to be
considered, and that the warning system in order to be efective should be able to distinguish
between them, applied different filters for every possible fault model (see figure 18) and using the
information provided by the filter about the quality of estimates it was possible to discriminate
between rival models using the Bayes rule [28].
Process
Filter 1
Normal process
Filter 2
Fault type 1
Filter 3
Fault type 2
Filter 4
Fault type 3
Figure 18. Multiple filter method for fault discrimination using the Bayes rule to determine the
model which fits best the measurements.
5. Conclusions
Early warning detection based on temperature derivatives is feasible and should serve as a
223
warning for plant operators. Moreover, it can be easily implemented on a microcomputer linked to
ordinary measurement devices.
On-line model based measuring techniques have the highest predictive power and selectivity.
Even though the development of a mathematical model is time consuming, once it has been builtup, it can be applied not only for safety but also for optimization purposes increasing the yield and
minimizing the time for each batch, and consequently can be justified economically.
224
NOTATION
A
A
B
C
Cp
Ea
H
k
K
IP
q
M
M
R
r
S
T
U
US
m
V
V
<w
%
y
depends on kinetics
pre-exponential factor
Dynamic matrix of the state -space model
Input matrix
Molar concentration
Specific heat capacity
Activation energy
Molar enthaply (liquid)
Reaction velocity constant
Gain vector of the Kalman filter
Covariance matrix
Thermal flow
Mass
Measurement matrix
Gas constant
Rate of reaction
Surface
Temperature
Heat transfer coefficient
Effective heat transfer coefficient
input vactor
Noise vector of the measurements
Volume
Noise vector of the state
State vector
Measured output vector
mol-m"3
JKg-^K" 1
J-mol"1
J-mol"1
depends on kinetics
W
Kg
J-moH-K"1
mol-m'^s" 1
m2
K
W-m-2-K-1
WK" 1
Greek symbols
Discrete state transition matrix
J-K-i
Thermal capacity
p
X
CO
Subs cripts
Actuator or External cold source
A
Stirrer or inserts
a
Bottom
B
L
m
M
Kg-m-3
s
rad.s"1
: Liquid
: Reaction mixture
: Measurement
225
c
d
E
e
h
i
0
1
P
R
r
sp
T
t
V
w
z
REFERENCES
1. Grewer, T., Klusacek, H., Loffler , U., Rogers, R.L., and Steinbach, J. (1989)
'Determination and assessment of the characteristic values for the evaluation of the thermal safety
of chemical processes', J. Loss Prev. Process Ind. 2, 215.
2.
Riesen, R. and Grob, B. (1985) 'Reaction Calorimetry in Chemical Process Development',
Swiss Chem. 7, 39-43 .
3.
Hub, L. and Jones, J.D. (1986) 'Early On-Line Detection of Exothermic Reactions',
Plant/Operation Progress 5, 221.
4.
King, R. and Gilles, E.D. (1986) 'Early detection of hazardous states in chemical reactors
with model-based measuring techniques', 5th International Symposium "Loss prevention and
Safety in the Process Industries", Cannes 15/19 September.
5.
Isermann, R. (1984) 'Process Fault Detection Based on Modeling and Estimation MethodsA Survey', Automatica 20, 387 .
6. Tufano, V. (1988) 'Modeling runaway reactions in reactors protected with suppression
systems', J. of Hazardous Materials 19, 225 .
7.
Himmelblau, D.M. (1978) Fault Detection and Diagnosis in Chemical and Petrochemical
Processes, Elsevier, Amsterdam.
8. Towill , D.R. and Payne, P.A. (1971) 'Frequency domain approach to automatic testing of
control systems', Radio and Electron. Engr.41, 51.
9. Casadei, R. (1977) 'Automatisierungstechnik im Wandel durch Mikroprozessoren' in M.
Syrbe and B. Will (eds.), INTERKAMA-Kongreb, Springer, Berlin, 179.
10. Luenberger, D.G. (1966) 'Observers for Multivariate Systems', IEEE Trans, on Automatic
Control 11, 190 .
11. Kalman, R.E. (1960) 'A New Approach to Linear Filtering and Prediction Problems', J.
Basic Eng. 82 D, 35-45.
12. Gilles, E.D. and Schuler, H. (1982), 'Early Detection of Hazardous States in Chemical
Reactors', Ger. Chem. Eng. 5, 69.
13. Bonvin, D. and Saner U. (1988) 'On Line Procedures for Supervising the Operation of Batch
Reactors', Comput. chem. Engng. 12, 371-376.
14. Snee, T. (1991), in this course.
15. OLIWA manual (1985), System Technik AG.
16. Oppenheim, A.V. and Shafer, R. W. (1975) Digital Signal Processing, Prentice-Hall,
Englewood Cliffs, NJ.
17. Seborg, D. E., Edgar, T.F. and Mellichamp D.A. (1989) Process Dynamics and Control,
Wiley , New York.
226
18. Spence, J.P. and Noronha, J.A. (1988) 'Reliable Detection of Runaway Reaction Precursors
in Liquid Phase Reactions', Plant/Operation Progress 7, 231 .
19. Seborg D.E., Edgar T.F., Shah , S.L.(1986) , AIChE Journal 32, 881.
20. P. de Valliere and D. Bonvin, Application of estimation techniques to batch reactors-II.
Experimental studies in state and parameter optimization, Comp. Chem. Eng., 13,11 (1989).
21. Soliman, M. A. and Ray, W. H. (1979) 'Nonlinear State Estimation of Packed-Bed Tubular
Reactors', AIChE Journal 25, 718-720.
22. Kuruoglu, N.R., Ramirez, W.F., Clough, D.E. (1981) 'Distributed Parameter Estimation
and Identification for Systems with Fast and Slow Dynamics', Chem. Eng. Sci. 36,1357 (1981).
23. Watanabe.K. and Himmelblau D.M. (1984) 'Incipient fault diagnosis of nonlinear processes
with multiple causes of faults', Chem. Eng. Sci. 39, 491-508.
24. King, R. (1985) 'Multiple Kalman filters for early detection of hazardous states', Proceedings
of Industrial Process Modelling and Control, Hangzhou, 6-9 June.
25. Lainiotis, D.G. (1971) 'Optimal Adaptive Estimation: Structure and Parameter Adaptation',
IEEE Trans. Automatic Control 2, 160-169.
26. Jazwinski, A.H. (1970) Stochastic Processes and Filtering Theory, Academic Press, New
York, 269.
27. Gelb, A. (1974) Applied Optimal Estimation, MIT Press, Cambridge.
28. Anderson , B.D.O. and Moore, J.B. (1979) Optimal Filtering, Prentice-Hall, Englewood
Cliffs, N.J.
29. Zaldivar, J.M., Hernandez , H. and Barcons C. (1990) Development of a Mathematical
Model and Numerical Simulator for a Reaction Calorimeter. FISIM, RC1 version, Technical Note
N 1.90.109, Commission of The European Communities, Joint Reasearch Centre, Ispra (Italy).
J.S. Duffield
CEC Joint Research
Centre.
Ispra Establishment
I - 1020 Ispra (Va)
Italy
1.
INTRODUCTION
228
(Design Institute for Emergency Relief Systems) project. DIERS consists of a consortium
of 29 companies under the auspices of the American Institute of Chemical Engineers and
was formed to generate experimental data on large-scale vessels undergoing either runaway
reactions or external heating (e.g. by external fires or loss of cooling), and to develop
methods for the safe design of emergency relief systems to handle these events. Significant
progress has been made but this knowledge is by no means complete.
This lecture will concentrate on what happens within the vessel during emergency relief,
and an attempt will be made to highlight the basic phenomena that are present (and that
have to be modelled) during the transient and how they interact and affect the course of the
transient. Reference will be made to the computer code RELIEF, which is currently being
developed at the JRC to model such situations and most of the illustrative calculations
presented in this lecture have been made using this code.
The important aspect of two-phase critical flow through the vent line will be left to the
next lecture. It is perhaps worth mentioning however, that it is the behaviour within the
vessel that determines the entrance conditions to the vent line, and the errors associated
with calculating these conditions usually have a much greater effect on the calculated vent
flow than the likely spread of values that would result from using different critical flow
models.
2.
The situation under consideration can be a batch reactor or storage tank containing a multicomponent liquid mixture in which a chemical (usually exothermic) reaction occurs. If due
to malfunctioning the generation of reaction heat in this mixture exceeds the heat removal
capacity of the equipment a thermal runaway process will occur which is strongly enhanced
by the Arrhenius-type temperature dependency of reaction rate. When this situation cannot be controlled by operational measures, the temperature will rise to levels where the
volatile components of the liquid reactant mixture start to evaporate. At high temperature
levels also gas may be produced as a result of undesired secondary decomposition reactions.
This volume production leads to an increase of system pressure and in order to prevent
over-pressurisation of the reactor vessel it is necessary to discharge the fluid mixture from
the vessel at an adequate rate. For sizing the emergency relief system, required by safety
rules, it is necessary to dispose of adequate computation models which must be based on a
correct description of the chemical conversion, of mass transfer between liquid and vapour
phase, of two-phase fluid dynamics and of the interactions between these processes.
2.1
LEVEL SWELL
Chemical reactions can be the cause of a rise in pressure in a closed system by increasing
the vapour pressure of the system and/or by generating non-condensible gases as reaction
or unwanted decomposition products. Even endothermic reactions can cause a pressure
increase if the reaction products are gases, or liquids which are more volatile than the
reactants. Exothermic reactions are potentially more dangerous as in addition they raise
the temperature of the reactants and hence accelerate the chemical reaction.
229
Often in the literature distinction is made between the mechanism of this pressure rise
so that simplifications can be made to the mathematical treatment of relief system sizing.
The following distinctions are usually made:(a) "Vapour pressure" or "tempered" systems, in which the pressure generated by the
reaction is due to the increasing vapour pressure of the reactants, products and/or
inert solvent as the temperature rises.
(b) "Gassy" systems, in which the pressure is due to the production of a permanent gas
by the reaction.
(c) "Hybrid" systems, in which the pressure rise is due to both an increase in vapour
pressure and permanent gas generation.
For such systems a number of analytical tools and formulae can be used to calculate the
vent size for a particular overpressure. These "hand calculational" methods usually treat
the vessel as a single calculational node having uniform properties. The obvious difficulty
arises when this assumption is not valid and when it is not known a priori
what type
of system is expected. An excellent review of existing vent sizing methods of this type is
given by Duxbury and Wilday [1,2].
When a runaway reaction is in progress there is a volumetric source in the liquid phase
resulting from evaporation and/or reaction gas production. The bubbles of vapour and/or
gas generated within the liquid tend to rise through the liquid and disengage at the liquid
surface. If the rise velocity is sufficiently high then droplet entrainment can occur.The
bubbles during their residence in the liquid occupy volume and so cause the liquid level to
rise or "swell". Figure 1. shows qualitatively this phenomenon.
When the set pressure is reached and the reactor or storage vessel relieves then the
pressure falls and the evaporation or "flashing" increases markedly. This causes the liquid
level, or to be more precise the "two-phase mixture level"*, to rise further and if this level
reaches the vent position two-phase venting will occur. The depressurisation rate of a
system is directly proportional to the volume flow rate exiting the system and since this,
under critical flow conditions, is inversely proportional to the mixture density entering the
vent line the capacity to reduce the system pressure by venting is strongly reduced when
the mixture level reaches the vent position. In a runaway situation if the volume production
rate due to evaporation and/or gas production is greater than the vented volume flow rate
the system pressure will increase. Therefore, the ability to describe the motion of the
two-phase mixture level is one of the most important aspects of reactor relief modelling.
2.1.1
Interfacial
Momentum
Transfer
The general area of interfacial momentum transfer in pool systems containing arbitrary
fluids and subject to depressurisation is still an open area of research. There is a paucity
of experimental data particularly for high viscosity fluids from which realistic models can
be developed. The problem reduces to the description of the motion of bubbles within a
The two-phase mixture level separates the region which is predominately liquid (possibly containing vapour bubbles)
from the region which is predominately vapour (possibly containing liquid droplets) and is usually deSned as the position
where there is a discontinuity in the axial void fraction profi/e
230
continuous liquid phase and the motion of droplets within a continuous vapour phase, and
how this motion changes with void fraction. In a bubbly liquid in a pool where wall effects
are negligible, the friction or drag exerted by the liquid on the bubble surface determines
how fast the bubbles rise within the liquid. This frictional force in comparison to the
buoyancy force decreases markedly as the bubbles increase in size. As the void fraction
increases the rate of slip of the vapour past the liquid increases until the situation arises
when the liquid begins to break up into droplets and the vapour becomes the continuous
phase. The drag between the droplets and the vapour increases as the droplet size decreases
and so the slip decreases. At the extremes of all liquid and all vapour flow the slip must
obviously be zero, and at some intermediate void fraction the slip will be a maximum where
the bubbles have their maximum size. Therefore, the phasic velocity difference should vary
with void fraction in a way similar to as is shown in figure 2.
To describe this motion one has to look at the phasic momentum transfer. The equations that describe this process are usually formulated in terms of mixture or drift flux
model [3] or a two-fluid model [4].
In the drift flux model the phasic momentum equations are replaced by a mixture
momentum equation and the relative velocity between the phases is described by a steady
state correlation. Its validity is limited to situations where acceleration and wall friction
forces can be neglected. For most practical cases it has been found that an equation of the
form:
Ur
Uv-U,
^00am(l-a)"
(1)
correlates experimental data for bubbly flow in vertical pipes well, where Uoo represents the
terminal rise velocity of a single bubble. The dependence of !/, upon fluid properties has
been determined experimentally by Pebbles and Garber[5], and for bubbles of diameter up
to 1 or 2 cms is given by,
Uoo = 1-53
gcrAg
1/4
(2)
2c
1/2
9Ql
(3)
then
Uoo = y/gn
(4)
The two-fluid model uses a separate partial differential equation to describe the motion
of each phase and if one assumes that the pressure difference between the vapour and liquid
phases is negligible, the momentum equations averaged over a constant cross section can
be written in the form:
231
dagvUv
dagvUv2
V
dt
az
dp
da
\-a + Am +pint = -n - F
az
az
wv
aeff + r m t / , n t
. ,
15)
'
+
'
+ (1 a)-f- - Am - pint = Ti-F w,-(lat
az
az
az
.
..
...
a)etg - TmUint (6)
Am
is the added mass term related to inertial effects. Its value is important only
for high velocity accelerating flows.
r;
^iu(v,i)
Pint ff
The interfacial drag force r< is the force that develops when one of the phases attempts
to move faster than the other and acts so as to retard the faster moving phase. The
magnitude of this force depends on the shape of the vapourliquid interface, the relative
velocity and the phasic properties.
To solve this system one has to specify a correlation for the interfacial drag force. If a
force balance is made on a single bubble it can be shown that [6]:
3ag,U?Cd
n =
8
(7)
rf,
The basic empiricism of the twofluid model enters in determining the values of C j and TV
In the particular case of pool boiling or flashing in large diameter vessels (such as
batch reactors or storage tanks) the velocities are low and the temporal, acceleration and
convective terms in the momentum equations can be neglected as too can the terms related
to wall friction and momentum transfer due to phase change. Thus in this case there is
little to choose between the drift flux approach and the twofluid model except that the
drift flux model generally requires less computing time.
We have chosen for the illustrative computations presented in this paper to adopt the
drift flux approach, and to describe the phasic velocity difference as a function of void
fraction. The expression chosen is given below which incidentally is equivalent to figure 2.
Uv-Ut=Upool
a m (1 a)"
"maxK1
_J
"max)
(8)
232
where amax is the void fraction which gives the term a m ( l - a ) " its maximum value.
The denomenator is a scaling factor which ensures that the maximum value of the slip is
given by Upooi irrespective of the value of void fraction at which it occurs. The coefficients
m and n describe bubbly flow and droplet flow respectively, they have been fitted to experimental data. Upooi is closely related to equation (2) and contains a physical property
sz
and
a pool situation.
can
With the constant 1.53, Upool is identical to equation(2) and corresponds to the churnturbulent regime as defined by Zuber [7]. This definition has much success in describing
drift flux experiments where gas is bubbled through a liquid column in reasonably small
diameter tubes, say up to lOcms. However, there is a general awareness that this value
significantly underpredicts the rise velocity in flashing pool situations. For a more detailed
discussion reference should be made to the following articles by, Filimonov[8], Styirkovich
[9], Gardner [10], and Kataoka and Ishii [11]. Presently we are using a constant that
has been obtained from fitting equation (8) to data from available flashing experiments.
To improve the modelling further more experiments are required, particularly for viscous
fluids.
2.1.2
Effect
of Level
Swell
on Vessel
Depressurisation
Recently a number of vessel depressurisation experiments have been carried out in the
Multiphase Multicomponent (MPMC) test facility of the JRC [12]. Shown in figures 3 and
4 are two similar tests with water, where the only difference was the initial filling of the
vessel. In figure 3 the initial filling was 98% which ensured significant two-phase venting,
whereas in figure 4 the filling was 65%. Subsequent analysis has indicated that at a filling of
65% the two-phase mixture level did not reach the top of the vessel and all vapour venting
occurred. Thus a comparison of figures 3 and 4 indicates what effect two-phase venting
has on the depressurisation. It is clearly seen that the rate of depressurisation is strongly
reduced when two-phase conditions occur in the vent line. This becomes more important
when the initial conditions are not as in the tests i.e. an inert fluid at steady state, but a
chemically reacting fluid undergoing a thermal runaway.
2.2
MULTICOMPONENT EQUILIBRIUM
So far we have restricted our discussion to a single component fluid, whereas in the process
industry we are usually dealing with multicomponent mixtures made up of fluids of significantly different volatilities and physical properties. To address this problem we assume
that the multicomponent mixture consisting of liquid and saturated vapour is in phase
equilibrium. This may be expressed as:
Xvi = KtXH
(9)
where Xvi is the mole fraction of component t in the vapour phase, Xu that in the liquid
phase and K{ is the phase equilibrium ratio. If the vapour mixture behaves as an ideal gas
233
the partial pressure p, of component i in the vapour mixture is proportional to the mole
fraction X< resulting in:
Pi = pXvi
(10)
If the liquid mixture also behaves ideally and follows Raoult's law the partial pressure
Pi is proportional to the liquid mole fraction Xu and to the vapour pressure of the pure
component at the same temperature:
Pi = P.iXu
(11)
As appears from eqs(10) and (11) for ideal mixtures the phase equilibrium ratio Ki is:
Ki = *
P
(12)
For ideal gas behaviour of the vapour mixture the mole fraction X, is equal to the volume
fraction ",. Eq (9) then can be transformed into:
Y,i = KiXu
(13)
The relations between the component mole fractions and the component concentrations
are:
x =
" i^h^
{14)
> 2^ i M(
""i L-i
u{
(16)
234
"dewpoint" Tdcw In this case, the condensed liquid will have a composition XAH of around
0.05.
2.3
CHEMICAL REACTION
An ongoing chemical reaction is a potential source of both volume (due to the generation of new products) and heat. Consider a general chemical process in the liquid phase
which involves two reactants A and B undergoing an irreversible reaction according to the
stoichiometric formula
nAA + nBBi*ncC + nDD
(17)
The instantaneous conversion rate of reactant A can be expressed by an Arrhenius type
temperature dependency:
n,A = k0\C,^\m\C,^]nexp[-E/RT)
(18)
where : k0 is an empirical factor, m and n are the reaction orders for A and B, E is the
activation energy (in J/kmol), R is the gas constant and X is the absolute temperature.
The heat source if the reaction is exothermic, or the heat sink if it is endothermic is
obtained from the enthalpy change of the reaction or from the differences in the heats of
formation of the reactants and products.
2.4
NON-EQUILIBRIUM EFFECTS
3.
MODELLING PRINCIPLES
Having described the key phenomena associated with reactor relief we wish to provide some
illustative computations to highlight the importance of these phenomena. For this we will
235
make use of the code RELIEF, which is being developed here at the JRC; the salient features
of which will be described below. Hand calculational methods, derived from the DIERS
programme, will not be discussed here and the reader is referred to paper of Duxbury and
Wilday [1,2] for further details.
In modelling the in-vessel fluid behaviour an attempt is made to separate as far as possible
the physical, physico-chemical and thermokinetic phenomena associated with the multicomponent phase-equilibrium and with the exothermic liquid phase chemical reaction from
the phenomena pertaining to the two-phase fluid dynamics. The rate of vessel depressurisation and the rate at which fluid is discharged from the vessel depend on the fluid dynamic
phenomena in the vent line and on the two-phase flow behaviour of the multicomponent
mixture inside the vessel. Both phenomena are modelled but major attention needs to be
given to the multicomponent two-phase fluid behaviour in the vessel because the vent mass
flow rates depend strongly on fluid dynamic conditions at the entrance to the vent line. In
the illustrative calculations shown later a simplified critical flow model has been selected
to compute the mass flow through the vent line, taking the stagnation pressure and fluid
mixture density as the prime governing parameters.
The treatment of volume production in the vessel, which is due to evaporation of volatile
components from the liquid and to gas production by chemical reaction, is an element of
key importance in the analysis. Vapour production during the venting process is associated
with external heat input, a liquid phase volumetric heat source due to ongoing exothermic
chemical reaction and the change of system pressure with time. Gas production can result
from the ongoing chemical reaction or through secondary decompostion reactions.
The key features of the in-vessel fluid flow model are:
* Vertical discretisation of the vessel into control volume elements for which conservation
laws pertaining to the separate phases are applied.
* Formulation of mass conservation equations for individual components in each of the
phases.
* Formulation of phasic energy equations for the component mixtures.
* Description of the relative motion between phases with an algebraic phase slip (driftflux) model.
* Irreversible chemical reaction in the liquid phase formulated as one of arbitrary order in
terms of reactant concentrations with a temperature dependency given by an Arrheniustype expression.
The following hypotheses and assumptions are presently made:
- miscibility of all components in the liquid phase
- uniformity of pressure over the flow cross section
- the wall friction and acceleration terms in the momentum description can be neglected
- the kinetic energy terms in the energy equations can be neglected
236
- axial conduction and axial mass diffusion in the fluid can be neglected
- thermal equilibrium between liquid and vapour phases (no superheat at onset of the
depressurisation process)
- distribution of components between phases governed by phase equilibrium relations
- phase equilibrium relations based on ideal gas behaviour (Dalton's law) and on ideal
liquid solution behaviour (Raoult's law)
It is worth pointing out that most of the above assumptions do not represent limitations
of the analysis. Their elimination however requires specification of additional relations
containing parameters that have to be determined experimentally (e.g. in case of nonequilibrium, interphase areas and interphase transfer coefficients, for non-ideal liquid mixtures activity coefficients e t c . ) .
3.1
CONSERVATION EQUATIONS
Conservation laws are applied to the separate phases in one-dimensional Eulerian finite
volume elements. For the sole purpose of simplifying the presentation of equations the
assumption is made that flow cross sections at the inlet and outlet boundaries of a finite
volume element are equal.
3.1.1.
Mass
Conservation
Here mass conservation for component i is expressed in terms of concentrations per unit of
mixture mass. This formulation allows one to separate the concepts of change in chemical
composition and change of mixture density and enables the derivation of phasic mixture
energy conservation equations containing heat source terms, associated with chemical reactions.
The vapour mass conservation equation in discretized form for a component t can be
written as:
A(agvfivi)
A(Gvfij)
+ rmi+ari
(19)
A((l-a)eifMi)
^
A(G, W; )
y
T mi + (1 - a)ru
(20)
237
In the above equations are: Q and QI the mixture density in the vapour and liquid
phase, /*,- and fin the component mass concentrations in mass per unit of mixture mass
(which is equivalent to the component mass fraction) in vapour and liquid phase, G and G/
the vapour and liquid mass flow rates, T m ; the component interphase mass transport rate
per unit of element volume, r,- and ru the component mass production rate in the vapour
and the liquid phase per unit of phase volume, given by equation(l8). Physical parameters
on the right hand side of equations(19) and (20) represent time average values during the
time step (time centered values). Parameters under the difference operator at the right
hand side pertain to the inlet and exit boundaries of the element.
3.1.2.
Energy
Conservation
Neglecting the contributions of kinetic energy and of axial conduction to energy transport
the energy conservation equations formulated in terms of enthalpies, in discretized form,
are for the vapour mixture:
H^BvK)
^
A(G/i)
Ap
y
+ n.,rm + qhv +a~^
I 21 )
A(GiM
y
(22)
. r
. . . . .
>API
nv,Tm + qhi + (l-a)
Where qhv and qh[ are the external heat input per unit of element volume to vapour
and liquid respectively, the mixture enthalpies are given by :
hv =J2^ih"
i
i = ^2^ihii
( 23 )
(24)
where hvi is the component enthalpy in the vapour phase at system pressure and temperature and hn the component liquid phase enthalpy at the prevailing liquid temperature
and component saturation pressure, and h, is the vapour mixture saturation enthalpy at
the vapour-liquid interface.
The left hand term and the first right hand term in equations (21) and (22), representative of energy accumulation and energy convection, contain a change in mixture enthalpy
(over time and distance respectively).
The above equations can be developed further by taking into consideration that the
component composition in the separate phases may change due to occurrence of chemical
238
reactions and mass transfer processes. The aim is to determine the total volume production
rate per element. This can be written as:
'/'tot = $p + V'chem + V W
(25)
Equation(25) expresses that the total volume production rate per element foot is due to
changes of pressure and temperature in the gas phase (\>p), a volume source term associated
with chemical conversion (V^m) and volume production by phase change (il>vap)The rate at which components are produced by chemical reaction [rvf,rti) is directly
related to the reactant conversion rate taking account of the stoichiometry of the reaction
equation. Note that if the term rpciiem dominates equation(25) then even an endothermic
reaction can cause a rise in pressure.
The total volume production rate in the vessel can be found by summing the contributions for all elements in the vessel. Equating the total volume production rate in the vessel
to the total volumetric flow rate exiting the vessel yields the value of the depressurisation
rate within the vessel.
3.2
With the known value of the elemental volume production rate the mixture volumetric
flow rates at the element boundaries can be successively evaluated and from phase slip or
interfacial friction relations, as described in section 2.1.1, the phasic elemental volumetric
flow rates and the element void fraction can be calculated. The elemental component
distribution over the vapour and liquid phases can then be calculated using the component
mass conservation equations and the phase equilibrium equations.
4.
ILLUSTRATIVE CALCULATIONS
Calculations will be presented that sequentially include the phenomena that have been
described above. The aim is to highlight the importance and relevance of certain parameters
during venting and to demonstrate the complexity of the process.
4.1
The hydrodynamic capability of RELIEF has been extensively tested in a recent benchmark
exercise carried out at the JRC [12]. Here various blowdown experiments using pure component fluids in different sized test facilities operating at different pressures were compared
to various code calculations. Some of these calculations were performed "blind". Figure 6
shows a comparison of such a blind prediction using RELIEF with the experimental values
for the depressurisation of a vessel (top venting) containing the refrigerant R114.
239
In the following eight figures (7 - 15) the importance of phase equilibrium, interfacial
friction and vent location will be considered. In all these cases chemical reaction is absent so the vapour production and subsequent level swell are due only to phase change
and vapour expansion. The situation considered is the venting of a typical type of batch
reactor or storage tank. The vessel is cylindrical and positioned vertically with the vent
positioned either at the top of the reactor or at the bottom. The vessel was given the
following dimensions: height=4.17 m, internal diameter=0.305 m, vent diameter=0.02 m.
It was descretised into 100 axial elements. In principle any number of components can
be treated with the model but for these calculations we will restrict our attention to two
components and will chose fluids that have dissimilar vapour pressures. The fluids chosen
are the refrigerants R - l l and R-22; figure 7 shows the vapour pressure curves for these two
substances. In all the calculations the vessel is initially filled to a relative height of 60%
with an initial mixture composition of 20% (mass) of the volatile component and 80% of
the less volatile component.
To test the importance of liquid-vapour equilibrium the results of two calculations will
be shown. The first treats the binary mixture as a true mixture in which the composition
changes during the transient due to phase equilibrium effects, the second keeps the composition constant (pseudo one-component) with the properties kept equal to those of the
binary mixture at the initial conditions. Figures 8 and 9 show the pressure and temperature histories for the two cases. Notice that the pressure of the pseudo one-component fluid
remains always above the two component fluid but that the temperature always remains
below.
4.1.1.
Phase Slip
Of crucial importance is the ability to describe the extent of the liquid "swelling'' which in
turn defines the onset and duration of two-phase condition at the entrance to the vent line.
In figures (10 - 15), the important effect of phase slip and vent location is demonstrated.
In these figures two situations are considered, ie. both top and bottom venting of a vessel
containing a true mixture of the two refrigerants R - l l and R-22. To illustrate the effect
of interfacial friction two values of slip have been taken; the first calculated using equation
(8) and is labelled "with slip" in the figures, the second is representative of a homogeneous
mixture or foam, obtained by putting slip to zero and is labelled "no slip".
Figure 10 shows the vessel void fraction evolution for the case of top venting with slip.
The initial swelling of the liquid is clearly seen and after about 5 seconds the mixture level
reaches the top of the vessel; this signifies the beginning of two-phase venting. Note that
a secondary mixture level is formed near the middle of the vessel after about 10 seconds
which steadily moves upwards and after about 50 seconds reaches the top of the vessel.
At this time there is a significant axial void profile in the vessel. Later in the transient at
times beyond 100 seconds (not shown in the figure so as to avoid confusion) the mixture
level moves downwards from the top and stops when the blowdown is complete, giving the
final situation with the remaining liquid in the lower part of the vessel. Figure 11 shows
the same void fraction evolution when slip is put to zero (homogeneous conditions) and a
characteristic foamy behaviour is seen. After 20 seconds the vessel is completely filled with
240
a uniform mixture, and the void fraction becomes a function of time only and not vessel
height. No settling out of the fluid will occur at the end of the transient.
Figures 12 and 13 illustrate what happens in the identical case with bottom venting.
For the case where there is phase slip one again sees a moving mixture level but in this case
it steadily moves downwards at an almost constant speed. At about 40 seconds the vessel
is voided and the vessel pressure will then fall rapidly (see figure 14). For the homogeneous
case with no slip the transient is characterised by the mixture level remaining at the height
of the initial filling and with a constant void distribution below this level.
The consequence of there being an interphase slip on the global parameters is vividly
shown in figures 14 and 15. The system pressure histories are shown in figure 14 and
demonstrates the strong influence of both vent position and phase slip. One should note
that the curves for top and bottom venting show even a qualitatively different behaviour,
and that for a certain period of time top venting can result in the highest pressure. Figure
15 shows the corresponding normalised mass inventory histories where significantly different
masses are vented. This can have major consequences when one has to consider the design
of downstream equipment.
4.2
Two types of chemical reactions will be considered, the first is typical of a polymerisation
reaction where the more volatile component is converted into the less volatile component,
the second is a decomposition reaction forming a gaseous product. Calculations were performed where slip was modelled, again using by equation(8) and where slip was put to zero.
The two components had vapour pressures that differed by a factor of ten and the initial
mass fraction for each component was set to 80% for the more volatile component and 20%
for the less volatile component. The vent line was positioned at the top of the vessel and the
relief valve was set to open at an overpressure of about 10%. Figure 16 shows the influence
of phase slip on the pressure history, and demonstrates that for this type of reaction a slip
between the phases reduces the tendency for runaway.
The influence of vessel dimension is shown in figure 17 where calculations have been
performed for two vessel having the same volume but with different lengths ( 0.4m and
4.0m ). A comparison of figures 16 and 17 leads one to believe that the effect of reducing
the vessel height has a similar effect to that of increasing the phase slip. The physical
explanation for this scale effect is that within the vessel bubbles rise with the same velocity
independent of vessel height so their residence time in the liquid is larger for longer vessels.
This results in an increase in the "swelling" of the liquid for the longer vessel and hence
increases the probability of obtaining two-phase flow conditions at the entrance to the vent
line. This in turn reduces significantly the pressure relieving capability of a given vent
geometry.
It is interesting to see whether these conclusions are universally valid or are specific
to reaction type. Figures 18 and 19 show similar calculations for the decomposition of
hydrogen peroxide. This reaction has been studied in detail because we have recently
performed such experiments here at the JRC. Here one sees that the tendencies are exactly
the reverse of those seen with the polymerisation reaction.
241
To explain this one has to appreciate that the liquid mixture is initially subcooled
and the gaseous phase is primarily composed of oxygen from the decomposition reaction,
which has no latent heat of vaporisation associated with it. Therefore, the pressure and
temperature within the vessel are effectively decoupled. Once the relief valve opens (sa 100
seconds into the transient for a relief set pressure of 2 bar) the vessel pressure remains close
to the outer pressure until the onset of runaway, and oxygen is vented from the vessel. For
homogeneous conditions (zero phase slip) or for slip when we consider the longer vessel
more liquid swelling occurs. If the mixture level reaches the elevation of the vent line then
two-phase fluid will be vented and since the temperature and pressure are decoupled there
is no feedback on fluid temperature (and hence reaction rate) of any increase in vessel
pressure due to this early two-phase venting. Of importance however, is that reactant mass
is vented in this case. Thus when runaway occurs there is less hydrogen peroxide in the
vessel if homogeneous conditions or the longer vessel are considered. This then results in
lower peak pressures than the corresponding cases where inter phase slip is modelled or a
shorter vessel is considered.
4.3
VENT SIZING
As an example of how RELIEF can be used to size a venting system, or to calculate the
consequences of using an existing system for a given reaction, we will consider the hydrogen
peroxide decompostion now in more detail. The results presented here refer to a vessel of 1
m 3 with an initial filling of 60%. The liquid composition is 20% H202 and 80% H20. The
relief valve was set to a very small overpressure so that the vent can be assumed to be open
at a time close to zero. Figures 20 and 21 show the results from calculations using three
different vent diameters. The first assumes a zero vent diameter and therefore represents
the closed adiabatic case. The other two calculations have vent diameters of 2cm and 5 cm
respectively.
Figure 20 shows the temperature histories and what is immediately apparent is that the
vent size has a negligible effect on the initial heat up phase (say up to 6 min). Thereafter,
up until the maximum temperature is achieved, the differences are mainly between the
closed vessel and the vented vessel. This means that for a vented vessel the time for the
onset of runaway is to a first approximation independent of the vent area. One should also
note that the value of peak temperature is little affected by the venting process, ranging
from 184 C for the adiabatic case to 175 C for a vent diameter of 5cm.
Figure 21 shows the corresponding pressure histories and here one sees that vent size
has a significant effect on the pressure. The temperatures and pressures are effectively
decoupled. The reason for this is that during the heat up and runaway the system pressure
is always above the saturation value of the mixture, which means that the gaseous phase is
virtually entirely made up from the oxygen liberated from the decomposition reaction. In
fact boiling of the liquid only occurs after all the H202 has been consumed and the pressure
falls below the saturation value. This occurs at 7min 40sec for the 5cm vent and at 8min
35sec for the 2cm vent where a distinct change in the gradient of the temperature trace can
be seen.
242
Norma.!
operation
Runaway
i<
Mixture
" o o
level
Vapou.r
vapour
Reactor
generation
reach
Fig 1.
situation
Mixture
all
location,
level reaches
two-phase
venting
level swell
0.50
_
>
0. 40
>o
LJ
/
0.30
u
ex
u.
u.
a
/ V
0.C0
o
o
_l
Lu
>
0.10
I/]
0.00
0. 20
VOID
Fig 2.
0.50
0. 70
FRACTION
vent
ver.tin"
243
2. 40
fV
^
2.20
2.00
N.
1.80
1.60
\ .
1.40
_
:
1.20
1.00
TIME
Isac)
Fig 3.
2.40
2.20
1\
\
2.00
1.80
1.60
^ ^ ^ - . _^^
1.40
1. 20
N.
~~"~~-
1.00
40.0
TIME
Fig 4.
Isocl
244
'dew
h
T
bub
r-t*
l \
-X
'
= 0.7
>v
rt^
-fi _ ri
T)
!
I
!
I
*.,,;= 0.25
.
'
1
I
100% B
Fig 5.
100% A
7.00
6.00
K
-
^
\ ,N
"
5.00
-z.
EXPERIMENT
\
\ \ /
\ X'
V
PREDICTION
n-
i
(ft
Id
u.
0-
^--V
2.00
1.00
0.00
TIME
Fig 6.
(S)
245
S"
S"
s-
*"
^ ^
R-22/
y
^^
sS^
^
s^
j r
ys y^
>^
y^
R-iy
/
240.
280.
320.
Temperature (Kalvin I
Fig 7.
T n*e t seconds 1
Timm (seconds)
Figs 8. & 9.
pseudo one-component
246
Voi d F r e c t ion
Fig 10.
:
1
;
0 sec.
60.0
50.0
'
40.0
I
*j
i
*
5s
30.0
20.0
10.0
i
]
0.<0
'
1
i
i
i
i
i
---'
j 10 s j 20 s
i
j
1
i 50 s
i
i
!
|
i
i
i
i
i
1
i ,
i .
0. CO
Void F r a c t I o n
Fig 11.
247
i-
eo.o
80. 0
i i |
i .
70.0
0 sec.
SO. 0
40.0
30.0
20.0
10.0
0.0
10 s
/
/
/
/
/
/ /
//
20 s
30 s
- i//
;
40 s
jS-;-
;.
r --7-;-0. 0
0. 60
Void
BO. 0
70.0
FroctIon
......... .....
BO. 0
:
:
'0 sec.
r
50. 0
':
40.0
'-
/
!
10 s\
20 s
30 s
30.0
40 sj
20.0
10.0
0.0
1.
. . . .
0.40
0.60
Void FroctIon
Fig 13.
248
60.
120.
Tl
Fig 14.
(seconds)
Pressure histories for different inter phase slip and venting modes
^ \
' \ ^ ^^"^
0. BO - \ \
^^^\
0 so
0.70
0.60
0. 50
0. 40
0.30
0.20
^^^^^^
' \\\\\ \\
^ ^ _ _ J
:
0.10
0 00
~~\
\
v_
slip
\ n o slip """""--^
\60..
T i r>
Fig 15.
120.
(seconds]
Mass inventory histories for different inter phase slip and venting modes
249
ao.
120.
lco.
20a.
TIHE ( seconds )
Fig 16.
Polymerisation
reaction
long vessel
a.
10- o
BO.
120.
rinE I atconda )
Fig 17.
250
90.0
|
t
H
BO. 0
70.0
Decomposition
reaction
i!|
60.0
with slip
1
i
50.0
40.0
1
1
1
1
30.0
20.0
//
10.0 h
\
\
\
\
/ /J /
\ \ n o slip
0.0
TlflE ( s t c o n d a
Figl8.
9;
20.0
400.
TIME I seconds
Fig 19.
500.
)
251
1
0 cm
.X
ii \
1 \
1
S cm
4:00
Fig 20.
Fig 21.
0:00
252
REFERENCES
1) H.A. Duxbury, A.J. Wilday,
1969.
1975.
5) F.N. Peebles, H.J. Garber, Chem Eng Progr, Vol49, 88-97, 1953.
6) M. Ishii, N. Zuber, "Drag coefficient and relative velocity in bubbly, droplet or particulate flows", J.AIChE, Vol25, No5, 843-855, 1979.
7) N. Zuber, J. Hench, Rept no. 62GL100, General Electric Company, Schenectady, N.Y.,
1962.
8) A.E. Filimonov, M.M. Przhiylkovski, E.P. Dik, I.N. Petrova, Teploenergetika 4(10),
22-26, 1957.
9) M.A. Styrikovich, A.V. Surnov, Y.G. Vinokur, Teploenergetika 8(9), 56-60, 1961.
10) G.C. Gardner, "Fractional vapour content of a liquid pool through which vapour is
bubbled", Int. J. Multiphase Flow, Vol 6, 399-410, 1980.
11) I. Kataoka, M. Ishii, "Drift flux model for large diameter pipe and new correlation for
pool void fraction", Int. J. Heat and Mass Transfer, Vol (30) 1927-1939, 1987.
12) A.N. Skouloudis, "Fifteen benchmark exercises on vessel depressurisation with nonreacting fluids", EUR 12602 EN, 1989.
13) L. Friedel, 5th Int. Symp. on Loss Prevention, paper 43, Cannes, 1986.
14) G. Friz, W. Riebold, W. Schulze, "Studies on thermodynamic non-equilibrium in flashing water", OECD Specialists meeting on transient two-phase flow, Toronto, Aug 1976.
253
NOMENCLATURE
C
C
E
Y
G
U
h
P
R
r
ffc
T
V
X
9
F
Cd
DH
K
Concentration
specific heat at constant pressure
activation energy
volume fraction
axial mass flow rate
velocity
enthalpy
pressure
gas constant
component mass production rate per unit of phase volume
bubble radius
absolute temperature
volume of computational element
component mole fraction
acceleration due to gravity
wall friction force
drag coefficient
hydraulic diameter
phase equilibrium ratio
kgm" 3
Jkg^K-1
Jkinol" 1
m 3 /m 3
kgs- 1
ms" 1
Jkg- 1
Nm" 2
Jkmol^K-1
kgm-3s_1
m
K
m3
kmol/kmol
ma - 2
kgm-2s-2
m
Greek symbols
a
Tm
\i
g
<7
rf>
n
7
Subscripts
t
/
v
s
int
component
liquid phase
vapour phase
saturated
interface
m 3 /m 3
kgm _ 3 s _ 1
kgm - 3
Nm-1
m3a_1
kgm_2s-2
A. BENUZZI
Institute of Safety Technology, JRC Ispra,
Commission of the European Communities
21020 Ispra. (VA), Italy.
A B S T R A C T . This lecture is a review of the methods developed to calculate the critical flow during the
emergency pressure relief of a chemical reactor or a storage tank. The fundamentals of the single-phase
and two-phase criticalflowsare discussed but the emphasis is put mainly on the simplified formulations
recently developed (DIERS) in order to provide quick and simple-to-use vent sizing methods appropriate
for safety requirements.
1. I n t r o d u c t i o n
The scope of this lecture is to review the calculation methods currently employed
to evaluate the critical mass flux G ( = pu) in the venting line when for some reason
(runaway reaction or external fire) the pressure in the reactor (or storage tank) reaches
the "set" value or the value at which the pressure relief device will be fully open. It is
well known t h a t under these conditions two-phase flow will most likely occur / l / and
t h a t this situation represents the most severe requirement for the vent sizing. In fact,
the pressure rise in the reactor is related to the volume increase produced by vapour or
gas generation due to the effect of the energy sources. Pressure relief requires a balance
between the volumetric vapour/gas generation rate and the volumetric discharge flow
rate. Fig. 1 taken from / 2 / and referred to styrene properties at 0.5 MPa, shows t h a t
both the mass flux G and the two-phase density p decrease as the inlet void fraction a
increases. In contrast, the volumetric flow G/p is found to increase as a increases; for
pure gas flow ( a = l ) G/p is about 20 times larger t h a n for all-liquid (a = 0) flow.
As a consequence, two-phase flows need larger vent areas compared to pure gas flows,
in order to assure the same relief capacity. If a limited pressure increase ( e.g: 20% of
the set pressure ) is tolerated during the venting, a significant reduction in the vent
area will be possible. Therefore the evaluation of the mass flux G for two-phase flow
conditions represents a key step in the emergency relief systems (ERS) design. This
lecture will be limited to flows in the following boundaries (see Fig. 2):
Tubes (long pipes) with cylindrical or annular geometry, constant or variable
cross section and arbitrary inclination. Friction losses are always important.
Nozzles characterized by variable cross section (converging/diverging) and
short length which determines negligible friction losses.
255
A. Bemtzzi and J. M. Zaldivar (eds.). Safely of Chemical Batch Reactors and Storage Tanks, 255-284.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
256
Orifices (diaphragms) which are supposed to represent tubes with zerolength. Special cases are rupture disks, safety relief valves and wall breaks
which are represented by combinations of the previously mentioned elements.
1.1. UNSTEADY-STATE SINGLE PHASE CRITICAL FLOW
Although the theory of single-phase critical flows is well known, nevertheless it is
worthwhile to recall here its basic aspects in order to make easier the extension of the
formulation to two-phase critical flows. Considering a flow system formed by a pipe
attached to a constant pressure reservoir on one end and to a large receiver on the
other end, when the downstream pressure is reduced below t h a t of the upstream value,
a pressure gradient is set up in the pipe and the flow begins. A steady-state is reached
when the pressure gradient forces are balanced by the frictional and inertial forces. If
the thermodynamic state of the fluid in the upstream reservoir is kept constant, decreasing the downstream pressure will produce an increased flow rate until a maximum
value is reached. At this point, further reductions in the downstream pressure will produce no effect. The maximum flow rate value is called "critical" and represents a limit
in the discharge process t h a t has to be determined with accuracy in many engineering
problems.
The critical flow usually occurs when the Mach number is equal to one at the
smallest cross section of the pipe. In this condition a downstream perturbation (e.g:
change of pressure) cannot be transmitted upstream and the flow rate is no longer
influenced by the changes in the receiver presssure. Let us consider the set of partial
differential equations describing the transient compressible flow in a rectilinear pipe of
variable cross section A and inclination 0 with respect to the horizontal plane.
at ( M ) + h ( M ) = 0
ft (pAu) + j ^ (pAu2)
j - t [Ap (h + ) ] + fz
(Mass)
+ f^ (Ap) = -xTw
+ Apg cos 0
(Momentum)
(Energy)
(1)
where p, u, p and h are fluid density, velocity, pressure and specific enthalpy, respectively. In addition, TW and qw are shear forces and heat flux at the wall of the pipe, x is
a coefficient obtained through appropriate correlations and g is the gravity acceleration.
Equations 1) have to be coupled with the constitutive law:
P = f(p,h)
to close the system. Equations l) can be expressed in matrix form / 3 / :
(2)
257
(4)
(6)
A, > 0
for some
for all
j' < 3
i f j
(7)
(8)
These are the mathematical conditions satisfied by the equations of motion for a
flowing fluid when reduction in downstream pressure ceases to cause an increased flow
rate. It is well-known t h a t the choked-flow condition for single-phase flow occurs when
the fluid velocity just equals the local sound speed.
1.2. STEADY-STATE SINGLE PHASE CRITICAL FLOW
For steady-state flow the time derivatives in equations 1) become zero and the matrix
equation 3) reduces to:
BAf = r
(9)
258
and the flow is critical when the following conditions are fulfilled somewhere along the
pipe:
det[B)
= 0
det ( B P > ) = 0
(10)
t'=
1,2,3
(11)
(12)
where v = l / p is the specific volume of the fluid and the continuity equation can be
expressed as:
Gv = u
(13)
dG
dv
dp
dp
dp
TP
= Gmax
Tp
(15)
= 0
dv
v + Glaxv-=0
, .
(16)
259
G,2
l'dv\
= - U J
(17)
where Gmax is the critical mass flux, and p and v the pressure and specific volume,
respectively, at the exit of the discharging line.
2. P e c u l i a r i t i e s of T w o - p h a s e Critical F l o w s
Two-phase critical flow of a single-component mixture is much more complicated than
the corresponding single phase flow due to a number of phenomena / 6 / . These are:
1) change of phase, boiling (flashing) or condensation; 2) slip or relative velocity between the phases; 3) flow pattern (or flow regime) and 4) possible departure from
thermodynamic equilibrium. The first of these phenomena occurs because there is a
pressure variation along the line. Generally, the exit quality (per cent vapour by mass)
increases for low initial quality values and decreases for high initial quality values, if
the expansion is assumed to be isentropic / 7 / . Slip takes place mainly because the
vapour phase is less dense than the liquid phase and, therefore is more intensively accelerated by the acting forces. Friction between the phases and liquid entrainment by
the vapour tend to reduce the relative velocity between the phases. These phenomena
are strongly influenced by the flow patterns (flow regimes) or, in other words, by the
topological structure of the interface between the phases. In Fig. 4 the main flow patterns are depicted. It is evident t h a t mass, momentum and energy exchanges between
the phases will be strongly dependent on the flow regime occurring. However the flow
regime prediction and the evaluation of the flow regime transitions are problems which
still need more research work. Lack of thermodynamic equilibrium, has been shown to
exist in orifices and nozzles or, in general, in short pipes (L < 0.1m) / 5 / . In singlephase critical flows, even if the velocities are high, molecular relaxation phenomena
are sufficiently rapid for the vapour to be regarded as in thermodynamic equilibrium.
For two-phase critical flows, on the contrary, relaxation times for the formation of new
interfaces (nucleation), heat, mass and momentum transfer, and the evolution of flow
patterns are comparable with the time spent by the fluid in the critical region of rapid
property change. With a saturated or subcooled liquid, departure from thermodynamic equilibrium can occur because of the delay in initiation of boiling as the fluid
flows into a region in which the pressure is below its saturation temperature. This
delay can occur because of the lack of nucleation sites for evaporation or because of
the short time of expansion, or both. In the case of saturated vapour which is cooling
during an isentropic expansion, nucleation sites are needed. If the radii of the nuclei
are small, large subcooling can be reached. In this situation a "condensation shock"
or the sudden formation of a cloud of fine liquid droplets, can develop, giving rise to
a pressure pulse. Nevertheless for long pipes (L > 0.1m) the effects of the lack of
thermodynamic equilibrium are, in general, negligible / 7 / . Fig. 5 illustrates the typical pressure profile in a long pipe and the presence of equilibrium and nonequilibrium
regions for two-phase flow.
260
2.1. GOVERNING EQUATIONS FOR TWO-PHASE CRITICAL FLOWS
(19)
and r is the source term vector which is the right hand side of the six equations given in
Table I. The mathematical critical flow conditions for unsteady and steady-state flows
are formally the same as those of the corresponding single-phase cases ( eqns 6, 7, 8,
9, 10, 11), except for the fact t h a t the order of the polynomials is 6 instead of 3. The
model presented in Table I, called "mechanistic" or "Two-fluid", in principle appears
capable of representing all the recognized nonequilibrium phenomena occuring in twophase critical flows. It has been used, with minor modifications, in various computer
codes related to nuclear reactor safety analysis. The fluid considered is water which is
very well characterized for thermophysical and for exchange properties in the two-phase
region. This is not the case for other single-component fluids, and particularly, for the
multicomponent mixtures encountered in the chemical industry. For such systems the
"Two-fluid" approach, due to the lack of knowledge on the coefficients of the equations
shown in Table I, is difficult to use. Nevertheless, from the theoretical point of view
there is no doubt t h a t it is the more rigorous method.
2.2. SIMPLIFIED FORMULATIONS (HEM, HFM, ERM)
A current simplification made for the evaluation of the critical mass flux is the assumption t h a t the two-phase mixture is an homogeneous fluid with liquid and vapour
having equal velocities and temperatures (Homogeneous Equilibrium Model - HEM).
The specific volume of the mixture is given by:
v = vt(l - x) + vgx
(20)
Where x is the vapour quality. The standard expression for the critical flow 17)
is coupled with a momentum conservation (Bernoulli) / 7 / equation which for a nozzle
takes the form:
261
dp
" i L Jpo
(21)
where subscripts 0 and 1 indicate stagnation and external (downstream) conditions, respectively. For a long pipe the Bernoulli equation takes the form:
2po
vi
(11 _ M
_ JUL fH
V
Vaj
viPo Jo
L
J0 ' v*dL*+
*dH
P n
H
K0v*0 +2{v{-
v*0)
(22)
dx
dvi
(23)
For a single component fluid 23) can be evaluated at local conditions according
to:
dvg
dp
IP
'xX-wr-^
dx^
dp/
-CT
h2
dvi
(vi
dp
\af
(24)
(25)
(26)
(27)
In equations 23) through 27) 7 is the isentropic coefficient, vig and h\g are the
specific volume and enthalpy change from vapour to liquid state, respectively, C is the
liquid specific heat, T is the absolute temperature, and aj is the sound speed in the
262
liquid phase. In 25) and 26) the evaluation of dx/dp is performed along isentropic
and isenthalpic paths, respectively. The local quality with reference to stagnation
conditions can be approximated by:
hlgoT
CT
(T
nig J o
n-ig
\10
hlgo ,
<:)h
(To T)
(29)
(30)
Hg
"Ig
(31)
Tvlg
dvi/dp has been utilized in arriving at equations 26), 27) and 29). Using 24),
25) and assuming dvi/dp = 0, the H E M local choking condition in the simplest form
becomes:
xvg
IP
vfg(CT-xhlg)
t.
(32)
vfg(CT-xhlg)
+N
(33)
*?.
N =
*i
2ApPlK$vfgTC
+ 10L
(33a)
263
where KQ is a discharge coefficient ( = 0.6 for sharp edges orifices) and A p the total
available pressure drop between the vessel and the end of the pipe. The difficulty
associated with this model is t h a t N (usually N << 1 for nozzle flows) is sensitive to
the flow geometry. In the Equilibrium Rate Model (ERM) isenthalpic conditions are
assumed with N = 1 :
xC0v0 Vr g
v,CT
Vjg
IP
(34)
As a special but interesting case, we consider the saturated inlet conditions [XQ =
0) and we get:
hi
Gc
(35)
dp (T
dT\C
3. M u l t i c o m p o n e n t T w o - p h a s e Flows w i t h E n e r g y Sources
The physical properties of a multicomponent mixture can be derived from the properties of each component according to the following simple relationships:
hig = ^
Yihlgi
vig
E r ^.
E*'C<
where Y{ and X{ are the vapour and liquid mass fractions of the ith component, respectively. Therefore, the H E M concept can be extended to multicomponent mixtures
which are the fluids to be considered in the venting process of batch reactors.
4. D I E R S M e t h o d o l o g y
The Design Institute for Emergency Relief Systems (DIERS) / 9 / of the AIChE has
developed research programmes to:
reduce the frequency, severity and consequences of pressure producing accidents
promote the development of new techniques which will improve the design of ERS
DIERS was formed in 1976 by 29 US and European Organizations and so far has spent
in research work approximately $1.6 million. The main result of the DIERS work was
the development of an ERS sizing methodology based on:
264
WW/g.
A small test sample (100ml) to assure safe and easy handling
Direct extrapolation to large scale: safe but not overly conservative
Relatively inexpensive
The key feature of the equipment is the use of a low heat capacity test cell (see
Fig. 8) which is maintained in adiabatic conditions through electrical heating provided
to keep the wall temperature equal to the sample temperature. This should reproduce
in the calorimeter the behaviour of a "small volume element" of the real reactor charge
with undesidered scaling effects.
The V S P can be used to measure homogeneous (foamy) versus non-foamy behaviour during relief of a RR and to discriminate between "tempered" versus "gassy"
systems. The apparatus can also measure thermal quantitites and other RR d a t a
(temperature, temperature rate and pressure time histories and heat of reaction versus
temperature). The information obtained by testing the R R system in the V S P allows
the identification of the relief behaviour (thermal and fluid dynamic) and permits the
use of the computer code or other calculation methods, with the appropriate selection
of models.
SAFIRE (System Analysis For Integrated Relief Evaluation) is the computer code
developed by FAI / l l / under a DIERS contract which incorporates most of the DIERS
analytical methods. The code describes the multiphase fluid dynamics of a batch-type
chemical reactor or storage vessel with an ERS. Mixture physical properties, reaction
stoichiometry and kinetics, vessel and vent line geometry are input by the user. Fluid
dynamic model assumptions are also made by the user. SAFIRE can calculate the
mixture behaviour (temperature and pressure response), prior to relief system actuation as well as the subsequent vessel mixture discharge process through the ERS.
265
where r\ = p/po
and e = V/VQ)
rCTpvf
"T^,2
CTpvfg,2
g
r^- = a 0 +
'-JSi2
v0hfa
v0hf
The correlating
(38)
{ - 2 [a; In ry + ^ - l ) ( 1 - 7 ? ) ] } *
W
(i-1) + 1
>
266
+ 2 w 2 ( l r?d) = 0
(40)
After 77c is found, its value can be substituted into equation 39) to obtain G*.
Alternatively, the local choking criterion given by:
(41)
c j=
can be used to obtain the same result. These solutions for choked flow through a
nozzle are represented graphically in Fig.9, where both flashing and nonflashing flows
are displayed. A correlation developed by / l l / gives results in close agreement with
the previous method: For w > 4 (low quality region):
G* = [0.6055 + 0.1356(lnw) 0.0131(lnw) 2 ] /w 0 " 8
(41a)
(416)
Furthermore, for the allliquid inlet condition, the correlation may be replaced by:
G = 0.9G L
(41c)
4/
2
D
G*2
7i - f?2
+ (1
w) 2
2 In
In
(1 u)ri2
(1
(l-u>)r)2
+w
W)T7I + C J
+C J /771
( l w ) 7 i + w \772
(42)
where 4fL/D
is the total equivalent pipe resistance, 771 = P i / p o and 772 = P2/P0 (see
Fig. 10). In order to calculate flow discharges from a large reservoir, both the inlet
and exit conditions have to be known. For the inlet, G* and 771 are related through
the relationship 39), with 77 = 77 j . Loss entry effects can be incorporated into the term
4/L/D.
For subsonic or unchoked exit conditions, P2 = pa where pa is the ambient
back pressure. For exit choking conditions, equation 41) with 77c = 772c provides the
relationship between G* and the critical exit pressure ratio, 772c In this way, for a
given 4fL/D,
we have threee equations for the three unknowns, G*, rji, T)2C- Fig. 10
displays the ratio of Gc for pipe flow to GQC corresponding to a nozzle, as a function of
267
4fL/D,
for various values of the parameter u>, covering both flashing and non-flashing
flow conditions.
Inclined pipe: Generalized solutions have been obtained by using a dimensionless
"Flow inclination" number:
Fi =
gDcos9
p0gLcos8
(43)
4f
fV,
[[1-W)V2+UTI]
\l-G'2p]dT]
(44)
which reduces to equation 42) for Fi = 0. Fig. 11 illustrates the effect of an inclination
F{ = 0.2 by comparison with Fig. 10 representing the Fi = 0 case. For more details
and for the extension of the "omega" method to include the effects of subcooling inlet
conditions and the effects of the presence of noncondensable gases in the mixture, see
/12/.
4.2. VENT SIZING FORMULAE
In this section the most commonly used methods for the design/verification of ERS are
presented. The case of pure vapour venting is not considered because it has limited
applicability / 1 3 / . Two-phase flow can usually be expected to occur from a runaway
reactor vent and the assumption of two-phase flow should usually be made as a safe
case. It is recalled t h a t the design/verification of an ERS in such conditions requires an
appropriate characterisation of the chemical system concerning the thermokinetic and
fluid dynamic aspects. This can be achieved only through adiabatic calorimetry testing
/ 1 0 / . One of the lectures of this Course is devoted to this topic. It is usually preferable
t h a t the relief device can operate at a pressure value pg ("set" value) well below the
maximum allowable working pressure (MAWP) of the reactor vessel. In addition,
the pressure during the venting transient may be allowed to rise up to a fixed value
Pm{Pm = l-2p s ) before starting the decay phase. In doing so, a significant reduction
of the vent area as compared to the case of zero overpressure (Ap = pm pa = 0) can
be obtained.
In order to decide which methods are applicable in a given case, it is necessary to
define the chemical system as one of the following types:
a)
268
b)
"Gassy systems", in which the pressure is due to the non-condensable gases generated by the reaction.
c)
"Hybrid systems", in which the total pressure is due to both vapour pressure and
gas generation.
"Vapour pressure systems". In Fig 12 the pressure and temperature time histories
of a runaway reactor in vented (dotted line) and non-vented (continous line) conditions,
are shown / 1 4 / . In the vented case the maximum pressure value p m is reached in a time
r p = Atp + Atv where Atv is the time required to generate heat for vapour formation
during venting and Atp is the time (adiabatic rise time) needed to reach pm in the
non-vented case. If the volatility of the system does not change during the venting,
the turnaround in pressure occurs at the same time of the turnaround in temperature
(r p = rt). The governing equations for the relief transient can be obtained considering
the mass balance:
dm/dt
= -GA
(45)
o r m = mo GAt
where m = pV and V is the volume of the reactor
{m3);
= mq-
GAVhlg/[mvig)
(46)
r = ra + g -
vh
'
' ,
(47)
= 0, we get:
(48)
V
'
269
(49)
(49a)
Gvhio
Fig. 13 / 1 4 / displays the vent area versus A p showing the benefits in the reduction
of the vent area which can be achieved with allowance of a small overpressure. Allvapour and all-liquid venting represent, respectively, top and b o t t o m venting modes
with complete vapour-liquid phase separation or disengagement. The treatment is
similar as the one previously described; the only difference in equation 46) is the ratio
V/m replaced by V| or v. For the zero overpressure, in the case of all-vapour flow the
required flow area is given by:
+ {dT/dt)a)
(50)
where dT/dt values are measured with the adiabatic calorimeter. Key parameters for
the use of this method are: V, m o , C, hig, v\g, and P T data.
For a non-reacting system subject to a constant energy input Q(VK) representing
the case of a storage tank exposed to an external fire, equation 46) becomes:
mCdT/dt
= Q - GAVhlg/(mvig)
(51)
which after using a treatment similar to t h a t in the runaway reaction case, yields the
following expression for temperature:
T +
ccln
GA
hig
o Cvig ( g j - t)
L GA
(52)
Q
W
GAC
,m0QVig
- 1 +
4-
-FT
m0Cvig
(53)
270
"Gassy systems". Let us consider the case of an ideal gas (accurate enough for
low pressure conditions) inside the contaiment vessel of an adiabatic calorimeter:
n = pVc/RTc
(54)
and Tc is
~ Vc/(RTc)dp/dt
(55)
then:
mg = Mwdn/dt
(56)
vg = RTp/[ppMw)
(57)
and considering t h a t :
we get:
Qg = mgvg
= VcTp/{ppTc){dp/dt)
(58)
where Mw is the molecular weight of the gas, pp is the peak pressure, Tp is the temperature at peak rate, and Qg is the volumetric gas generation rate evaluated at the
peak pressure during venting. For an adequate vent size, we equate the volumetric
generation rate to the volumetric discharge rate:
A = m0Qg/{maGv)
(59)
where m,Q is the reactor charge (kg), ma is the test cell charge (kg), v is the homogeneous two-phase specific volume V/mo where V is the reactor volume (m3), and G
is the two-phase discharge mass flux [kg/m2s),
or:
A = m20Qg/{meVG)
(60)
(61)
(62)
271
VcTp/{m3ppTc).{dp/dt)max{m3J{V2Ap)*
A =
(63)
"Hybrid systems" are those in which a non-condensable gas is produced but some
of the reactor contents vaporize during the runaway, providing a tempering effect. The
vent size formula is the equation 49) modified to account for the fact t h a t the total
pressure is the sum of the partial pressures of the vapour and the gas (p = pv + pg)The term hig/vig is replaced by (Tpv/p)dp/dt
/ 1 5 / and the equation 49) becomes:
qm0
A =
(^)*
The term (pv/p)dp/dt
given by / 1 5 / :
(64)
+ (CAr)i]
, ,Pv dp
+ (
2T
7A'
(65)
=
=
=
=
=
=
=
=
13.16
9500
0.45
482.5
0.493
0.54
492.7
0.662
m3
kg
MPa
K
K/s
MPa
K
K/s
They have been obtained via adiabatic calorimeter testing, The following mixture
property d a t a have been used:
For p
0A5MPa:
vi
vg
C
hig
=
=
=
=
0.001388
0.08553
2470
310.6
m3/kg
m 3 /kg
U/kgK
kJ/kg
272
For p = 0 . 5 4 M P a :
vi
vg
C
hig
=
=
=
-
m3/kg
m3/kg
kJ/kgK
kJ/kg
0.001414
0.07278
2514
302.3
= 10.2K
1.426kJ/kgs
GA =
2 =
, ,
256kg/s
{(9501O6XXO3OB4M)'+ (2470X10.2)*}
310600 /
\ '
0.08414 \ 2 4 7 0 x 482.5,/
= ^ ^ = 0.084m 2
and
, ,2
= 3040fco/m
s
y/
Using t h e more traditional vapour-venting model, the zero overpressure case (equation 49b) gives:
GA =
, .
37kg/s
*'
and the critical mass flux, evaluated with equation 41b) and with u> = 1.4 gives:
0.66
/ 41
1.4 0 - 39
\0.
37
2
1310 = 0.026m
= 1310fco/m2s
and
Dv = 0.189m
(all-vapour venting)
In Fig. 14, t h e pressure time histories, during venting and calculated using t h e
two-phase homogeneous vessel method, vent diameters D2p and Dv, are shown. It is
evident t h a t t h e vent size based on all-vapour flow will lead to vessel failure.
Pressure
A d d e d Interfacial
Mass Pressure
d7 (AkOk)
dk
dt
AKL>k
Dktih
D(
Ak
B k + A p ,
oz
Aahekfhk+ -^-1
Where:
dk
fi
d
= + u
dt
ot
oz
Bk = ( - 1 )
n =
A^n,agi<mv,
= A/'k
ut
Dg ?>g
D | l'|
o
dt
+ uk
Gravity
Cross
Section
Changes
I-a
Ark
[v,-vk)
= Ark(hk +
+ Pk
Interfacial
Wall
heat
heat
transfer transfer
Interfacial Wall
Friction Friction
Phase
Change
+ Arlk
+ Aakykgz
*k^kw
- ^
fl
f l a ^ p
fit
+ A p
*k<Pwk + A a k 0 k g z u k
/V-l-irr
OZ
Aqkl
+ Rk
O ^
ot
R k =lR
(l- k )p,4^
4
oz
;i-b
[I-c
274
103
100
Styrene
g = 500 kPa
600 -- ^ ^ ^
60
400 --
1 - 40
e ^
Q;
g 200
20
CD
- TJ
u>
. 03
W
- CO
6
E
- Q.
ra
n
.*:
m
:
4
I-
G/Q^C
100 --
10
1<-
ci)
Q.
>.
60 "
^^^
H
\- 4
40 -
ra
ti
o
"a:
0)
i _
CI)
20-
10
CO
"- 2
(/I
)
ra
m
.w
O
X
D
CO
CO
at
10
E,
0.2
E
o
0.4
0.6
Inlet void fraction
0.8
1.0
275
Nozzles
y "if"
Orifices (diaphragms)
Fig. 2 : Geometries considered for critical flows.
!76
mm
Slug or
plug flow
o ^0
o Q
o
a O a
" a
"o00 o
0 o
O a o
"-
I
Non -equilibrium
\^t
. |
Equilibrium
10.
8. re
6
x
H20
P0= 1.1 MPa
xo=0.001
6.
Homogeneous Equilibrium
Gc-
rP
s'
-\ 1/2
2 / vdP
y
L
P0
J
G= -
^V/-"
1
1
1
1
\
1
^=0.88
i
2.
0.20
G -
d v i -1/2
dP.
i
0.40
0.60
1
1
1
1
1
\
\
\
\
\
1
I
'
1
li
0.80
_P_
P.
1.0
278
Experimental
T
o LVD = 0
LVD = 12 I
D = 6 i
LVD = 40 I Fauske
LVD = 3
A LVD = 3.5
T LVD = 18 I Sozzi and Sutherland
D = 12.7 mm
a LVD = 6.5
* LVD = 29 I
Friedrich
LVD = 1.5
D = 6.0 mm
* Uchida & Narig
LVD = 12
D = 4.0 mm
o Zaloudek
LVD = 6
D =12.7 mm
Model Predictions
Upper bound flow (Frictionless)
Homogeneous Equilibrium Model (Frictionless)
Moody (Frictionless)
Henry & Fauske (Frictionless)
Edwards LVD = 12
Henry LVD = 12 (Frictionless)
12 O r
10.0
9.0
o
-
8.0
a 7.0
15
30 45
60 75
90 105
Stagnation pressure (bars)
120
279
EXHAUST
& SUPPLY
Fig. 8 : Small-scale test equipment with closed and open test eel
designs (from reference /10/).
280
_
1.2
1.0
08
Non-flash ng flow
Flashing flow
Pc/Po ^
: -r\
'
^_Pc
06 0.4
TD
: -J
Gc
Gc
n
0.01
V Po Co
02
i
0.1
0)= nn
1 1 1 1
^ ^
1 I 1 1 1 1 1
10
100
u> = a0 + Q 0 C P T 0 P 0 (vveo/hveo)z
100
4fU/D
Fig. 10 : Choked flow discharge from horizontal pipes /12/
281
100
4fL/D
Fig. 11 : Choked flow discharge from inclined pipe, with Fi 0.2
/12/.
System type
Sealed
Venting
282
u.o
Styrene example
13.16 m2 9500 kg
0.7
0.6
Computer simulation
(SAFIRE)
0.5 [
0.4
0.3
0.2
0.1
0
0
i
4.5
10
20
30
40
50
Percent overpressure
i
60
i
5.0
5.5
6.0
6.5
Maximum pressure bar abs.
70
i
7.0
90
Time (s)
120
180
283
5. R e f e r e n c e s
/l/
/2/
Leung, J.C., Fisher, H.G.; (1989),J. Loss Prev.Process Ind., April, Vol. 2
/3/
/4/
/5/
/6/
PI
/8/
/9/
Fisher, H. G.; (1988), Spring Nat. Meeting AIChE, New Orleans, March 6 - 10
/10/
/ll/
/12/
/13/
Duxbury, H. A., Wilday, A. J.; (1989), Int. Symp. on Runaway Reactions, Boston,
March 7 - 9
/14/
/15/
/16/
6. N o m e n c l a t u r e
o
A
A
B
C
D
f
/
Fi
G
g
h
H
K
L
n
m
Speed of sound
m/s
Surface area
m2
Matrix of the coefficients of the time-derivative terms of the eqns of motion
Matrix of the coefficients of the space-derivative terms of the eqns of motion
Liquid specific heat
J/kg.K
Cross section diameter
m
Vector of unknown variables in the eqns of motion
Friction factor (Fanning)
Flow inclination number
Mass
flux
kg/m2.s
Acceleration of gravity
m/s2
Specific enthalpy
J/kg
Elevation
m
Discharge coefficient
Length
m
Number of moles
Mass
kg
284
N Non-equilibrium coefficient
P Pressure
q Specific power
Qw, Heat flux
R Gas constant
T Source term vector in the eqns of motion
Q Power input
Qsj Volumetric gas generation rate
s Specific entropy
t Time
T Temperature
u Velocity
V Specific volume
X Vapour quality
X Liquid mass fraction
Y Vapour mass fraction
z Spatial coordinate
Pa
W/kg
W/m2
m2Js2K
W
m3/s
J/kg.K
s
K
m/s
mZ/kg
a Volumetric fraction
j3 Coefficient of the added mass terms in the eqns of motion
7 Isentropic coefficient
e Dimensionless specific volume
r Coefficient of the interphase exchange terms in the eqns of motion
9 Angular coordinate
77 Dimensionless pressure
A Eigenvalue of the matrix of the eqns of motion
u> Correlation parameter
p Density
X Coefficient in the wall exchange terms of the eqns of motion
T Time interval
r t K Shear stress
6.2. SUBSCRIPTS
0
c
t
k
j
g
v
/
m
max
s
p
Initial value
Calorimeter
Component (1,2)
Component (1,2)
Component (1,3) or (1,6)
Vapour/gas phase
Vapour phase
Liquid phase
Maximum value
Maximum value
"Set" value
peak value
kgm/s
m/s
kg/m3
l/m
s
Pa
J.C.LEUNG
Fauske & Associates, Inc.
16W070 West 83rd Street
Burr Ridge, Illinois 60521
U.S.A.
Introduction
This paper and the accompanying paper address the pressure relief
vent sizing methods for runway reaction systems. In addition, the
fire emergency sizing for non-reactive (storage) systems will be
included because of its prominence in the chemical industry. Runaway
reaction systems can generally be classified as tempered or nontempered types, see Table 1. A tempered system is one in which the
reaction heat can be removed by latent heat of vaporization, thus
controlling the runaway reaction from escalation in temperature.
This of course is accomplished by pressure relief venting.
Many of
these systems are in fact normally operated in the refluxing mode for
temperature control. The latent heat can be provided by either the
reactant(s) or the solvent(s).
However, the latter case presents
special concern If the system loses its solvent via boil-off due to
fire.
Note that a tempered system can accommodate a reaction that
gives off noncondensable gases as long as the reaction temperature
can be controlled, we call this type a tempered hybrid system. On
the other hand, a non-tempered system exhibits little or no latent
heat of cooling at all, this is typical of a low vapor pressure
system. If the reaction products are also of low vapor pressure, the
pressure relief requirement can be quite small. But quite often the
reaction product(s) are noncondensable gas(es), these are so-called
"gassy" systems. For these non-tempered systems, the heat release is
largely retained in the runaway reaction mass and if left unattended,
without cooling re-initiation, quenching or dumping, may lead to very
large temperature and pressure excursions.
As might be expected, the vent sizing methods differ depending on
the system type. In this paper only the tempered vapor (or sometimes
being called tempered volatile) system will be addressed. The other
systems will be covered in the accompanying paper.
In the past, several methods of calculation have been proposed for
sizing emergency relief system (ERS) for runaway reactions.
One
common, but frequently nonconservative method, is based on vapor
venting alone. As noted by several early observers (Boyle, 1967;
Harmon and Martin, 1970; Huff, 1973), the most realistic case should
285
A. Benuzzi and J. M. Zaldivar (eds.). Safely of Chemical Batch Reactors and Storage Tanks, 285-298.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
286
be based on the release of a vapor-llquid mixture with two-phase
discharge in the relief system. This conclusion was supported in
part by actual pressure relief experience from industry.
This was
also the recommendation of the Design Institute of Emergency Relief
Systems (DIERS) (Fisher, 1985; Huff, 1988), a consortium of 29 companies under the auspices of AIChE formed in 1976 to develop improved
methods for ERS design. The methods presented here are either taken
from this technology or extensions of it.
Theory of Pressure Relief
Simply stated, pressure relief of a runaway reaction requires a
balance between the volumetric vapor generation rate V
and the
volumetric discharge flow. This criterion can be derived from first
principles via consideration of mass and energy balance (Leung,
1986).
The above statement must be satisfied at the maximum
(turnaround) pressure during relief. Codes may require this maximum
pressure not to exceed 1.1 times or 1.21 times the design (gage)
pressure of the vessel. In equation form this criterion can be
written as
W
V - - - GA v
v
p
(1)
^ '
where W is the so-called relief vent rate (kg/sec), G is the discharge mass flux (kg/m sec), A is the vent area (m ) , p is the fluid
density (kg/m ) discharging from the reactor, and v is the fluid
specific volume (1/p).
In general the two-phase fluid density
depends on the vapor volume fraction, a (or void fraction) via
p - a p
+ (1 - <*)P = (1 - * ) P j i
(2)
where p
and p.
are the vapor density and liquid density, respectively. The discharge mass flux in Eq. (1) also depends on the fluid
conditions and is often limited by the choking (sonic or critical
flow) phenomenon at the minimum discharge flow area.
The volumetric vapor generation rate necessary to remove the exothermic reaction heat is given by
Q
m q
,-,
rxn
^rxn
V - r
-
v
v
h .p
h v
vi v
vi
...
'
v(3)
where Q
is the total heat release rate (J/sec), q
is the
specific neat release rate (J/kgsec), m is the reacting mass at the
point of pressure turnaround, and h . is the latent heat of vaporization.
Strictly speaking, Eq. (3) should be written as V
=
(Q
/h )v where v denotes the net increase in volume per unit
p
rxn' vi' vi
. vi
^ n
n _
mass upon vaporization. But at low pressure v . - v
- v . - v and
Eq. (3) results. The traditional vapor vent sizing formula can be
obtained from Eqs. (1) and (3) with p p and m - m , the reacting
mass at the start of relief venting, i.e.,
287
m qn
o rxn
W r
vi
._.
vapor venting
,,
(4)x
(5)
q
p
oo nrxn
-rxn
vi p v
m q
oo ^rxn
-rxn r
vi *v
, ,,
. -i
(6)
According to this formula, the term G/p, which can be regarded as the
volumetric flow per unit area, plays a key role in determining the
vent size. It is therefore of interest to examine its variation for
a wide range of two-phase flow inlet conditions.
Using styrene properties at 500 kPa, Fig. 1 shows that both the
choked flow mass flux G and the fluid density p decrease with increasing void fraction a entering the relief piping.
In contrast,
the volumetric flow term G/p is shown to increase rapidly with increasing void fraction. For example, we find in this illustration
that the volumetric flow per unit area is nearly 20 times larger for
all-vapor venting (a 1) than for two-phase venting with a = 0.2.
This would translate to vent area of 20 times larger for this twophase venting case than for vapor venting, if in fact no overpressure
(i.e., pressure above relief set pressure) is permitted. As will be
shown, by allowing for overpressure, this difference in vent size can
be greatly reduced due to rapid mass (chemical "fuel") depletion in
the case of two-phase venting. By the time the pressure turnaround
criterion is approached (i.e., Eq. (3)), the mass remained can be
significantly less than the Initial mass prior to venting (i.e., m
mQ).
Venting With Overpressure
The simplest method of incorporating the effect of overpressure on
vent size is that due to Boyle (1967). He defined the required vent
area as that size which would empty the reactor before the pressure
could rise above some allowable overpressure.
This can be represented mathematically as
m
^ G A T
(7)
P
Here the emptying time, equated to the pressure rise time At , is
based on the pressure history obtained from adiabatic runaway "computation or data in a non-vented system (see Fig. 2). Thus according
to Eq. (7), the vent area continues to decrease with increasing
overpressure. However this method does not yield realistic result as
no overpressure is approached. For the case of zero overpressure,
Eq. (7) would give an infinite vent size (since At - 0) which is at
288
odds with the result presented earlier, Eq. (6). This discrepancy is
due to the neglect of the energy balance consideration in Boyle's
method, i.e., the effect of tempering and latent heat cooling has not
been considered.
This limitation was overcome in Leung's formulation. The relief sizing formula for homogeneous-vessel (or uniform
froth) venting is given as (Leung, 1986),
m
q
o ^
h 11/2
m
(8)
(V T )
where
q " 0.5 C p |^|
+ |^| |
(9)
Here q represents the average heat release rate during the overpressure venting period, (dT/dt) and (dT/dt) are the self-heat rates
corresponding to the set pressure (temperature) and the turnaround
pressure (temperature), respectively (see Fig. 2). For no allowable
overpressure (AT - 0), Eq. (8) reduces to Eq. (6) exactly.
The effect of overpressure on vent size can be clearly illustrated
by Eq. (8) together with the experimental data reported by DIERS.
Figure 3 is a dimensionless plot of A/A
versus AP/P
showing the
homogeneous-venting prediction of Eq. (8) and the styrene relief data
observed in both large (2000 ) and small (32 i) scale facilities
(AIChE/DIERS, 1986a). The agreement is encouraging. The discrepancy
is due to the observed vapor-liquid disengagement behavior which
results in lower overpressures than the idealized (homogeneous-vessel
venting) case. Note that by allowing for an overpressure corresponding to AP/P of 0.1, Eq. (8) suggests a vent area reduction of about
six times from the no overpressure case (i.e., A/A
0.15).
This
reduction is quite significant and is mostly due to the mass depletion via venting, resulting in a much lower total energy release at
the pressure turnaround point.
Figure 3 also depicts that at higher overpressure the relative
reduction in vent area becomes increasingly smaller. This portion of
the curve is in essence operating on Boyle's emptying time theory
with most of the energy release being stored as sensible heat in the
reacting mass. The ultimate turnaround in pressure is due primarily
to the emptying of the reactor. Due to the Arrhenius behavior typical of most reactions, the simple averaging approximation for q, as
suggested by Eq. (9), gets progressively worse at higher overpressure, resulting in some cases a minimum in the area
versus
overpressure curve (see Leung and Fisher, 1989). For these situations better approximation of the non-linear heat rate can be given
by
q - C
AT
(10)
289
where At
is based on the pressure history obtained from adiabatic
runaway computation or data in a non-vented system.
An alternative
would be to borrow the results from thermal explosion theory
(Townscend and Tou, 1980) where
2
s
E
A (dT/dt)s
T
2
T
m
(dT/dt)r
(11)
can be used to approximate the pressure rise time in Eq. (10). Here
E
is the apparent activation energy and E /R can be obtained from
the slope of an Arrhenius self-heat rate plot. It is worth noting
that Eq. (8) in combining with Eq. (10) would reduce to Boyle's
formula, Eq. (7) as high overpressure is approached.
The abovementioned vent sizing methods require the evaluation of two-phase
mass flux G. We shall postpone this discussion towards the end of
the accompanying paper where a complete and generalized treatment for
the mass flux will be presented.
Finally, a very simple method for sizing was advanced by Fauske
(1984a, 1984b) and this has come to be known as the nomograph method.
In the most recent form it can be represented by (FAI, 1989)
A = 1.5 x 10
m (dT/dt)
s
(12)
2
where the particular units are: A[-] m , m [-] kg, dT/dt[=] C/min,
P [=] psia. This equation was developed based on an overpressure to
set pressure ratio AP/P
of 0.2. Note that this is different from
the traditional meaning or 20% overpressure since by convention the
percent overpressure is defined based on the gage set pressure.
Vapor-Liquid Disengagement
The homogeneous-venting model is an idealized flow regime with no
phase separation at all inside the reactor (Huff, 1982). Because of
its simplicity in treatment and its conservatism, this uniform-froth
model offers a useful tool for ERS design.
While a real venting
process falls somewhere between all-vapor venting and homogeneous
venting, the amount of phase separation or disengagement depends
largely on the particular fluid behavior and the vapor superficial
velocity. From studies of gas sparging through liquid columns, two
rather distinct flow regimes have been identified - the bubbly and
the churn-turbulent regimes. These flow regimes are best described
by the drift flux model (Zuber and Findlay, 1965; Wallis, 1969) and
have been used extensively to characterize the in-vessel fluid behavior during the course of the DIERS research program (AIChE/DIERS,
1986b). The complete vapor-liquid disengagement criterion can be
stated as follows.
If the vessel free-board volume fraction (void
fraction) a exceeds the disengagement void fraction a , then vapor
venting is predicted. The latter is governed by the following drift
flux relationships for bulk heating:
290
^ -
- a (1 - a )
J*.
2a
1 - an
(13)
(14)
CO
where j
is the superficial vapor velocity at the top of the vessel.
At the turnaround condition, j
is given by
.
rxn
g=> " h ,p A
/\z,\
rxn
" h .p A
vi V X
Ci:>;
V* V X
&(Pjl
Pw)
1/4
Here U
iV4
is
the
bubble
(16)
2 + IA
churn
flo w
(17)
(18)
where i p = Q
/ ( h . p A U ) , can be defined as the dimensionless
rxn
vi v x ^
superficial velocity.
These solutions are presented graphically in
Fig. 4. If a
> a n , vapor venting (complete disengagement) is
predicted.
If a < a., then two-phase discharge upon relief opening
will occur. Under such situations, analytical equations have been
developed (Leung, 1987) for both the bubbly flow regime and the churn
flow regime. However, these solutions are not as simple or explicit
as Eq. (8) for the homogeneous-vessel venting case, but they can be
readily solved by numerical method.
Figure 5 is an illustration of the influence of flow regime on vent
size based on a styrene runaway example (Leung, 1986; Leung, 1987).
Note that the bubbly regime yields nearly the same vent size as the
homogeneous regime. The churn regime, on the other hand, yields
significantly smaller vent size, and in this example the vent size is
nearly the same as the all-vapor venting area if a 40% overpressure
is allowed.
Finally, some useful approximations for these flow regimes are
given here. For bubbly flow the venting process resembles closely to
that of the homogeneous-vessel case particularly when i j> > 0.25.
According to Fig. 4 or Eq. (17), no vapor disengagement is possible
when the dimensionless superficial velocity (i p) is greater than 0.25.
For churn flow, significantly more vapor disengagement takes place.
An asymptotic solution was suggested which yielded both simplicity
and acceptable accuracy. The vent area as given by Leung (1987) is
291
(")
where
(1 - r) - - J
|V2
% -
<F\
(21)
Crozier (1985) has summarized various sizing formulae for fire emergencies which were all based on vapor venting assumption. His
discussion dealt mainly on the underlying fire heat fluxes assumed in
these formulae.
The NFPA (1984) chart was found to be most widely
followed. For a wetted area of less than 19 m
(200 ft ), the
average fire heat flux is 63 kW/m (20,000 Btu/h-ft ). For larger
vessels with A > 19 m , the average fire heat flux actually falls
off due to the fact that larger vessels are less likely to be fully
exposed to fire. It should be noted that a local fire heat flux as
high as 110 W/ra
(35,000 Btu/h-ft ) has been measured (API, 1976;
Moodie and Jagger, 1987). This value applies to direct flame impingement situations only. Unless the whole vessel is fully engulfed
by the fire, the average flux of 63 W/m is widely accepted (A < 19
m ).
The fire emergency sizing also depends on the flow regime behavior
of the liquid under consideration. First of all, if the liquid is a
foamy fluid, i.e., the regime is in bubbly flow, then the relief
sizing treatment resembles closely to that for the runaway reaction.
In the case of fire, the dimensionless superficial velocity TJ> is
given in terms of
Q
^-h-T^A-ir
<22>
V f V X <*>
(23)
292
If ip > 0.25, homogeneous-vessel venting is a close enough approximation. The required vent area is given implicitly by the following
equation (Leung, 1986)
V h
Vi
(24)
m C v .
o p vJ
If TJ> < 0.25, the venting behavior in bubbly regime can differ
substantially from the uniform-froth case. For lack of any better
method, the relief sizing equations for constant heating in bubbly
regime as given by Leung (1987) are suggested. At present insufficient data exists for further refinement (although preliminary data
suggests that the above methods yield too conservative vent sizes).
For non-foamy system or churn-turbulent fluid, the influence of
external wall heating differs significantly from that of bulk heating.
The vapor is found to nucleate on the vessel wall and upon
detachment forming a boiling two-phase boundary layer, see Fig. 6.
An analysis that describes the void distribution and thickness of the
boundary layer, both of which increase with vertical distance along
the wall has been presented (Grolmes and Epstein, 1985). The amount
of liquid swell as a result of the rising bubble in the boundary
layer was found to be significantly less than the case where bubbles
are created and rising throughout the bulk of the liquid. In fact, a
reported experimental investigation suggests that in venting assessments of liquid-filled vessels subjected to external fire, it is
acceptable even to ignore the effects of liquid swell for non-foamy
liquid (Fauske et al., 1986). However, the possibility of two-phase
flow to the vent still exists if the velocity of the venting vapor
within the free-board volume is such that a droplet spray is pulled
off the liquid surface and carried by the vapor to the vent (Epstein
et al., 1989). Usually this occurs at a very high quality (vapor
mass fraction) flow condition. Nonetheless, it causes a significant
increase in pressure drop across the vent and particularly detrimental to atmospheric-type storage tanks where allowable overpressure is
very limited. Epstein et al., (1989) has developed a criterion for
the onset of two-phase venting via entrainment consideration, see
Fig.
6. From this criterion, a minimum head space
clearance h . can
r
, D, . ,
min
be derived,
AT -
GAC
in
m
o Q F V vi
V GA h
vi
1/2
2 p h ,U
v vi E
(25)
"gP,
1/4
(26)
293
111
tr
I
<
DC
HI
0.
LU
I-
~^
1X1
cc
ID
CO
CO
HI
tr
0_
- 1 l
600
;-
l l
B81031 H A
STYRENE
~- ^^-~~~
~
-
j-
P ^
<
LU
CC
Q:
200
"
<
CO
10,
rn
ra
CO
6 -
*-
4 -
<
ni
i
o
5
1
60
40
~'^--"
~
CC i'
10 LU E
\ - 6
\-
i-
G//3 . , /
LU
Q
III
o
>
CO
CO
<
l I 1 1
0.2
0.4
1
0.6
1 1
0.8
294
0.5
D I E R S ICRE S e r i e s
Scale(J)
0.4
Venting
32
32
2000
Top
Bottom
Top
0.3 \
<
v.
< 0.2
Homogeneous
N J~ V e n t i n g
0.1
Vessel
0
o " - -
I
0.2
0.|1
0.3
0.4
0.5
AP/R,
=r I
'
'
'
Hi
a"
Bubbly Flow
'
"
o
0.1
0.01
Figure 4.
0.2
0.4
0.6
0.8
DISENGAGEMENT VOID FRACTION, (XD
T^TII mil
<
295
PbAK PRESSURE, bar(abs)
5
5.5
6
6.5
Styrene Polymerization
9500 kg in 13.16 m 3 Tank
Homogeneous
Venting
0.05
Bubbly
Vapor
-Venting
Area
Churn
20
40
60
PERCENT OVERPRESSURE
Figure 5. Vent area versus overpressure for
styrene runaway example in different venting
regimes.
Vapor
ICL.010204
C.A
296
TABLE 1. Classification of reactive systems
TYPE
CLASS
SOURCE OF PRESSURE
Vapor System
Tempered
Hybrid System
Tempered
Hybrid System
Non-Tempered
Gassy System
Non-Tempered
Q
F
= h-^
>
f 0 r h > h
P h U_,
r
v vi E
<27>
min
0.61 |P p
for vapor choked
h . , then an alternative is to
just below the droplet entrainet al., 1986),
for h < h .
mm
(28)
Conclusion
This paper provides a state-of-the-art summary of practical vent
sizing methods for tempered vapor systems. Both runaway reaction and
fire emergency of non-reactive storage have been discussed.
The
effect of two-phase discharge on vent size can be elucidated and
recommendations are provided for different venting flow regimes.
References
American Institute of Chemical Engineers (1986a), "Phase III
Large Scale Integral Tests, DIERS III-5, Experimental Results
for Series III Tests, DIERS III-6, Experimental Results for
Series IV Tests, Analysis and Program Summary", Reports submitted by Fauske & Associates, Inc. for AIChE/DIERS.
American Institute of Chemical Engineers (1986b), "Emergency
Relief Systems for Runaway Chemical Reactions and Storage
Vessels:
A Summary of Multi-Phase Flow Methods", Report
submitted by Fauske & Associates, Inc. for AIChE/DIERS.
297
American
Petroleum
Institute
(1976), API RP-520, Fourth
Editions,
"Recommended
Practice
for
the
Design
and
Installation of Pressure-Relieving Systems in Refineries, Part
I - Design and Part II - Installation", API, Refinery
Division, Washington, D.C.
Boyle, W. J., Jr. (1967), "Sizing Relief Area for Polymerization
Reactors", Chem. Eng. Prog., 63 (8), p. 61 (August).
Crozier, R. A., Jr. (1985), "Sizing Relief Valves for
Emergencies", Chem. Eng. Prog., p. 49, October 28 issue.
Fire
Reactor Vent
298
Kutateladze, S. S. (1972), "Elements of the Hydrodynamics of
Gas-Liquid Systems", Fluid Mechanics - Soviet Research, Vol.
1. P. 29.
Leung, J. C. (1986), "Simplified Vent Sizing Equations for
Emergency Relief Requirements in
Reactors
and
Storage
Vessels", AIChE Journal, 32 (10), pp. 1622-1634.
Leung, J. C. (1987), "Overpressure During Emergency Relief
Venting in Bubbly and Churn-Turbulent Flow", AIChE Journal, 33
(6), pp. 952-958.
Leung, J. C , Buckland, A. C., Jones, A. R., and Pesce, L. D.
(1988), "Emergency Relief Requirements for Reactive Chemical
Storage Tanks", Institute of Chemical Engineers Symposium
Series, No. 110, p. 169.
Leung, J. C. and Fisher, H. G. (1989), "Two-Phase Flow Venting
from Reactor Vessels", J. Loss Prev. Process Ind., Vol. 2, p.
78.
Moodie, K. and Jagger, S. F. (1987), "Flow Through Pressure
Relief Devices and the Dispersion of the Discharge", Inst, of
Chem. Engrs., Symp. Series No. 102, p. 215.
National Fire Protection Association (1981), "National Fire
Codes", NFPA-30 Flammable and Combustible Liquids Code, NFPA,
Quincy, MA.
Townscend, D. I. and Tou, J. C. (1980), "Thermal Hazard
Evaluation by an Accelerating Rate Calorimeter", Thermochimica
Acta, Vol. 37, p. 1.
Wallis, G. B. (1969), One-Dimensional Two-Phase Flow. Chapter 4,
McGraw-Hill, New York.
Zuber,
N.
and
Findlay,
J. (1965), "Average Volumetric
Concentration in Two-Phase Flow Systems", Trans. ASME J. Heat
Transfer, Vol. 87, p. 453.
J.C.LEUNG
Fauske & Associates, Inc.
16W070 West 83rd Street
Burr Ridge, Illinois 60521
U.S.A.
Introduction
In the previous paper, vent sizing methodology for tempered vapor
systems has been presented. In this paper the gassy system and the
hybrid system are examined. As indicated in Table 1 of the previous
paper, two general types of hybrid system can be found. The type
that does not exhibit any sustained tempering is the non-tempered
type (either due to solvent boiled off or insufficient latent heat of
cooling) and this system exhibits essentially the same characteristic
as the gassy system - the source of pressure is dominated by the
evolved gas mostly. The other is the so-called tempered hybrid
system and as expected, this system exhibits similar tempering
(latent heat of cooling) characteristic as the vapor
system.
However, if not properly relieved, the accompanying gas evolution can
cause the reactor pressure to rise much more rapidly.
We will examine the relief requirement for these systems and the sizing methods
in this paper.
Theory of Pressure Relief
In general the pressure relief of a runaway reaction requires a
balance between the volumetric vapor/gas generation rate and the
volumetric discharge flow. This criterion can be represented mathematically as
m q
o^rxn ,
T
+ m m
\t
- ^ r
(l)
q
n
0 u
:
rxn
'v
'g>
299
A. Benuizi and J. M. Zaldivar (eds.). Safety of Chemical Batch Reactors and Storage Tanks, 299-310.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
(2)
300
where p
and p are the vapor and gas density, respectively. From
ideal gas law, p S - P M y'RT and p
- P M /RT.
Again the above
equations apply to the case of no^overpressure with an initial batch
size of i , It should be noted that according to the partial pressure definition, the total volumetric generation rate is not equal to
the sum of the vapor volumetric rate (V - m q
/h p ) and the gas
volumetric rate (V = m m /p ) . In Eqs. (1; and (2; , m denotes the
specific mass rate of gas evolution. Experimentally it ^can be estimated from pressure rise rate data via
VM
-m
Ut
v
/mt
-'meas
where V
is the free-board volume the gas occupies in the test apparatus, (dP/dt)
is the measured pressure rise rate, and m
is
the test sample mass.
Note that in the case of vapor relief, the
vent rate requirement is simply
Va r
wfJventingJ
P 1 - mo
q
rxn
r
+ m
V wv
g wg
Thus the difference in relief vent rate between homogeneous venting
and vapor venting is very large and can be given by
.[
..(homogeneous^
W
I venting I
wf vapor 1
,.
.
(1 - a )pr
o P
_.
m /V
o'
p
r
+ pH
p
H
+ pH
(6)
^ '
^ventingj
where a is the initial free-board volume fraction in the reactor.
In what follows we will discuss the vent sizing implication and the
merit, if any, for allowing for overpressure in both gassy and hybrid
systems.
Gassy System Vent Sizing
The non-tempered gassy runaway system is deficient in latent heat of
cooling which is used in vapor system to control the runaway by
venting.
For the gassy system, the temperature runaway history
remains essentially identical to the adiabatic case even in a vented
environment.
Figure 1 illustrates a typical venting scenario for a
gassy system. Due to the continued accumulation of gas given off
from the runaway reaction, the rupture disk soon reaches its relief
pressure and causes the reactor to depressurize quickly to ambient
pressure or to the back pressure imposed on the downstream equipment
301
(Pt. 1 in Fig. 1). Two-phase flow is not expected since the gas
evolution rate is quite low at this early stage. The assumption of
two-phase discharge would be non-conservative (unsafe) here since it
causes early emptying of the reaction mass. After this initial
"blow-off" of the gas, the reactor temperature continues to rise
while the system remains at the prevailing back pressure (Pt. 2 in
Fig. 1). Note that the use of a reclosure-type relief valve would
keep the system pressure at close to the set point with periodic
opening to bleed off the excess gas, but the temperature heatup
profile remains unaltered.
As the reaction approaches the maximum
rate prior to complete conversion (decomposition), pressurization of
the reactor will take place (Pt. 3 in Fig. 1 ) , and this will most
likely be accompanied by two-phase discharge.
The worst case to
consider
is therefore the onset of homogeneous-vessel venting
coinciding with the peak reaction rate (as indicated by maximum dT/dt
or dP/dt rate), see Fig. 2. Here the general vent rate equation, Eq.
(5), for gassy system reduces to
m
W
g.max
m
o
g
Evaluating this requirement at the maximum allowable
(after substituting Eq. (3) into (7)) yields,
w
max
(7)
max
P
m
V /-.-> max
_ _o _cfdP]
m
pressure
P Idtl
mv J \
(8)
To estimate the required vent size for the allowable overpressure (AP
= P - P ),
m
s
W
A = ^ f
(9)
where G is to be evaluated at P and at the reactor loading of m /V.
For gassy system, the two-pnase flow can be characterized as nonflashing flow discharge; this is in total contrast to the flashing
flow regime for the vapor system.
A unified treatment for both
regimes has been developed and will be presented later on.
Starting with Eq. (8) and after making a number of conservative
approximations, Fauske arrived at a simple formula in terms of the
maximum pressure rise rate measured in either the RSST (FAI, 1989) or
the VSP equipment,
AA
m (dP/dt)
vK _
meas
- . ^Jl
,, .
<10>
m
where the coefficient K is 3 x 10" for the RSST and 3.3 x 10"
for
the VSP (this difference accounts for the different containment
vessel free-board volume). Here the particular units are: A[=] m ,
dP/dt[=] psi/min, P [-] psia.
So far no attempt nas been made to account for the mass loss due to
decomposition
and
venting.
The latter involves a transient
302
(unsteady) calculation during the venting process.
This effect is
expected to result in about a factor of two reduction in vent size
from the above methods. Early venting prior to the peak rates requires
significantly more analysis and physical (flow regime)
characterization. This is best investigated experimentally by the
direct scaling approach as discussed below.
The so-called area-to-charge direct scaling approach (Leung and
Fauske, 1987) offers an alternative method which usually yields a
smaller vent size than the calculational methods given above.
This
is because early loss of reactant from the reactor due to two-phase
flow is an advantage for non-tempered systems. Since this loss is
always more effective in the reactor than in the test apparatus due
to much larger superficial velocities, direct scale-up in apparatus
like VSP using top venting is hence possible. A vent size that allow
safe venting of the test sample and empties its content completely
(this is most important) can be safely extrapolated to full scale
based on area-to-charge scaling. Often a number of tests may be
required to narrow in on the size that limits the overpressure to
just below the allowable level.
It should be mentioned here that the Boyle's emptying time formula
(Eq. (7) of previous paper) can yield nonconservative results also if
the pressure rise time At is evaluated at a point where the rate is
substantially less than the peak reaction rate. On the other hand,
it has been found that if the peak reaction rate (m
) was used to
2 max
calculate At , the resulting vent size would be overly conservative.
Hybrid System Vent Sizing
The tempered hybrid system possesses volatile
or
condensable
component(s) which can be counted on to control the runaway temperature via latent heat of cooling.
Figure 3 illustrates a typical
venting scenario for a hybrid system. Although this illustration
uses a reclosure-type relief valve (PSV), the general venting trend
is not much different in the case of a rupture disk. Here PSV would
"lift" first when the system pressure attains the relief set pressure
and bleed off the accumulated gas (Pt. 1 of Fig. 3). The lifting and
reclosing of the PSV might occur many times while the vapor pressure
of the volatile component(s) continues to increase due to rising
temperature. It is not anticipated that two-phase discharge would
occur during these early periodic relief intervals. However, as the
tempering point is approached, the vapor pressure contribution
noticeably increases and the total vapor and gas volumetric rate
rises rapidly. Typically this is the point when two-phase discharge
simultaneous with overpressure venting would take place (Pt. 2 of
Fig. 3). For an adequately sized vent, the pressure would turn
around before the maximum allowable pressure and the temperature
would turn around also. Due to the effect of gas accumulation, the
pressure turnaround and the temperature turnaround do not occur at
the same time (Pts. 3 and 4 of Fig. 3).
The so-called tempering condition is achieved when the evaporative
cooling becomes equal to the reaction heat release.
This condition
is attained at the temperature turnaround point (Pt. 4 of Fig. 3).
303
The tempering condition can be determined experimentally in apparatus
like the VSP or RSST. Mathematically we can write the ratio of the
partial pressures as
m
q
^rxn
-g_
(11)
MW
h .M
vx. wv
. S.
v
v
In addition, these partial pressures add up to the system total
pressure P,
For vent sizing purpose, one is most interested in
T\
emperature T
at the relief device set
pressure P . Thus with P equal to P , we can solve for the tempering temperature T__ which yields the following vapor pressure
P
x /M
-g _ g w g _
P
x /M
P (T
(12)
) 1 +
A m
(13)
1/2
+ (C AT)
P
AP
(dP/dT)
AP
vi
Vvi
m m RT
g
a VM (dT/dt)
o wg
's
(14)
304
system).
A recent development based on simple mixture consideration
has been suggested for the calculation of the discharge mass flux G
in Eq. (13). This will be presented in the next section.
Another vent sizing method for hybrid system is that based on the
RSST technology and the simple equation approach (FAI, 1989; Creed
and Fauske, 1990)
Based on AP/P of 0.2, the following formulae are
suggested:
m
6
A 5 . 6 x l 0
(dP/dt)
(for RSST only)
p3/2
(15)
s
(dT/dt)
1.5 x 10
(16)
vlo
+ (1 a )p.C
T P
o r SL p o vo
vio
+ (1
G* ===
0*0
305
non-flashing component. The relative amount is conveyed by the inlet
gas mole fraction y
in the vapor phase. The generalized solution
for the normalized two-phase mass flux in a nozzle is given by (Leung
and Epstein, 1991)
a yy I n r r 6 - + (1 - a )y
o go
P
o -'go
6
6
go
G*
w(l
V l n p~ +
"
"P
(1
gJ
) d - y g0 >
1/2
(17)
vo
+ 1
g g
+ (1
(19)
Thus with the downstream pressure slightly below P , Eqs. (18) and
(19) can be solved for P /P
and P /P , which can then be substituted into Eq. (17) for calculating G*. This G* will correspond
to unchoked (subsonic) flow condition. However, as the downstream
pressure is reduced, G* reaches a maximum at the so-called choking
condition.
By definition further reduction in downstream pressure
has no effect on G*. At this choked condition, the exit pressure is
given by the choked or critical pressure which is found from the
following expression
p )
- a yJ
I n r 2 - + (1 - a )y
1
o 6go
P
o 'go
- i*go
l
gJ
P
- -(i - y g 0 ) l n
f~ + (i - ) d - y g 0 )
vo
2
2
y
(1 - y 0> fP 1
fP
1
1
R
-gV
+
2 a g P
u
P !0
o
.
I gJ
-.2
+ 1
P 1
V
P
vo
(20)
306
TIME
Figure 1.
Homogeneous Venting
Non-Tempered System
(Gassy)
ca
o
CO
o>
o
Tempered Systems
(Vapor/Hybrid)
a.
T7
a
T3
TP
'MR
TEMPERATURE ( R e c i p r o c a l S c a l e )
Figure 2.
307
LU
PSV
cc
I-
<
DC
LU
Q_
LU
I-
LU
CC
Z)
CO
co
LU
cc
CL
TIME
Figure 3.
1.4
1
^ - ^
I C L . 9 0 0 4 1 6 A.A
1 I I I lll|
I I lll|
1 II I
N o n - F l a s h i n g Flow
system.
1I
I I I III
Flashing Flow
0.8
0.6
0 4
0.2
10
w
Figure 4.
100
--a0+PoCPTopo(Whvlo)
308
0.2
Figure 5.
j_
_L
0.4
0.6
0.8
INLET VOID, aQ
1.0
309
This
is
or P /P
EO
"v
as it
VO
- 0.9 - ^
V
vio
1
T C
(21)
J op
2Q
References
Boyle, W. J., Jr. (1967), "Sizing Relief Area for Polymerization
Reactors", Chem. Eng. Prog., 63 (8), p. 61.
Creed, M. J. and Fauske, H. K (1990), "An Easy, Inexpensive
Approach to the DIERS Procedure", Chem. Eng. Prog., 86 (3), p.
45 (March).
Fauske & Associates, Inc. (1989), "Reactive System Screening
Tool (RSST) System Manual - Methodology and Operations",
Report No. FAI/89-73.
310
Leung, J. C. (1986), "Simplified Vent Sizing Equations for
Emergency
Relief
Requirements
in Reactors and Storage
Vessels", AIChE Journal, 32 (10), p. 1622.
Leung,
J. C. and Fauske, H. K. (1987), "Runaway System
Characterization and Vent Sizing Based on DIERS Methodology",
Plant/Operations Progress, 6 (2), p. 77.
Leung, J. C. and Grolmes, M. A. (1988), "A
Generalized
Correlation for Flashing Choked Flow of Initially Subcooled
Liquid", AIChE Journal, 34 (4), p. 688.
Leung, J. C. (1990a), "Two-Phase Flow Discharge in Nozzles and
Pipes - A Unified Approach", J. Loss Prevention Process
Industries, Vol. 3, p. 27.
Leung, J. C. (1990b), "Similarity Between Flashing and NonFlashing Two-Phase Flows", AIChE Journal, 36 (5), p. 797.
Leung, J. C. and Epstein, M. (1991), "Flashing Two-Phase Flow
Including the Effects of Noncondensable Gases", ASME Trans. J.
of Heat Transfer, 113 (1), pp. 269-272 (February).
J.L. GUSTIN
RHONE-POULENC INDUSTRIALISATION
24, AVENUE JEAN JAURES
69151 DECINES CEDEX B.P.166
FRANCE
ABSTRACT. This paper gives an overview of the DIERS methodology for vent sizing
for runaway reactions. The DIERS vent sizing methods rely on experimental data
obtained under adiabatic conditions. The theoretical background of practical interest
for pseudo-adiabatic calorimetry is given. This theoretical background provides
methods to correct the data obtained in pseudo-adiabatic calorimetric devices to true
adiabatic conditions.
The experimental methods available for the system characterization
acquisition are described. These experimental methods are :
and
data
311
A. Benuzzi and J. M. Zaldivar (eds.), Safety of Chemical Batch Reactors and Storage Tanks, 311-354.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
312
1 -
INTRODUCTION
Emergency Relief Systems (E.R.S.) are provided on vessels containing a
chemical compound or a mixture prone to undergo a runaway reaction, to
prevent these vessels from exploding.
The previous methods issued by the A.P.I, which only took into account gaseous
release could lead to undersizing by a factor of 6-10, should a runaway reaction
be initiated (1).
The DIERS Methodology for vent sizing includes the following basic steps
Step 1 : Definition of the worst credible deviation of the process, to provide the
design case for vent sizing.
Step 2 : Characterization of the reacting system behaviour, using pseudoadiabatic experimental techniques. The reacting systems are divided in
three classes :
- High vapor systems
- Gassy reactions
- Hybrid systems.
Step 3 : Acquisition of the experimental data necessary for vent sizing. The
nature of the data required depends on the nature of the reacting
system. The data must be obtained under conditions close to adiabatic
for a correct simulation of the runaway behaviour.
313
Step 4 : Choice of the vent sizing method and of the two phase flow calculation
method, according to the system behaviour.
The purpose of this paper is to describe the experimental methods available for
the system characterization and data acquisition for vent sizing and to discuss
the relevance of the various methods on an experimental and theoretical basis.
The definition of the worst credible deviation of the process and the discussion
of the choice of the most suitable method for vent sizing and two phase flow
calculations, are beyond the scope of this paper.
High vapor systems are tempered systems : If E.R.S. actuation allows control
of the pressure in the vessel, the temperature is also controlled because the
pressure vs temperature relationship is imposed by the vapor-liquid
equilibrium.
The control of the temperature during the vent actuation allows the control of
the reaction rate in the reaction mixture, provided that the reaction follows
the ARRHENIUS law.
These features allow the thermal runaway to be easily controlled. The vent
area required is usually less for a high vapor system than for a gassy
reaction.
314
For high vapor systems, the pressure in the vessel does not depend on the
filling ratio. This allows laboratory experiments to be performed in a closed
test cell, with a high filling ratio, giving nearly adiabatic conditions yet no risk
of cell rupture.
For high vapor systems the vent area is estimated by considering the
adiabatic heat rate at the temperature Tg related to the vent set pressure P s
by the vapor-liquid equilibrium.
The following data is required for vent sizing :
- the adiabatic heat rate curve of the reaction mixture,
- the pressure vs temperature relationship.
2) Gassy reactions
Gassy reactions are reactions in the condensed phase which produce noncondensable gases like CO, C 0 2 , N 2 , NO, NgO, 0 2 ...
The pressure in the enclosure is due to non-condensable gases only. There is
no vapor-liquid equilibrium.
For gassy reactions, the pressure in the vessel depends on the filling ratio.
The higher the filling ratio, the higher the pressure rise and the pressure rise
rate. In laboratory experiments using closed cells, the filling ratio must be
kept low to prevent cell rupture. As a consequence, the experimental
conditions are far from adiabatic.
Vent sizing for gassy reactions is based on the maximum rate of gas
generation measured under adiabatic conditions. It is safe to consider that
this rate is obtained even if the vent is actuated, and to design on this basis.
315
3) Hybrid systems
A hybrid system is a tempered system where non-condensable gases are also
produced by the runaway reaction. Neither the vent sizing equations nor the
two phase vent flow equations are well established or validated.
Due to tempering, the vent size is estimated by considering the heat rate and
the non-condensable gas generation rate at tempering conditions e.g. at the
temperature T s related to the vent set pressure Pg by the vapor-liquid
equilibrium. The discussion of the vent sizing equations for hybrid systems is
beyond the scope of this paper.
In addition the behaviour of several hybrid systems may be considered.
316
3 -
PSEUPO-ADIABATIC CALORIMETRY
3.1 Thermal Inertia
Experimental methods in calorimetry where the sample enclosed in a test
cell, is allowed to react with no heat exchange with the outside, are called
"Pseudo-Adiabatic Techniques". This is to emphasize the fact that it is the
sample and the test cell which are under adiabatic conditions.
In these techniques the heat produced by the chemical reaction causes a
temperature rise in both the sample and the test cell.
The experimental set up is usually able to prevent any heat loss from the
test cell by maintaining the temperature of the surroundings equal to the
sample temperature or to the test cell wall temperature.
The way the heat produced by the sample causes an increase in its own
temperature and that of the test cell is characterized by the Phi factor :
When a runaway reaction occurs in a chemical plant, the Phi factor is 1.05
or less.
The use of pseudo-adiabatic techniques to provide data for vent sizing
relies on the possibility of achieving a Phi factor close to 1 and/or
correcting the experimental data to true adiabatic conditions.
317
The experiments are performed by charging the cold reaction mixture in the
test cell and then by slowly raising the temperature to initiate the reaction.
When the sample exhibits self heating, the apparatus assumes adiabatic
conditions for the sample and the test cell.
The maximum heat rate is also reduced in the most sensitive equipment,
due to more reagent depletion before this point is reached.
The maximum heat rate observed in different types of apparatus must be
adjusted to the same onset temperature.
3.3 Theoretical basis
The theory by D.I. TOWNSEND and J.C. TOU [2] for the Accelerating Rate
Calorimeter (A.R.C.) holds for the other pseudo adiabatic techniques. This
theory assumes that the temperature rise observed is proportional to the
degree of conversion and that the rate constant depends on the
temperature according to the ARRHENIUS law. The theory provides a
method of determining the kinetic parameters of the ARRHENIUS law from
the experimental data e.g. the heat rate curve, dT/dt in log.scale vs T in
reciprocal scale.
The Phi factor is introduced to take into account the deviation from
adiabatic conditions due to the heat capacity of the cell.
This theoretical background provides methods to correct the heat rate curve
dT/dt vs T for the deviation from true adiabatic conditions.
The adiabatic temperature rise AT * _ 1 is obtained from the experimental
temperature rise 4 T *
- by the straightforward relation
318
4T<D
= 1
4 T ( ] ) > 1
Tf = To + 0
ATQ^
A method is proposed to correct the whole heat rate curve but this method
is not satisfactory.
It is necessary to be familiar with the theory by D.I. TOWNSEND and
J.C. TOU to have a good understanding of the correction methods, despite
the fact that few findings of this theory are used for vent sizing purpose.
319
1
where
M =
Ln 0
activation energy
Phi factor
This correction takes into account the fact that for a given onset detection
sensitivity of the experimental device, the onset temperature would have
been lower if the Phi factor was 0 = 1 .
Then the heat rate curve is corrected for the Phi factor.
The adjusted temperature T for
temperature T^| by :
T'AA
"=
T'on + 0
0=1
n \M, - T nV)
I dt/0 > 1
dT
dt_
0=1
= 0- EXP
by
/1
(-\ T 'M
dT
'A/J
dt
0> 1
.20
Correction for a different onset temperature
The reaction rate at any temperature above the onset temperature is
influenced by the onset temperature.
The onset temperature of the reaction may be determined experimentally by
a temperature scan in a pseudo-adiabatic apparatus. In this case, the onset
temperature depends on the onset detection sensitivity of the experimental
device. No exotherm can be measured below the base line heat rate on the
heat rate curve. The base line heat rate is given by the rate of temperature
scan.
The reaction rate under process conditions can be obtained from this
measured heat rate, adjusted for the difference in the onset temperature
and in Phi factor.
R
. Ln (J)
E
321
2) If a different onset temperature T^ is considered, any
temperature is adjusted for this new onset temperature :
+<
f-rrM - T
measured
This correction method seems realistic for simple one step ARRHENIUS
reactions. It allows the correction of the whole heat rate curve and does
nothing more than changing the onset temperature on the "O" order line
once corrected by the Phi factor.
If not, the gas generation rate must be adjusted using this method in the
absence of any other method.
3.4 Pressure
322
A deviation from an initial vapor pressure curve may indicate the production
of non-condensable decomposition gasses.
Alternatively polymerization of the reaction mixture will give a lower
pressure than expected from the initial pressure curve.
For gassy reactions, experiments in closed cell will often lead to cell rupture
for reasonable filling ratios. A preliminary experiment using an autoclave
will provide information on the system behaviour and allow the choice of a
filling ratio which will avoid cell rupture even if the reaction is gassy. A
preliminary examination of the reaction using DTA and the autoclave should
be carried out before any pseudo adiabatic experiment.
323
Description of the ARC
The basic principle of this apparatus involves maintaining the sample and
the test cell under adiabatic conditions once an exothermic reaction is
detected.
The calorimeter jacket and the spherical test cell are shown in figure 1. The
test cell normally contains between 1 and 10 g of liquid or solid material.
The test cell is connected to a pressure transducer (operating range 0-2500
PSIA) and suspended inside a calorimeter jacket constructed from Nickelplatted copper. The jacket contains eight heaters and three thermocouples.
A fourth thermocouple is attached to the outside of the test cell. The
operating temperature range is 0-500C. All ARC operations are controlled
by a microprocessor. The results are obtained on a line printer (raw data)
and on an X-Y plotter (graphs).
Figure 2 shows the heat-wait-search operation of the accelerating rate
calorimeter.
The sample is first weighed in the test cell to determine the value of the Phi
factor. The test cell is connected to the line to the pressure gauge inside
the jacket and the number three thermocouple is attached on the test cell
wall.
describe the
desired
At the end of the experiment the computer will process the data. The
following graphs are obtained :
324
PRESSURE
TRANSDUCER
CELL THERMOCOUPLE
THERMOSTAT JACKET
TEST CELL
FIGURE 1 :
TEMPERATURE
WAIT
PSEUDO ADIABATIC
SEARCH
TIME
FIGURE 2 :
325
TEMPERATURE
FIG. 3
-TEMPERATURE VERSUS TIME
TIME
HEAT RATE
(LOG. SCALE)
FIG. 4
HEAT-RATE VERSUS TEMPERATURE
FIG. 5
ACTIVATION ENERGY
VERSUS TEMPERATURE
TEMPERATURE
A. PRESSURE
LOG. SCALE
FIG. 6
PRESSURE VERSUS TEMPERATURE
326
dP/dt (LOG.SCALE)
FIG. 7
RATE OF PRESSURE RISE
VERSUS TEMPERATURE
TEMPERATURE
(RECIPROCAL SCALE)
dP/dt
(LOG.SCALE)
FIG. 8
PRESSURE RISE RATE
VERSUS HEAT RATE
dT/di
dj.
(LOG.SCALE)
(LOG.SCALE)
FIG. 9
PSEUDO CONSTANT k
VERSUS TEMPERATURE
5
TEMPERATURE
(RECIPROCAL SCALE)
327
dT/dt
(LOG.SCALE]
->
TEMPERATURE
(RECIPROCAL SCALE)
TEMPERATURE
(RECIPROCAL SCALE/\
>
TIME (LOG.SCALE)
328
- The observed heat rate in Log.scale versus the temperature in reciprocal
scale fig.4,
- The activation energy versus the temperature fig.5,
- The pressure in Log.scale versus the temperature in reciprocal scale fig.6,
- The rate of pressure rise in Log.scale versus the temperature in reciprocal
scale fig.7,
- The rate of pressure rise versus the heat rate both in Log .scale fig.8,
*
- The pseudo constant k in Log.scale versus the temperature in reciprocal
scale fig.9,
- The correlation of the heat rate data by an estimated heat rate curve
fig.10,
- The time to maximum rate in Log .scale versus the temperature in
reciprocal scale fig.11.
Kinetic interpretation
The activation energy is obtained from the curve in fig.5 or from the slope
of the heat rate curve in fig.4. The order of the reaction is obtained from
the following relation :
(If " T i
EA
R
M2
where
Tf
= final temperature
TQ
= onset temperature
TM
329
The estimated value of n is checked by the linearity of the plot of pseudoconstant versus reciprocal temperature in fig.9.
If the linearity is not satisfactory the value of n needs further adjustment.
Then the adjusted heat rate curve may be compared with the observed heat
rate curve as in fig.10.
Results of practical interest
The heat rate curve and the kinetic interpretation will provide the kinetic
parameters of the ARRHENIUS law k , E, n, near the onset temperature.
Due to low onset temperature and high Phi factor the heat rate curve may
show that the reaction splits in several separate steps. This is an interesting
result.
The P versus T curve will allow the system characterization
experimental temperature range.
in the
The heat rate versus pressure rise rate curve may prove that the heat
release and the gas generation are due to the same reaction.
If the reaction is a simple one step ARRHENIUS reaction the heat rate curve
and the pressure rate curve can be corrected to true adiabatic conditions.
330
Phi factor
If the sample reaction produces non-condensable decomposition gasses,
the filling ratio of the test cell must be low enough to prevent cell rupture
i.e. to keep the pressure in the operating range. The sample weight may be
as low as 1 g and the Phi factor 5 to 8 using a S.S. bomb.
If the sample reaction behaves like a high vapor system in an autoclave
experiment, the filling ratio in the ARC test cell may be high enough to
achieve a Phi factor as low as 1.2 using a titanium bomb.
Sensitivity of the onset detection
Due to the heat - wait - search procedure, the heat rate sensitivity threshold
for the exotherm onset is as low as 0.02C/mn. Depending on the sample
heat capacity, the sample mass and the Phi factor, the sensitivity is better
than 1 watt/kg sample. This is better than most DTA/DSC equipment.
Effect of Phi factor and sensitivity of onset detection
The ARC is an apparatus with a good onset detection sensitivity. The onset
temperature where the apparatus assumes a pseudo adiabatic behaviour is
very low compared to other techniques.
The Phi factor typically ranges from 2 to 5 (extreme values 1.2-9).
The Phi factor is high compared to other techniques.
These two features will cause low measured heat rates and will allow
decoupling of complex kinetics allowing the separation of complex
reactions into separate steps.
Experimental temperature range
331
The experimental temperature rise (Tf - TQ) = A T A B / 0 is reduced due to a
high Phi factor.
The experimental temperature range of the exotherm is shifted toward lower
temperatures. This may cause a subsequent exotherm not to be included in
the runaway, which would have been triggered in adiabatic conditions.
Let us take an example :
If the onset temperature of an exotherm measured in the ARC is 30C and
the experimental temperature rise 100C with 0 = 3 , the experimental
temperature range is 30C - 130C.
In a less sensitive but more truly adiabatic device, the onset temperature of
the same exotherm may be 60C and the experimental temperature rise
could be 300C with <J) ~ 1. Then the experimental temperature range is
60C - 360C.
In these two experimental temperature ranges, the reaction is not the same.
Further decomposition reactions may be triggered between 130C and
360C wich will increase the heat rate and the exotherm. In such a case, the
runaway behaviour is not satisfactorily measured with the ARC.
This heat rate curve shows all the above mentionned features
-
ARC
N"
208 - 209
I'HEHOL
'
FORMALDEHYDE
REACTION
or/ut
.0)
.M
''IAJ
-JIII
ldu- 1 -
- ^ E L M CO
333
and cannot be corrected to adiabatic conditions by the known simple
correction methods.
In conclusion the ARC is not well suitable for vent sizing because it
operates too far from adiabatic conditions and gives onset temperatures
and maximum heat rates which are too low.
The other pseudo-adiabatic techniques are more suitable for vent sizing.
On the other hand the ARC is still a valuable tool for obtaining thermal data
near the onset temperature and detecting complex reactions which may
look simple under runaway conditions.
334
The demand for the product is limited by the high level of expertise
required to operate it and interpret the results.
The same results
experiments.
may
be obtained
through
closed
Dewar
flask
Alternatively an open test cell can be used, especially for gassy reactions.
The heating device is a heating coil separated from the test cell by an
insulation assembly. The auxiliary heater is used to heat the test cell, when
in temperature scan mode, with a minimum heat rate of 0.3C/mn. The
outer guard heater is placed on the outside wall of an aluminium can
separated from the can by a uniform layer of insulation. A thermocouple is
attached to the inner surface of the aluminium can.
Once the sample heat rate is higher than the temperature scan heat rate,
the auxiliary heater is switched off but the guard heater is used to keep the
temperature of the surroundings equal to that of the test cell.
The original design of the VSP did not include a Heat - Wait - Search
routine.
The heater temperature and containment vessel pressure are controlled by a
computer.
335
THERMOCOUPLES
NITROGEN
EXHAUST/SUPPLY
PRESSURE
BY PASS
GUARD HEATER
TEST CELL
INNER HEATER
OUTER CAN
INSULATION
CONTAINMENT VESSEL
336
Test procedure
The VSP can be used to characterize runaway chemical reactions in a
closed cell as well as in an open or a vented cell. The experimental
information obtained includes Thermal data (closed cell) and flow regime or
viscosity characterization (open cell).
The pressure in the containement vessel is controlled to follow the test cell
pressure.
When the reaction is completed, the power is turned off to the guard heater
and the sample is allowed to cool slowly.
The thermal data includes :
- The temperature T^ versus time curve
- The pressures P^ and ?2 versus time curve
- The heat rate curve
- The pressure rate curve
- The pressure versus temperature curve.
The closed cell procedure is suitable for obtaining thermal data for high
vapor (tempered) systems where the pressure is due to a vapor-liquid
equilibrium.
337
For gassy reactions or hybrid systems the closed cell bursts readily due to
the high filling ratio. This exemplifies the need for a preliminary test using
an autoclave to detect the production of non-condensable gasses. Only the
high vapor systems should be used in a VSP closed cell test.
For testing gassy and hybrid systems in the VSP the "constant volume
mode" is best used where the gas production rate is obtained from the
pressure build up in the containment vessel using a open top cell.
To demonstrate the use of the VSP in obtaining thermal data, the heat rate
curve of the phenol+formaldehyde runaway reaction with the recipe of ref.5
is given fig.14. Note the high onset temperature and the high maximum
heat rate obtained in the VSP experiment.
Flow regime testing
Flow regime characterization is obtained from blow-down tests using a top
vented test cell.
For the current test cell design, the vent line diameter of 2.5 mm leads to
linear flow velocities of 30 - 60 cm/s. Normally 40 % to 60 % of the liquid
will be left behind following blowdown transient if non-foamy behaviour
prevails.
If a foamy regime prevails, no liquid is left behind in the test cell.
The test procedure is the following : (see fig.15)
- The containment vessel pressure is set to the value Pg.
- The runaway reaction is initiated and allowed to reach tempering at Pg.
- The containment vessel pressure is decreased to atmospheric, thereby
initiating the blowdown process.
- When the test cell pressure is equal to atmospheric, the containment
vessel is immediately repressurized to prevent further mass loss from the
test cell. The sample mass left behind in the test cell is determined by
weighing it afterwards.
338
I I I 1 II I I I 1 I 1 I I I I I 1 !
1 1 1 1 1 1 Ll 1
iiiiiiiiiiiiiiliiiiiii'inniiiiiiiiiiiui'
rf**
.....
1
1
]
| .
1
\ jgr
..... . _ _ _ .4?'
cF
en
_43T
rff
'
: :
*r
~f
&k ~B
ifi
',
a
3
!
i
1
" I I
I ' l l '
6(]
1
0
3
r; ; iV-Z-t~:';E?"
'~~"Jr i
*f \
-am
3
sr
li.
J_LLi!.!J.L!.
8()
I
1 M 1 I
100
1 I I ! ! 1 1 1 11 M I i i m i l
120
HO
HI
llll
170
M l i r 11 M M M 1 r
200
(C)
Fig. 14
339
,
?< , te
TEMPERING
Ts
REPRESSURIZAT1
l\
'W
\ \
BLOW DOWN
TIME
BLOW DOWN
TIME
FIGURE 16 : BLOW DOWN TEST USING BOTTOM VENTING FOR VISCOSITY TESTING
340
Viscosity testing
The VSP experiment using a bottom vented test cell and the blowdown
technique is used to characterize the flow regime (turbulent or viscous). A
vent line with L/D ~ 30 and a line length = 100 mm is used to assure
equilibrium flashing flow conditions.
The test procedure is the following : (see fig.16).
dT Y
Cp L
If the experimental flow rate is less than the ERM value the flow is viscous.
341
Drawbacks of the VSP
Open cell testing appears to be expensive because of damage to the
heating device.
The heat rate base line corresponds to an onset detection sensitivity of
0.3C/mn. This is a poor sensitivity compared with the ARC. Constant
temperature exposures are difficult to achieve because of the temperature
drift of the VSP.
The pressure follow-up in the containment vessel could be better.
VSP clones
Once the basic design of the VSP is understood the performance of the
apparatus can be easily improved by the users. It would be unfair to present
these improvements as a new concept which could justify a trade
competition.
during
constant
temperature
342
- Pressure correction for the Nitrogen pad.
by providing a solid
343
HEAT RATE
10"
103
10 2
IO1
10" !=
r.
'^rrpts^SfrrfK.-f.,
-rrp*-f***n:'ln" '
lo-'b
-1000/T(K)
10-2
-3
.4
-2.8
(. _|
-2.6
1.(.
100
-2.4 .
-2.2
-2
-1.8
-1.6
14||(Jk_JII|l_J__+_4;----l--t-4-l-4-4-+
150
200
250
300
350
F i g u r e 17
Pressure = f(T)
103
^
10 2
, . . /
\
10'
-looomK)
inn
1
-2.6
-2.4
-2.2
-2
-1.8
-1.6
\.+- 4J....4-444 441~4~-1l~4-J-4-4-4-4-4--l"M-4-4-4-4
250
300
350
150
200
100
Figure 18
344
DP/Dt = r m
10 J
10 3
_
,*'
102
.-"""
10' h
.3
"
10"
1 ' " " '
..-y^f
10-'
"
io-2
10-3
-1000/T(K)
10 J
-2.8
-2.6
4-+[;I
100
-2.4
-2.2
-2
-1.8
-1.6
Ji!JI1--iI1kI14-44-4^-1-4-4-4-4-;150
200
250
300
350
Figure 19
345
4.3 The closed Dewar experiment
The closed Dewar experiment was formerly used to determine reaction
onset temperature and adiabatic induction time. For this reason it is a widespread testing method.
The use of closed Dewar experiment to provide the experimental data for
vent sizing has been emphasized more recently by R. ROGERS (I.C.I.) (7)
and the author (Rhone-Poulenc) (5). This is made possible by the availability
of low Phi factor stainless steel Dewar flasks achieving a Phi factor of 1.1.
In continental Europe the best Dewar is made from the domestic items sold
in the supermarkets by Camping Gaz. The same items are also sold in
England by Thermos. These Dewar flasks are manufactured in Japan and
Korea. The price for 1 I flask is 300 F.F. The plastic cap on the original item
is replaced by welding a thread to hold a cap equipped with
thermocouples, a pressure gauge, a rupture disk and a blow-down valve.
The Dewar flask is heated in a furnace the temperature of which is
controlled so as to be equal to the sample temperature, thus eliminating the
heat losses which are essentially from the cap.
The whole assembly, Dewar and furnace can be shaken if necessary and is
shown in fig.20.
346
The pressure is corrected for the Nitrogen pad.
The outputs include :
-
1
T
dT\
dt
To illustrate the use of the closed Dewar to provide the experimental data
for vent sizing, the heat rate curve and the pressure vs temperature curve
obtained for the Phenol + Formaldehyde runaway reaction with the recipe of
Ref.5 are given figs.21 and 22.
347
SAFETY VE1T
BLOW DOWN
VALVE
V O ^ * SAMPLE
PRESSURE GAUGE ^ T E ^ T U R
FURNACE
[ITEHRATURE
.-. .1
348
Sul9t
FicniiP :
PHENGL /FOnHOL
a: 2 9 0 6 9 0
l a l Mo 1
Teagoratura
loC )
( Eaeal au 2B/6/90 )
Log P - . 2 3 9 3 3 - 2 0 9 1 . 2 8 / (T+2271
da a a c o a o o a l t l o n : 38
;y
1
1
1
1
I /
/
!/
[
I
-t
<
\ ' A
i
,/
1
1
. V I I
>
i
i
/i i i
/ i !
1
1
!
,
>
'
'
1
i
/[
/ /
Sutat
:
Flcnur :
Temoetature
loci
PHENOL /FORMOL
a: 2SG690
( Eaaal mj 2A/B/90 1
349
consequences of the uncontrolled runaway reaction.
In some cases where very fast reactions are obtained with moderate final
pressure, the Dewar flask is the best choice because the VSP closed cell
ruptures during the fast pressure build up. A typical example of these fast
reactions is the bulk polymerization of vinyl acetate initiated by a peroxide.
The test cell holds a single heating element that compensates for heat
losses and allows a temperature scan in the test cell.
350
FILL
_r
.THERMOCOUPLE
HEATER
=3
PRESSURE
GAUGE
I PSV
NITROGEN
SUPPLY
TEST CELL
-INSULATION
ASSEMBLY
CONTAINMENT.
VESSEL
351
The heater is controlled by the sample temperature measurement, to
overcome the heat losses and produce a specified temperature rise rate at
that temperature. Nevertheless the temperature control is not by a feed
back loop but by a pre-programmed heating.
For a reactive system the linear heat up rate is added to the reaction energy
release so that the heat rate is measured under external heating conditions.
The manufacturer claims that heat rates as low as 0.1C/mn can be
obtained.
Test procedure
The heat rate data is obtained by setting the RSST containment pressure to
some Maximum Allowable Pressure and letting the reaction be initiated by a
temperature scan. Then the heat rate is measured under conditions close to
the adiabatic or under external heating conditions.
When the boiling point of the sample under the test pressure is reached,
the vaporization heat sink compensates for the heat rate this allowing the
detection of the boiling point.
Gassy reactions
352
Hybrid systems
A hybrid system is characterized by repeated tests under different back
pressures. If the heat rate and the gas generation rate are influenced by the
preset back pressure, the system is hybrid in nature.
The author is not familiar with the RSST and so these comments are based
on the experience of other workers.
The results of the Round Robin-test on 25 % Hydrogen Peroxide solutions
show a good agreement between heat rates obtained in RSST and VSP
experiments.
It is nevertheless likely that as with the VSP, a low heat rate base line
cannot be obtained with the RSST.
The heat rate data obtained with the RSST are influenced by :
- The heat rate of the temperature scan,
- The insulation packing,
- The position of the heating resistance with respect to the thermocouple in
the test cell,
- Any detail which may influence the heat exchange between the test cell
and the outside.
Given that adiabatic conditions are difficult to produce and that so many
details may influence the heat rate and gas generation rate data, the results
obtained from RSST experiments look surprisingly good.
In a well equipped process safety lab the VSP is the better choice. The
RSST could be purchased to take advantage of its possible operation on
plant sites.
353
CONCLUSION
This paper demonstrates the need for experimental techniques
operating conditions close to the adiabatic for vent sizing purpose.
achieving
Due to conditions far from the adiabatic, the ARC should not be used to obtain
data for vent sizing.
A good experimental device should achieve :
-
measure
the
ultimate
consequences
of
The VSP can satisfy most of these requirements, once improved by the
customer.
The close Dewar experiment is also a valuable tool but with a pressure range
restricted by the pressure resistance of the Dewar flask.
The RSST may be useful as a screening tool and is of interest for measurements
on plant sites. The VSP is a more useful device than the RSST.
One should never rely on one technique to produce data for vent sizing. Tests
using at least two or preferably three techniques, including the ARC, provide the
opportunity to detect unexpected reaction changes as well as gelification
processes which would compromize the reliability of the ERS design.
354
The sample must go through DTA and autoclave tests before using pseudoadiabatic techniques to detect extremely fast reactions and to adjust the
experimental procedure taking into account these basic results.
Hopefully the reader has been convinced of the great utility of pseudo-adiabatic
techniques in the process safety laboratories.
LITERATU RE
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
TREATMENT
OF
RELIEVED
FLUIDS
INTRODUCTION
A. Benuzii and J. M. Zaldivar (eds.). Safety of Chemical Batch Reactors and Storage Tanks, 355-370.
1991 ECSC, EEC. EAEC, Brussels and Luxembourg. Printed in the Netherlands.
356
2.
357
3.
VAPOUR DISPOSAL
358
Most flares are able to convert close to 99% of hydrocarbons
but other chemicals (e.g. HCN) may be only poorly treated. In
such instances, incinarators may be used. These subject the
gases to a more controlled temperature history and give a
minimum residence time necessary to achieve conversion. Also,
catalysts are frequently employed to deal with species that
are difficult to treat by heat alone.
4.
DESIGN CONSIDERATIONS
4.1
Knock-out-drum
V* k[(pL
with
m/s
TtD2/4
Quench Tank
<Tr ~ Tf) + Mz X X
Cq(TfT0)
'
359
assumes that the amount of non-condensible gas is negligible
which is usually acceptable for design purposes.
As stated earlier, Tf should be at least 10C lower than the
expected condensation temperature of the reactant vapours.
The vapour/gas leaving the quench tank will now be determined
by:
. non-condensible gas from the reactor
. uncondensed vapour
. further continued reaction in the quench drum
The tank volume above the liquid will act to separate this
gas/vapour from entrained liquid. If the flow rate of this
gas/vapor is known, then the top section of the drum can be
sized as a knock-out drum using equation (1).
If it can be shown that the reaction does not produce
non-condensible gas and if the reaction is suppressed after
quenching, then it is possible to totally contain the
incident.
4.3
r<Vr>
(3)
360
Equation (3) is applicable if the reactor is vented at a
temperature above its atmospheric boiling point (i.e. Tr > Tb) .
If the reverse is true the liquid will remain at the reactor
temperature Tr (no flash cooling) and the gas will be that
vented from the reactor plus further reaction in the drum.
Also, in such cases the reaction rate will be exactly the same
as in the reactor since the reactants are not cooled.
For the situation where the reactants flash down to a lower
temperature, the reaction rate will be quite different in the
knock-out drum, normally much lower corresponding to the
reduction in temperature. If the activation energy E for the
reaction is known, then the self-heat rate in the drum (dT/dt)d
may be obtained from that in the reactor (dT/dt)r using:
i-m,'
fl
El 1
R[ Tz
-HSrIBtf
(4)
Tb
(1 - x)
(dT/dt)d
(5)
in
in
to
of
(typical
for
361
decomposition r e a c t i o n s ) , then clearly this gas will need to
be relieved from the quench drum.
The rate of flow will be
identical to that from the reactor plus further gas generated
within the quench drum if the liquid continues to react. The
rate of reaction will be determined by the extent of cooling
and dilution by the quench fluid.
5.
5.1.
TESTING OP REACTOR/QUENCH
Runaway Reaction
COMBINATION
Simulation
362
15C/minute, PHI-TEC II can reach up to about 200C/minute.
5.2.
363
If a decomposition reaction is being studied this will be
clear from the results and the rate of gas generation can be
measured.
6.
6.1.
Quench Test
The temperature-time data for the reaction cell and the quench
cell is shown in figure 7. The vent valve was opened after
about 13.8 minutes into the exotherm - this is clearly
indicated - and then closed a few seconds later. When the
vent was opened, note the rapid initial rise in temperature
within the quench cell as it suddenly warms up from ambient.
364
When the vent valve was shut, the contents in the reaction
cell continued to self-heat, as did the quench cell contents.
The initial information derived from this test is:
*
= 21Exp
*1
1 7
-lxl
1.987
(_i-\ - (-1-)
\375/ 1345J
(6)
= 2 .9C/minute.
This is compared with the experimental value of about
7C/minute. The difference is at least partially due to the
fact that when the reactants were flashed in the quench cell,
lighter components will be preferentially stripped. Thus the
concentration will be quite different and extremely difficult
to calculate.
6.3.
365
could for example be directly connected to the flare.
6.4.
- QL)
(7)
7.
CONCLUSIONS
366
FROM REACTOR
FROM REACTOR
CLOSED VENT
_Jt
QUENCH LIQUID
367
Solenoid
Valves
n
Nitrogen
1. Thro R o d l a n l H o s i e r s
1. S i m p l o T h e r m o c o u p l e
3. M a g n e t i c Bar
A . To o I C e l l
5. U M a g n a !
0. V e s a u l Pressure
7. B ample P r e a t u i t
transduoor
Tranaducor
HEATING COIL
368
UJ
en
in
tn
en
40
60
50
70
TIME[min]
80.0
t-
~r
60.0
40.0
.v_i
Li_
20.0
0.0
25.0
55.0
*s-
85.0
115.0
TEMP(C)
145.0
L _
175.0
369
-.
130.0
8.0
12.0
TIME (min)
UJ
20.0
370
0.6
0.6
2
fe
0.4
0.2
0.0
'
0.5
NO QUENCH
1.5
'
2.0
2.5
3.0
3.5
Mq/Mr
REFERENCES
1.
2.
3.
4.
J Singh, PHI-TEC: Enhanced Vent Sizing Calorimeter Application and Comparison with Existing Devices,
International Symposium on Runaway Reactions, Boston,
March 7-9, 1989, AIChE
5.
T. J. SNEE
Health and Safety Executive
Buxton
Derbyshire SK17 9JN
UK
1. INTRODUCTION
A wide range theoretical and experimental techniques can be used to evaluate hazards
associated with runaway exothermic reactions. This paper illustrates how some of these
techniques are applied to a batch chemical reactor. A reaction system with well established
physical and chemical properties has been chosen in order to demonstrate how thermo
analytical data can be used to determine safe operating procedures. The methods discussed in
the present work would form part of a formal hazard evaluation procedure.
372
NOTATION
symbol
quantity
units
to
tf
m
me
ms
T
Tm
Te
T0
time
onset time
final time
mass
container mass
sample mass
temperature
reactant temperature
jacket temperature
onset temperature
s
s
s
kg
kg
kg
K
K
K
K
AT
heat of reaction
activation energy
pre-exponential factor
reaction rate constant
order of reaction
conversion
rate of heat generation
rate of heat loss
surface area
heat transfer coefficient
external heat transfer coeff.
internal heat transfer coeff.
thermal resistance of wall
specific heat capacity
specific heat of sample
specific heat of container
kJkg" 1
kj mol" 1
s-1
s-1
kJs" 1
kJs" 1
m^
W m - 2 K-l
W i n " 2 K-l
W m " 2 K-l
m 2 W- ; ' K
kJkg-1 K-l
kJ kg-1 K-l
kJ kg"1 K-l
-lif-1
kJmol-tK
ad
AH
E
A
k
n
X
^g
qi
s
u
hi
h0
Rw
^
V
S
4>
Vc
R
373
2. REACTION SYSTEM
In order to illustrate the principles discussed in this paper it was necessary to choose a
reaction system with the following characteristics:
* a simple reaction system which produces good thermo-analytical data providing reliable
source terms for the theoretical models.
* a reaction system where reliable data on the physical properties of the reagents and
products are available.
* a reaction system in which the rate of heat generation could be controlled by the addition
of small quantities of catalyst to allow a systematic study of the conditions which can lead to
exothermic runaway.
A simple esterification reaction was chosen to meet these requirements so that the salient
features of exothermic runaway can be explored:
CH 3 CH(OH)CH 2 CH 3 + (CH3CH2CCO2O -> C 2 H 5 C02(CH3)C2H5 + C 2 H 5 C0 2 H
sec.butyl alcohol
propionic anhydride
propionic acid
This moderately exothermic reaction can be catalysed by the addition of small quantities of
sulphuric acid.
3. HAZARD EVALUATION
Systematic determination of the risk and possible consequences of exothermic runaway should
ensure that safe and cost effective measures for controlling the hazard or mitigating the
consequences can be established. A typical theoretical and experimental assessment would pass
through the following stages:
* Determination of the exothermicity of the desired reaction and of any undesired side reactions. (If
the total potential energy release is very low, the risk of exothermic runaway can be discounted, and
further evaluation is unnecessary.)
* Estimation of the maximum temperature and pressure that could be achieved if the heat from
exothermic reaction were released very quickly (with no heat loss). In this way it is possible to
assess whether, under "worst case" conditions, there is a risk of vessel rupture, release of toxic
reagents or products, secondary reaction, auto-ignition or detonation etc. (If maximum temperatures
and pressures are relatively low, it may be possible to accept the risk of runaway, while ensuring
that the specifications of the reaction vessel are sufficient to achieve total containment.)
374
* Determination of the temperature dependence of the rate of heat generation and the rate of heat
loss so that the critical conditions which can lead to exothermic runaway can be predicted in order
that suitable control measures can be identified.
* Assessment of the reliability of control measures and the hazard assessment procedure. (If there
remains a significant risk of loss of control, it is necessary to adopt additional measures to mitigate
the consequences of exothermic runaway. These additional measures could include the provision of
an emergency pressure relief system, scrubbing systems, additional containment or plant isolation.)
The information required for the hazard evaluation can be divided into four main areas:
1) Physical properties of reagents and products - usually from published data or simple
measurements.
2) Specifications for process vessels etc. - from data supplied by manufacturers.
3) Reaction exothermicity and kinetic constants.
4) Heat transfer characteristics of reactor.
Physical properties, vessel specifications and heat transfer data are usually readily obtainable.
Definition of the thermo-kinetic properties can be more problematical. The present discussion will
place particular emphasis on the the determination of the thermo-kinetic parameters of an exothermic
reaction from thermo-analytical data.
As a simple example of a hazard evaluation procedure, a proposal to carry out the esterification
reaction in a 250 1 glass-lined reactor will be examined.
4 .PHYSICAL PROPERTIES
Relevant physical properties, and flammability and toxicity data for the reagents and products of the
esterification reaction are listed in Table 1. This type of information is readily obtainable for simple
chemicals. However, in the case of more complex reactions it may be necessary to determine some
of these quantities experimentaly. Uncertainties in, for example, the toxicity data for complex
substances, may necessitate additional safety measures which are not appropriate for simple
substances with well established properties.
375
TABLE 1. Physical properties and flammability and toxicity data for reagents and products of
esterification reaction.
sec.butyl
alcohol
propionic
anhydride
sec.butyl
propionate
propionic
acid
boiling point
(C)
99.5
169
132
141
specific heat
(kJkgK" 1 )
2.848
1.796
1.944
2.119
latent heat
(kJ kg"1)
544
306
283
431
flash point
(Q
24
74
(open test)
32
(n-butyl)
54
long term
expos, limit
(mg m -3 )
300
20
(acet. anh.)
950
(s.butyl acet.)
30
short term
expos, limit
(mg. m" 3 )
450
20
(acet. anh.)
1190
(s.butyl acet.)
45
5. PLANT SPECIFICATIONS
The present discussion will consider a proposal to perform the esterification as a pilot-scale (250 1)
batch reaction. The specifications for a standard, industrial, glass-lined, jacketed reactor are listed in
Table 2. Specifications for auxiliary equipment such as pumps, feed vessels and pipework, and the
standard operating and handling procedures would form part of a general safety assessment.
Hazard evaluation procedures concerned specifically with runaway reactions would focus particular
attention on the temperature and pressure rating of the reactor and possible catalytic effects from the
materials of construction.
It will be assumed, for the purposes of the present discussion, that the temperature of the heat
transfer fluid in the reactor jacket can be chosen according to the results of the hazard evaluation
and the requirement for efficient and economic chemical production.
376
jacket
working pressure
(barg)
6to-l
6to-l
design pressure
(bar g)
6.6 to-1
6.6 to -1
test pressure
(barg)
11.4
11.4
temperature range
(Q
-25 to 200
capacity
(litres)
250 (nominal)
334 (total)
93
material
mild steel
6. THERMAL ANALYSIS
Four types of apparatus have been chosen to illustrate how thermal analysis can be used to
determine the thermo-kineteic parameters of an exothermic reaction:
Differential Scanning Calorimetry (DSC)
Accelerating Rate Calorimetry (ARC)
Adiabatic Calorimetry with Vent Sizing Capability (VSP, PHI-TEC)
Heat Flow Calorimetry (RC1)
6.1 Differential Scanning Calorimetry [1]
The DSC sample (around 10 mg) is held in a sealed container and placed in a measurement cell
whose temperature is increased at a constant rate. The rate of heat evolution (or absorption) from
the sample is measured as a function of temperature with respect to an inert reference material.
6.2 Accelerating Rate Calorimetry [2]
The sample for ARC analysis (around 10c) is placed in a container (designed to withstand
substantial pressures) which is subject to stepwise heating, but is held in an adiabatic state between
each temperature step. Adiabaticity is achieved by matching the temperature of the surroundings to
377
that of the sample container. Exothermic reaction or decomposition is detected when, after a
temperature step, the temperature of the system continues to rise due to self-heating of the sample.
If the rate of temperature rise exceeds the detection threshold (0.02 K min"l), stepwise heating is
suspended and an adiabatic environment is maintained as temperature and pressure are recorded as
the reaction proceeds.
6.3 Adiabatic Calorimetry and Vent-Sizing
The operation of the VSP [3] and PHI-TEC [4] is similar to ARC, but thin-walled sample
containers are used so that that the experimental data is not significantly influenced by the thermal
mass of the container. Rupture of the sample container is prevented by applying external pressure to
match the pressure developed inside the container. These instruments are provided with special
facilities for the design of emergency pressure relief systems for chemical reactors.
6.4 Heat Flow Calorimetry [5]
The rate of heat generation is measured using heat-flow reaction calorimmetry by determining, in a
small jacketed reactor (capacity approx. 2 1), the rate of heat transfer from the reactants to the
surrounding heat transfer fluid. The rate of heat flow is proportional to the difference between the
reactor and jacket temperature. This difference can be related to rates of heat generation, after the
heat transfer characteristics of the vessel have been determined by calibration using electrical
heating.
7 REACTION EXOTHERMICITY
A range of thermo-analytical techniques can be used to determine the heat of reaction. The choice of
technique depends on factors such as: available sample size, required temperature range, maximum
temperature and pressure etc.
7.1 Heat of reaction from DSC
The DSC scan for the esterification reaction is shown in Figure 1. The reaction is first detected at a
temperature of 40C (the onset temperature) when the rate of heat evolution is sufficient to cause a
discernable departure from the baseline. As the temperature is increased further, the rate of heat
generation reaches a maximum value at 75C and then subsides as the reagents become depleted
until the trace returns to the baseline at 120C.
The heat of reaction can be determined by integrating the rate of heat generation between the onset
temperature and the final temperature.
1
A H = - Jqg.dt
to
This integral is given by the area between the DSC curve and the baseline.
(1)
378
CO
0>
64
72
80
88
96
112
120
Temperature 'C
Figure 1. DSC scan for the reaction between propionic anhydride and sec.butyl alcohol catalysed
by 0.8% H2SO4 (scan rate 8 K min"1).
n
+ PH-TECdota
DP-J
^^H+HH+^^Hf.,.
x ARC data
+
+
+
+
"E
1)-^
10"
-1
300 312
324
336
348
360
372
384
396
408
420
Temperature (K)
Figure 2. ARC and PHI-TEC plots of self-heat rate against temperature for the reaction between
propionic anhydride and sec.butyl alcohol (no added H2SO4).
379
7.2 Heat of reaction from ARC
The ARC plot of self-heat rate against temperature for the esterification reaction is shown in Figure
2. The reaction is detected at 310 K when the rate of self-heating exceeds the detection threshold.
Under adiabatic conditions, the rate of temperature rise increases to a maximum and then decreases
until a final temperature of 392 K is reached when the reaction is complete. The total heat evolved
is distributed between the sample and the sample container and the heat of reaction can be calculated
using the expression:
AH = <j).ATad.Cp
(2 )
where
ms.Cps + mc.Cpc
*=
EIT^I
<3>
A typical thermal dilution factor (<j)) for an ARC experiment would be 1.4.
7.3 Heat of reaction from PHI-TEC
The PHI-TEC plot of self-heat rate against temperature for the esterification reaction is shown in
Figure 2. The detection sensitivity and the form of the self-heat rate plot is similar to ARC data.
However, the lower thermal dilution factor for the PHI-TEC gives rise to higher self-heat rates and
a larger adiabatic temperature rise. The heat of reaction can be evaluated using Equation 2. PHITEC measurements can be made with thermal dilution factors close to unity.
7.4 Heat of reaction from RC1
Temperature records for the esterification reaction in the RC1 are shown in Figure 3. The reaction
was performed as a semi-batch process (metered addition of the anhydride to the alcohol) with the
reactor temperature maintained at 75C. At the start of the addition, the jacket temperature rises
above the reactor temperature in order to compensate for the addition of cold fluid to the hot reactor.
After 60 minutes the addition is complete and the jacket temperature falls below the reactor
temperature to compensate for the heat generated as the exothermic reaction proceeds.
The heat of reaction is determined by integrating the difference between temperature of the reactor
and the jacket multiplied by the heat transfer coefficient with corrections for the heat required raise
the temperature of the anhydride to 75C.
AH =
m JU.S.(Tm-Te).dt
(4)
to
At a reactor temperature of 75C, heat losses to the surroundings through the lid of the vessel are
substantial, and this introduces large uncertainties in the calculated heat of reaction, when the rates
of heat generation are relatively low. At 75C, with no catalyst, the rate of esterification, and
therefore the difference in temperature between the reactor and jacket, is relatively low and the
380
reactor temperature
jacket temperature
343
340
2800 5600 8400 11200 14000 16800 19600 22400 25200 28000
Time (s)
Figure 3. Temperature records from semi-batch esterifictation in RC1 reaction calorimeter (no
added H2SO4).
381
hazardous but not catastrophic. The preliminary hazard assessment indicates that reaction kinetic
and heat transfer data should be obtained in order to identify the critical conditions which could lead
to exothermic runaway, so that suitable control measures can be adopted.
heat of reaction
(kJ kg"1)
-261
-281
-301
reaction calorimetry
-306
9. REACTION KINETICS
The rate equation for a simple chemical reaction or decomposition can be written in the form:
-k.(l-x)*
(5)
The temperature dependence of the rate constant (k) can often be predicted using the Arrhenius
equation:
k = A.exp(-E/RT)
(6)
Under adiabatic conditions ,with constant Cp, reactant conversion is directly proportional to the
change in temperature
x=
T-T0
AT
ad
(7)
382
hence the rate of self-heating under adiabatic conditionsis given by:
dT dx AH
1W
dt = W
<* Cp.<j>
AH.A.(l-x) n
.exp(-E/RT)
Cp.<j)
(8)
at low conversion (x 1)
log(dT/dt) = log(AH.A/Cp.<t>) - E/RT
(8a)
The critical conditions which can lead to exothermic runaway are normally reached before reactant
depletion causes a significant reduction in the rate of heat generation. This means that, for the
purpose of hazard evaluation, the kinetics of exothermic reaction can be specified by determining
the Arrhenius constants A and E using thermal analysis. The addition of small quantities of
sulphuric acid to the esterification reaction mixture shows how the kinetic parameters can be
strongly influenced by catalytic effects.
383
0.KH2S04
02%H2S04
o
0.4%H2S(H
0.S5H2S04
390 400
40
Temperature (K)
Figure 4. DSC scan for the reaction between propionic anhydride and sec.butyl alcohol catalysed
by various concentrations of sulphuric acid.
iu :
uU
*
10^
2*:
0.4%H2S04
10-1
in" 2
HJ
0.KH2SO4"
."' /
o
o
c^"
")
o^
oo^
t <S>
0O
o-""
o
o
a =
0ZUQSM,
0% H2S04
1
-3.50 -3.40 -3.30 -3.20 -3.10 -3.00 -2.90 -2.80 -2.70 -2.60 -2.50
-1000/T (K-1)
Figure 5. ARC plots of log(self-heat rate) against reciprocal temperature for the esterification
reaction catalysed by various concentrations of sulphuric acid.
384
Reliable ARC data could not be obtained at sulphuric acid concentrations greater than 0.4 %
because, with a reactive compositions, significant reaction occurs during the time required to install
the sample in the instrument. This effect causes a reduction in the measured adiabatic temperature
rise. (The addition of sulphuric acid would not be expected to affect the heat of reaction or the
temperature rise that occurs if adiabatic conditions are established at the start of the reaction.)
9.3 Reaction kinetics from PHI-TEC
The PHI-TEC allows reagents to be introduced into the sample container after it has been installed
in the apparatus. Reactive compositions can be studied by operating in adiabatic mode as soon as
the sample has been assembled. Figure 6 shows PHI-TEC data for the esterification reaction
catalysed by 0.8 % H2SO4. The difference between initial and final temperature (110 K)
coresponds closely to the temperature rise measured for less reactive compositions (Figure 2). This
indicates that adiabatic conditions have been maintained from the start of the reaction and during the
course of the reaction when the rate of temperature rise reaches a maximum value of 500 K min" 1.
9.4 Reaction kinetics from RC1
Figure 7. shows the experimental record for batch esterification reaction (catalysed by 0.8%
H2SO4) in the RC1 with the reactor temperature maintained at approximately 31C. Equation 5
predicts that, at constant temperature, the maximum rate will occur at the start of the reaction (x =
0). Figure 7 indicates a maximum rate of heat generation only after substantial proportion of
reagents have reacted (x = 0.5), indicating auto-catalytic kinetics. Auto-accelerating reactions can
result in supercritical (runaway) conditions at temperatures lower than those predicted assuming
normal kinetics. Detailed discussion and quantification of this effect will be discussed elsewhere.
Figurre 7 shows how reaction calorimetry can be used to determine the concentration dependence of
reaction rate. Reaction calorimetry can be used to deterrmine the temperature dependence over only
a limited temperture range due to the substantial temperature dependence of the heat loss
characteristics of this type of apparatus and problems associated with the detection of rapidly
changing temperatures. The adiabatic techniques (ARC and PHI-TEC) are, in general, more
suitable for determining the Arrhenius parameters of an exothermic reaction.
9.5 Kinetic parameters for the esterification reaction.
The initial sections of the ARC and PHI-TEC plots of ln(heat rate) against reciprocal absolute
temperature (Figures 5 and 6) are approximately linear, indicating that the temperature dependence
of the reaction rate can be predicted by the Arrhenius equation. Values for the activation energy and
pre-exponential factor evaluated for the esterification reaction are listed in Table 4
385
10 3 ^
10 2 ^
-2
10
0XH2SO+
0.025* H2S04
3.50 -3.38 -3.26 -3.14 -3.02 -2.90 -2.78 -2.66 -2.54 -2.42 -2.30
-1000/r (K-1)
Figure 6. PHI-TEC plots of log(self-heat rate) against reciprocal temperature for the esterification
reaction catalysed by various concentrations of sulphuric acid.
40
37-1
34
reactor temperature
31
28
25-|
22
E
jacket temperature
19-1
16
13
X)
850
Figure 7. Temperature records during esterification reaction catalysed by 0.8% H2SO4 carried out
as a batch process in the RC1 reaction calorimeter.
386
TABLE 4 . Thermo-kinetic parameters evaluated from ARC and PHI-TEC data for the esterification
reaction.
sulpihuric
acid cone.
(%)
activation energy
(kJ mol"1)
pre-exponential factor
(s-1)
ARC
PHI-TEC
ARC
PHI-TEC
0.1
94
95
3.6*10 n
4.6*10n
0.2
97
94
1.8*1012
6.6*10 n
0.4
95
94
1.8*1012
7.9*10 n
0.8
98
7.5*10 12
(9)
where the heat transfer coefficient (U) comprises contributions from the thermal resistance of the
reactor wall and the internal and external film heat transfer coefficients :
J_ _1_
U~h0
w +
_J_
lu
(10)
Techniques for determining U are discussed elsewhere [6]. A value of U= 180 W m"2 K~l would
be typical for an organic medium in a standard industrial glass-lined reactor.
387
11 CRITICAL CONDITIONS
Conditions in a well-stirred jacketed reactor correspond closely to those assumed in the Semenov
model [7] for predicting the critical conditions for exothermic runaway. Under these conditions
(assuming Arrhenius-type temperature dependence for the rate of heat generation) the critical jacket
temperature is given by:
Vc=
m.AH.A.E.exp(-E/RTe)
UZRTT
i
e
(u>
This expression assumes that the critical conditions are reached before the depletion of reagents has
had any significant influence on the rate of heat generation (pseudo-zero order kinetics).
The Semenov model has been used to predict the critical conditions for the esterification reaction in
a 250 1 glass-lined reactor. Critical jacket temperatures calculated using ARC and PHI-TEC data are
listed in Table 5.
TABLE 5. Critical jacket temperatures calculated for the esterification reaction in 250 1 glass-lined
reactor.
sulphuric acid
concentration (%)
PHI-TEC data
0.1
296.0
295.8
0.4
285.1
288.9
0.8
285.6
It is not normally possible to test this kind of prediction experimentally, because of the attendant
hazards. However, in the case of the esterification reaction, where the properties of reagents and
products are well established, experiments to determine the critical conditions which could lead to
exothermic runaway could be performed safely in an isolated 250 1 reactor. The temperature-time
profiles which were observed when the esterification was performed as a batch reaction with water
circulating at 285 K (12 C) through the reactor jacket are shown in Figure 8.
388
OJ
68-
51I
34-
0.8%H2S(H
CL.
0.4% H2S04
^
- \
E
CL>
17-
00
0.1% H2S04
4800 9600 14400 19200 24000 28800 33600 38400 43200 48000
Time (s)
Figure 8. Temperature records during esterification reaction (catalysed by various concentrions of
H2S04) as a batch process in a standard industrial , pilot scale reactor (250 1 capacity) with the
jacket temperature maintained at approximately 295 K.
The drop in temperature which can be seen at the start of of each of the experimental records
(Figure 8) is due to endothermic mixing of reagents.
The results for 0.1% H2SO4 show that, after the initial mixing of reagents,the temperature of the
reactor contents remain at approximately the same temperature as the water circulating through the
jacket. (Analysis of the contents of the reactor after 48 hours indicated that the reaction had
proceeded isothermally to completion.) At a sulphuric acid concentration of 0.4% exothermic
reaction causes the temperature of reaction mixture to rise above the jacket temperature but the rate
of cooling is sufficient to prevent a large increase in the temperature of the reactor contents. At a
sulphuric acid concentration of 0.8% the rate of cooling cannot prevent an acceleration in the rate of
heat generation and a runaway excothermic reaction was observed.
The experimental data indicate that for the 250 1 reactor at a jacket temperature of 295 K the critical
acid concentration which can lead to a runaway reaction lies between 0.4% and 0.8%. This result is
broadly in accord with the prediction of the simple Semenov criterion (Equation 11) using thermoanalytical data from ARC or PHI-TEC (Table 5 ) .
Autocatalysis detected using isothermal reaction calorimetry (Figure 7) would lead to critical jacket
temperatures lower than those predicted in Table 5. Experiments over a wider range of acid
concentrations and jacket temperatures than those reported here would be necessary in order to
quantify this effect.
389
12 CONCLUSION
The application of a simple theroretical model can provide a reasonably reliable prediction of the
critical conditions which can lead to exothermic runaway in a batch reactor. More sophisticated
treatments which take account of the concentraion dependence of reaction rate would provide a
more accurate prediction of the critical jacket temperature. A range of thermo-analytical techniques
should be applied in order to determine reliable input parameters for the theoretical models.
REFERENCES
1) Rogers R.N. and Smith L.C., (1970) 'Application of scanning calorimetry to the study of
chemical kinetics', Thermochimicha Acta, 1, 1
2) Townsend D.T. and Tou J.C., (1980) 'Thermal hazard evaluation by an accelerating rate
calorimeter', Thermochimica Acta, 37, 1.
3) Fauske H.K. and Leung J.C., (1985) 'New experimental techniques for characterising runaway
chemical reactions', Chemical engineering progress, 81(8), 39..
4) Singh J., (1989) PHI-TEC: Enhanced vent sizing calorimeter application and comparison with
existing devices', International Symposium on Runaway Reactions', AIChE, 313.
5) American Society for Testing Materials, (1979) 'Arrhenius kinetic constants for thermally
unstable materials', ASTM E698-79, Nov. 79, Committee E-27, SCE27.02.
6) Chapman F.S. and Holland F.A., (1985) 'Heat transfer correlations in jacketed vessels',
Chemical Engineering, 15, 175-182.
7) Semenov N.N, (1928) Z. Phys. Chem.,48, 571.
P. F. NOLAN
Department of Chemical Engineering
South Bank Polytechnic
Borough Road
London SE1 OAA
UK
ABSTRACT. The available calorimetric methods are discussed in terms of
the data, which they provide in relation to the stages in the
development of a process. The data produced from calorimetric and other
thermal methods is placed within an overall assessment strategy
incorporating literature searches, calculations and formal quantitative
hazard assessment procedures and these may lead to the definition of a
basis for safety. Against this background a procedure is outlined which
will allow a beginner to establish a chemical reaction hazard assessment
laboratory for an industrial company.
1. INTRODUCTION
The majority of reactions carried out in the fine and speciality
chemical industries are exothermic and have the potential for overheating to occur in their normal batch or semi-batch methods of operation.
The safety record of the chemical industry is generally good when placed
against the background of the number of reactors in operation and the
quantities and types of materials processed. However, problems have
occurred in the past [1,2] due to:
a. inadequate understanding of the process chemistry and thermochemistry
b. inadequate engineering design for heat transfer
c. inadequate control systems and safety back-up systems
d. inadequate operational procedures, including training.
Various categories of hazard are inherent in chemicals manufacture.
These can be defined as chemical, operational, toxic and environmental.
The potential for such hazards, singly or in combination, can be present
during the stages of manufacture. (see Table 1)
Incidents have occurred in each of the stages below and hence it is
necessary to be in a position to predict any hazard producing potential
at every manufacturing stage. In particular, it is essential to have an
assessment strategy which links the nature of the chemical reagents, the
reaction with the equipment/plant in use and the operating procedures.
In general, chemical reaction hazards arise from:
391
A. Benuzzi and J. M. Zaldivar (eds.), Safety of Chemical Batch Reactors and Storage Tanks, 391408.
1991 ECSC, EEC, EAEC, Brussels and Luxembourg. Printed in the Netherlands.
392
TABLE 1.
a.
b.
c.
The implementation of the safety measures and their inclusion into the
process and plant design needs to be compatible with production and the
economics of manufacture. Finally, it is necessary to regularly monitor
the safety measures and ensure that they remain adequate.
2. STAGES IN PROCESS DEVELOPMENT
Gibson [3] has defined the stages in the assessment procedure against
the background of the normal development stages of a process. Process
development usually occurs in four stages, primarily defined by the
scale of operation - concept, laboratory bench, pilot plant and
manufacturing scale plant. At each stage, the potential reaction hazard
requires definition. Such continuous definition and in some cases redefinition leads to a comprehensive knowledge of the reaction at the
final, manufacturing stage. Whilst a continuous development of safety
considerations is recommended, it has to be recognised at the outset that
a substantial amount of the experimental work on reaction hazard
evaluation occurs, in practice, between laboratory bench and pilot plant
scale operations.
393
2.1. Laboratory Scale Development
After the conceptual stage of reaction design, it will be necessary to
carry out some bench-scale work in a fume cupboard. However, prior to
such activity it is necessary to:
a. check the literature for data on potential chemical hazards e.g. use
of Bretherick [4], NFPA [5] and Austin [6]
b. calculate the heat of reaction and the adiabatic temperature rise for
instantaneous reaction. The latter will enable the maximum
attainable process temperature to be calculated
c. ensure that the materials being handled will not detonate or
deflagrate and that the process does not involve a violently
exothermic reaction. The initial screen can involve:
i.
an examination of the chemical structure of the molecules
involved. Groups such as aromatic, nitro, nitrate ester and
nitramine are closely linked with explosibility and azo,
azide, nitroso, peroxide and acetylenic groups can also form
part of explosive structures
ii.
calculation of the oxygen balance. Almost all of the
recognised explosives have balances between -100 and +40, but
any balance more positive than -200 should be regarded as a
potential high risk
iii.
use of a computer program, such as CHETAH [7]. This allows
the estimation of the heat of reaction, thermodynamic
properties of individual substances and the prediction of
whether the compound or mixture has the tendency to propagate
a deflagration or detonation.
Other calculations may also be useful. Craven [8] described how the
approximate exothermicity of a reaction or decomposition can be predicted
from the atomisation energies of the reactants and the assumed products.
Atomisation energies can be determined from the summation of average bond
energies given in the literature [9]At this stage, it is also necessary to carry out some initial screening
tests (see Section 3.1) to answer specific questions on the exothermicity
of the proposed reaction. The main questions are:
a. What are the normal rates and quantities of heat and gas evolution?
b. At what temperature will runaway commence?
c. What are the consequences of runaway in terms of heat and gas
evolution rates?
Therefore, any sequence of initial screening tests must be able to
provide the following basic information:
a. temperature for onset of thermal decomposition
b. quantity and rate of heat release
c. rate of gas evolution as decomposition proceeds
d. catalytic effects caused by potential materials of construction of
the envisaged plant
e. autocatalytic effects
f. induction time effects
394
g.
395
occur from the normal process definition. The information required to
characterise a runaway or decomposition is:
a. the onset temperature for the specific plant system being studied
b. rate of heat evolution at runaway
c. rate of gas evolution at runaway
d. maximum pressure developed in a closed vessel when runaway occurs.
2.3. Manufacturing Scale Operations
Following bench scale and pilot plant operations, a number of further
decisions need to be made. The assessments carried out at the two
previous stages will need to be re-evaluated and some further testing
may be required to relate the data to the specific plant and operating
conditions. The manufacturing scale plant can be:
a. an existing reactor
b. a multi-purpose reactor, with possibly some modification to
accommodate the process
c. a new purpose-designed and built reactor.
The effect of scale of operation is a very significant factor, although
it is not always fully appreciated. Many incidents occur when processes
are scaled up, some having been performed successfully for years on a
slightly smaller scale. The reasons are clear when the following are
considered:
a. the rate of heat generation increases exponentially with temperature
b. the total heat generated is proportional to the batch volume (a cubic
term)
c. the rate of cooling is proportional to the surface area (a squared
term).
Any increase in batch size will substantially increase the total heat
generated. Unless the cooling capacity is increased accordingly then
the reaction temperature will rise. This leads to an increase in the
reaction rate and the reaction can go out of control.
The required cooling area for batch operation can be calculated by a
number of methods. The method described by Hugo [10] requires data on
the rate of reaction and the adiabatic temperature rise. Depending on
the reactor size, 2 - 5 m m
can be installed in reaction vessels with
only jacket cooling and 4 - 8 m m
if the vessel is also provided with
an internal cooling coil. The determination of the required cooling area
may mean taking the decision about moving from batch to semi-batch
operation. The latter allows the option of feed control to minimise
potential runaway conditions.
Any available safety system must be adequate for all situations, which
may arise. The bench scale studies can define the most suitable
agitator, the inside film heat transfer coefficient for use in scale-up
correlations and the design of normal operating and emergency relief
systems. The materials of construction of any plant needs careful
consideration, since it may catalyse undesired side or secondary
396
reactions. Laboratory scale tests may include samples of the materials
of construction or their decomposition products, e.g. rust.
The design of a reactor and the chosen operational procedures should be
accompanied by a series of formal hazard assessments. The easiest
technique for the nonspecialist is the use of HAZOP [11]. The Hazard
and Operability study can be carried out at various stages of the process
development. More quantitative techniques, e.g. the application of fault
tree analysis [12], require some specialist knowledge.
The manufacturing scale also raises the possibility of designing for
inherent safety; however, where this is not possible then reliance must
be placed on process control and associated safety back up systems, where
necessary. The basis of process control is set by the experimental
results from calorimetric investigations. These include the definition
of onset of exotherm temperature, the safety margins, amount of cooling
required, the temperature limit to prevent accumulation of unreacted
material and the effect of loss of agitation.
2.4. Modified Processes
Many incidents follow modifications to plant and/or process. Any
assessment procedure must cater for ongoing changes in the running of
processes and plants. If a hazard assessment of the plant and process
is available and the modification does not invalidate the basis for safe
operation then the formal recording of this decision and the reasons for
it are all that is required. Alternatively a detailed evaluation and
possibly alternative or additional recommendations to ensure safe
operation of the modified process or plant may be necessary. It is
essential that for any suggested modification, checks must be made on the
chemical and operational hazards and the basis of safety clearly re
defined.
There is currently no standard procedure or no one test for the assess
ment of reaction hazards; however, the type of data required, calls for
the use of calorimetric instrumentation.
3. CALORIMETRIC AND OTHER INSTRUMENTATION
Calorimeter is the generic term for a range of instruments used for
gaining thermodynamic, kinetic and heat transfer data to aid process and
plant design and operation. Calorimetry ranges in complexity from basic
heating tests to the sophistication of simulating an entire, as written,
process in a model benchscale, computercontrolled batch or semibatch
reactor. The main types of equipment, which can be used, after the
preliminary calculations and any necessary explosibility screening tests
are:
a. basic screening tests These are used primarily at the laboratory
scale development stage and prior to any studies outside the
laboratory environment.
b. isothermal calorimetry This is used to gain data on the normal,
397
c.
d.
398
sizing of emergency relief vents. The advent of commercial equipment
will aid the individuals/companies who are unable to develop their own
instrumentation, due to lack of workshop facilities or internal working
practices. However, the literature contains some very detailed
descriptions of easy to construct and use equipment, which can provide
the fundamental data on thermal stability and this includes onset
temperature, magnitude of an exotherm and its induction time. From such
data the size of the potential problem and the time available to correct
it can be estimated.
When expensive commercial instrumentation, such as the Seteram C-80,
Thermometries Thermal Activity Monitor, Columbia Scientific Industries
Accelerating Rate Calorimeter, is available some companies use them as
basic screening tests, although they are usually more expensive to run
and can take longer to provide the preliminary data. Although, of course,
they also provide additional data to that obtainable from the other
cheaper tests and instrumentation.
3.2. Reaction Calorimeters
Numerous commercial and "in-house" reaction calorimeters exist and the
choice of one particular instrument over another very much depends on the
type of data required (and the money available). Nearly all reaction
calorimeters can be said to operate in a number of modes, e.g. isothermal
and adiabatic. However, despite claims by some instrument manufacturers
their actual suitability for use in one or more modes can only be defined
for each specific study undertaken. Many of the instruments have common
features, although the method of gaining the calorimetric data may vary
slightly.
Dewar-based calorimeters are easy and cheap to construct [19]. Due to
their basic construction they can simulate the thermal characteristics
of the plant and measure the heat output of the reaction by measuring the
changes in the batch temperature. Experiments to determine the cooling
rates of large reaction vessels [20] have shown that the rate of heat
losses from 0.5 to 2.5 m vessels correspond to those from 250 ml and
500 ml Dewar flasks. Hence, measurements of the temperature rise when
reactants are charged to Dewar flasks at known rates enables the rate and
quantity of heat evolved in large scale plant to be found.
In the simple isoperibol calorimeter, the heat transfer medium in the
jacket is held at a constant temperature and the heat flow measured by
the change in reactant mass temperature. Unfortunately, it is usually
non-linear at high power outputs and both reaction kinetics and heat
capacity are affected by allowing the temperature to rise. The results
are only approximate. Power compensation calorimeters [21,22] are
relatively easy to design, construct and operate. The temperature of the
heat transfer medium in the reactor jacket is set below the desired
reaction temperature which is maintained by a heater in the reactants or
base. Any change in heat flow is compensated by a corresponding change
in the electrical power to the heater, which provides a direct measure of
399
the heat flow caused by the reaction.
Heat flow calorimeters monitor the temperature difference between a
rapidly moving heat transfer medium, e.g. silicone oil, in the reactor
jacket and the reaction mass. The heat (power) output of the reaction
is quantified by measuring the flow of heat between the batch and the
heating/cooling system. Heat balance calorimeters monitor the
temperature difference between the inlet and outlet points of the heat
transfer medium, which is flowing at a measured slow rate in the jacket.
All the above types of reaction calorimeters are ideal for studying
desired, normal reactions. They can provide temperature-time curves for
reactions, provide evidence of self-heating and can provide the data for
calculating heats of reaction and heat capacities. If assumptions can
be made about the reaction mechanisms, then it is possible to gain some
thermokinetic data on the reaction. The simplest type of analysis
assumes that the rate of reaction is proportional to the power output or
rate of temperature rise and is based on dimensionless rates and
concentrations. They can also be used to model semi-batch operation.
The heat flow calorimeter is an ideal instrument for gaining data on the
inside film convective heat transfer coefficient, which can be used as a
basis for scale up, in the standard heat transfer correlations in stirred,
jacketed vessels [23,24]. Geometric factors for the different types of
agitators can also be obtained from experiments using heat flow calorimeters [24]. However, it must be remembered that this is only possible
with reasonably mobile systems; too viscous materials cause a breakdown
of the necessary heat transfer paths between reactant-reactor wall-heat
transfer medium.
Reaction calorimeters can also be used to examine the problems of
accumulated heat, inhibition of the desired reaction and delayed
initiation. They are increasingly used for process development and
optimisation. It is possible to examine reactions under reflux
conditions with the addition of further equipment [25]. Reactions, which
generate limited amounts of gas can also be studied. Gas evolution
during a normal, desired reaction can be measured using an upturned
measuring cylinder filled with a suitable collecting fluid. Timing the
rate of gas collection can provide a quick assessment of the evolution
rate. An automated gas burette has been developed by staff at ICI pic
[26]. Lambert and Amery [27] have described the use of a thermal mass
flowmeter for measuring the gas evolved in both reaction calorimeters and
on full scale plant.
Commercial reaction calorimeters include the Toledo-Mettler Reaction
Calorimeter RC 1, the Thermometries Reaction Monitor RM 1 and the
Columbia Scientific Industries QRC (Quantitative Reaction Calorimeter).
However, numerous individuals and companies have designed and use their own
reaction calorimeters to great effect. These incorporate reflux
facilities [25] and the use of pressure vessels to extend the range of
reactions which can be studied [28]. Steele et al [24] has converted a
400
-3 3
675 x 10
m computer-controlled pilot plant reactor to operate as a
heat flow calorimeter.
3.2.1. Adiabatic Calorimeters
Uncontrolled reactions can originate from:
a. the desired reaction going out of control
b. a secondary reaction
c. a decomposition
The screening tests give some indication when runaway occurs but because
of the low heat loss conditions on industrial plant the accurate determination of the minimum temperature at which self-heating will start on
the plant requires the use of adiabatic calorimeters. While a number of
calorimeters can operate under an adiabatic mode, some specialist
instruments are available for examining the runaway condition.
Test methods can fail to detect the onset of an exotherm due to heat loss
from:
a. the sample to its surroundings
b. the sample to the test cell.
The latter is particularly important when the thermal capacity of the
test cell is large compared with that of the sample. A parameter, 0 is
used to characterise the effect and is termed the thermal inertia
0
1 + MC
c vc
MC
s vs
(C = specific heat at constant volume
v
subscripts s and c relate to the sample and container respectively
where M = mass
401
threshold, now down to 0.01 C min , i.e. any thermal excursion above
this threshold is recognised as an exotherm. The instrument is then
maintained under adiabatic conditions and temperature-pressure-time data
under runaway conditions is obtained. The data can be manipulated to
provide pseudo-kinetic information [29] as well as all the data offered
by the basic screening tests. Mores [31] has used ARC data, in
combination with self heat models, to gain self accelerating decomposition
temperatures [32] and for the modelling of sequential and parallel
reactions.
A number of other commercial instruments can provide similar data to the
ARC, e.g. Sikarex. The latter instrument can be operated under both
adiabatic and isothermal modes of operation. Pressure-temperature-time
data can also be obtained from instruments, which have been primarily
designed for the acquisition of information needed to size emergency
relief vents using the DIERS methodology [33,34]. Considerable attention
must be paid to the sensitivity of instruments over their defined range
of application and in terms of the type of substances investigated.
3.3. Instrumentation for Emergency Vent Sizing
The AIChE sponsored Design Institute for Emergency Relief Systems
produced a calorimeter, which uses a pressure equalisation system and a
weak test cell, with a low thermal inertia. The original Vent Sizing
Package (VSP) [33] uses a 120 ml heated test cell with a pressure control
system which balances the internal and external cell pressure and hence
the integrity of the test cell is maintained. The cell can operate in a
closed or open mode, the latter using a vent pipe into an outer containment vessel. Temperature-pressure-time data can be obtained, together
with the flow behaviour of the discharging runaway reaction masses. The
experimental data is combined with data from the actual plant (i.e.
ullage, total reactor volume and relief set pressure) and physical
properties of the reaction mass, i.e. density, to allow the calculation
of the size of the emergency vent. It is possible to simulate the
presence of an external fire on a reactor.
A similar device is the Phi-tech from Hazard Evaluation Laboratory Ltd.
Fauske and Associates, the originators of the VSP have now developed the
RSST [18]. This is a relatively cheap instrument, which operates in a
quasi-adiabatic mode. It is necessary to use some modified equations
from the DIERS programme for vent sizing. The vast majority of the
original VSP users have carried out substantial development work to
create improved instrumentation, yet retain the basic concept of
pressure equalisation to maintain the integrity of thin walled containers.
Whether a low thermal inertia factor results from this technique is
debatable. Some very sophisticated instruments now exist in the
specialist laboratories of industrial companies and academic establishments. Most developments are readily made available to others wishing
to modify existing equipment or to build their own vent sizing equipment.
402
A. BASIS OF SAFETY
The data from calorimetric experiments and the adoption of formal hazard
assessment procedures permit decisions relating to the safe design and
operation of the process and plant. Safe operation can be achieved by
preventative and protective measures. In order to select such measures,
it is necessary to define the worst case condition. The possible
measures include:
a. process control
b. process control + containment
c. process control + emergency venting
d. process control + inhibition of runaway reactions
A reactor can be designed to withstand the maximum pressure produced in
the runaway situation or alternatively a normal reactor can be installed
in a concrete and steel bunker. Since the products of a runaway can be
toxic, flammable, corrosive or foul smelling, containment offers the
possibility of restricting their release into the environment. Venting
is a normal operation in the course of chemical manufacture. Relief
vents sized to cope with gas pressure and/or fire engulfment are rarely
if ever adequate to provide protection against uncontrolled runaway.
The sizing of relief vents for worst case conditions has been covered by
the DIERS programme, although it is becoming increasingly important to
consider alternatives to emergency relief venting, due to economics, the
containing/disposing of the relief discharge and the problems of dealing
with the rate of rise in pressure found with some reactions. Inhibition
can take several forms. Quenching and dumping are common methods. The
sizing of dump tanks has been addressed by Grossel [35]. Occasionally,
it is possible to use reagents to remove free radicals from the reaction
mass.
In any considerations relating to the definition of the basis of safety,
it is necessary to always consider the reliability of the operators and
the instrumentation.
Process control can employ hard-wired instrumentation and microprocessors. Duplication of sensors monitoring key parameters is usually
recommended. The correct level of protection can be defined by the use
of Hazan [36].
5. ESTABLISHING A REACTION HAZARD EVALUATION LABORATORY
With the ever increasing legislation requirements and the quest for
better manufacturing routes to both old and new products, there is a
great need to support such developments with appropriate data on the
potential hazards and to provide information for the specification of
the design and safe operation of processes and plants. However, it is
easy to recognise that a need exists but it is a problem to actually
satisfy that need. It is important to establish "in-house" testing,
primarily because the individual company should know its processes
better than any external consultants and because the use of a hazard
403
assessment schedule also requires the incorporation of local process and
plant experience. Simultaneously, such testing may provide opportunities
for process development and optimisation. The initial steps are always
the most difficult because they involve the identification of the
problems of the individual company concerned. The literature and the
advertisements for commercial equipment can be confusing to the novice
in the field. Once the problems of the company have been identified,
the next stage is normally to seek advice from others, who have already
started such work. They will normally suggest the use of simple, cheap
methods presented in the literature or in specific publications e.g. the
ABPI Guidelines for Chemical Reaction Hazard Evaluation. Some of
Europe's major companies have their own "in-house" test methods and all
are prepared to give advice on tests, their results, the limitations and
their interpretation with respect to company activities in general terms.
The construction and use of cheap but effective methods for reaction
hazard screening will give confidence to the staff in an embryonic
hazard evaluation laboratory and it will be possible to see how the
measured data relates to past process and plant operating experience.
It is also possible to "calibrate" the obtained data by comparing the
results obtained on well defined materials with those given in the
literature for the same materials but found using more sophisticated
equipment. Based on experience, conversations with specialists and
comparative studies the level of confidence in the application and interpretation of data should grow.
At this stage, a decision must be made as to whether further equipment
and methods are needed. If developments are required, then the choice
of either purchasing commercial equipment or designing specific
instrumentation must be made. The work at the South Bank Polytechnic,
in London and its associated reaction hazard specialists based in
industrial companies has shown that it is possible to design, construct
and successfully use every type of calorimeter, including the next
generation of all-purpose calorimeters. Obviously, the development of
"in-house" instrumentation takes resources but they can be specifically
designed to solve particular problems. The purchase of commercial
equipment reduces the instrument development time but they all require
time for training of operators and for gaining sufficient experience to
apply the manufacturers operating instructions to specific industrial
reactions. An "in-house" instrument is usually easier to maintain and
modify. It must be stressed that in order to solve particular industrial
problems it may be necessary to operate commercial instruments in a
variety of configurations, other than that described in the manual of
operation. User forums certainly help the beginner to appreciate the
variations of operation and interpretations of data which are possible.
Gradually, the beginner is accepted as a specialist and user of
particular methods and a number of groupings have established themselves
for the free exchange of information. The European DIERS group meets to
discuss the uses of equipment, methods, data and designs. It is open to
industrialists and academics with practical experience. It is not
404
involved in training or in the promotion, marketing, sales of
commercial equipment. However, the people involved in reaction hazard
evaluation tend to meet regularly and exchange their experiences. They
have an open policy to welcome and guide new people into the subject area.
Experience of the very latest commercial equipment or some new idea to
permit the measurement of say, the rate of gas evolution is readily
discussed. The basic steps of setting up a reaction hazard evaluation
laboratory are given in Table 3.
TABLE 3.
i.
405
6. CONCLUSIONS
Reaction hazard evaluation must be considered at each stage of
development and scale of operation of a process. The details relating
to potential hazards need continuous expansion and re-definition. A
considerable number of methods and associated calorimeters exist and
their selective use can gain data, which can either directly, or when
used in appropriate formulae define a basis of safety for the
manufacturing scale plant.
The results from reaction hazard evaluation need to be combined with the
findings of formal hazard assessment procedures, e.g. HAZOP, FTA and
HAZAN, to define the worst case condition and the level of protective
measures required.
The basic steps of setting up a reaction hazard evaluation laboratory
include the identification of manufacturing problems and the acquisition
of raw and manipulated data. Confidence in the use of such data to
design and operate new plant and processes can be established on the
basis of relating laboratory-generated data to the past experience of
plant and process operations of the manufacturing company.
7. REFERENCES
1.
2.
3.
4.
5.
6.
7.
Manual of Hazardous
406
8.
9.
Sanderson R. T. (1971).
Press. New York.
Academic
10. Hugo P. (1980). Chem. Ing. Techn. 52, 712 and Hugo P., Konczalla M.
and Mauser H. (1980). Chem. Ing. Techn. 52, 761 included in
Westerterps K. Ft., van Swaaij W. P. M. and Beenackers A. A. C. M.
(1984). Chemical Reactor Design and Operation. Wiley. Chichester.
11. Kletz T. A. (1984). Hazop and Hazan - notes on the identification
and assessment of hazards. IChemE (Loss Prevention). Rugby.
12. Thomson J. R. (1987).
Harlow.
Longman.
London.
407
21. Wright T. K. and Butterworth C. W. (1987). Isothermal heat flow
calorimeter. IChemE Symposium Series No. 102. Hazards from Pressure.
Manchester.
22. Blitz J. L. (1989). The design and construction of a power
compensation heat flow calorimeter for the study of fermentation
processes. PhD CNAA South Bank Polytechnic, London.
23. Chapman F. S. and F. A. Holland (1965). Heat transfer correlations
for agitated liquids in process vessels. Chemical Engineering 1,
153 - 158.
24. Steele C. H. (1988). Calorimetric techniques for reflux analysis
and scale-up for the design and operation of batch reactors. PhD
CNAA South Bank Polytechnic, London.
25. Steele C. H. and Nolan P. F. The design and operation of a reflux
heat flow calorimeter for studying reactions at boiling.
26. Gibson N., Maddison N., Rogers R. L. (1987). Case studies in the
application of DIERS venting methods to fine chemical batch and
semi-batch reactors. IChemE Symposium Series No. 102. Hazards from
Pressure. Manchester. Pergamon Press.
27. Lambert P. G. and Amery G. (1989). Assessment of chemical reaction
hazards in batch processing. AIChE/IChemE International Symposium on
Runaway Reactions. AIChE Center for Chemical Process Safety,
Cambridge, Boston, MA.
28. Tee D. G. (1991). Heat flow calorimetric studies of ethylene oxide
reactions. PhD CNAA South Bank Polytechnic, London.
29. Townsend D. I. and Tou J. C. (1980). Thermal hazard evaluation by
an accelerating rate calorimeter. Thermochimica Acta 37, 1 - 30.
30. Rogers R. L. (1989). The use of Dewar calorimetry in the assessment
of chemical reaction hazards. IChemE Symposium Series No. 115.
Hazards X Process Safety in Fine and Speciality Chemical Plants.
Manchester. Pergamon Press.
31. Mores S. (1991). Enhanced control and data evaluation for
accelerating rate calorimetry. PhD CNAA South Bank Polytechnic,
London.
32. Mores S. and Nolan P. F. (1991). Determination of SADT from thermal
stability data generated using accelerating rate calorimetry.
Submitted to Journal of Loss Prevention in the Process Industries.
33. Fauske H. K. and Leung J. L. (1985). New experimental techniques
for characterising runaway chemical reactions. Chemical Engineering
Progress 81, 8, 39 - 46.
408
34. Fisher H. G. (1985). DIERS research program on emergency relief
systems. Chemical Engineering Progress 81, 8, 33 - 36.
35. Grossel S. S. (1990). An overview of equipment for containment and
disposal of emergency relief system effluents. Journal of Loss
Prevention in the Process Industries 3, 1, 112 - 124.
36. Lees F. P. (1980). Loss Prevention in the Process Industries.
Butterworth-Heinemann. Oxford.
O U T L I N E OF T H E M O D E L L I N G A C T I V I T I E S I N V E N T I N G
A. N. SKOULOUDIS
A B S T R A C T . The present work describes various aspects of modelling the loss of containment and
covers several problems which could be encountered in the theoretical calculations with several numerical
models (codes). The key features of four different codes which are suitable for the analysis of venting
transients are illustrated and the parameters which characterise the emergency relief of the reactor vessel
are identified. Six practical venting problems are analysed and comparisons are made with the data
calculated by the relevant codes.
1. G e n e r a l A i m s
The containment of multicomponent mixtures in vessels under high temperature and
pressure has been always an important subject for the safety of Chemical Industries.
There are several processes occurring during the emergency venting of vessels which
are directly related with the initial conditions which are frequently close to saturation.
Thus, after the opening of a relief device the contents of reactor vessel are emptied
under multiphase flow conditions. The following are typical processes identified during top venting transients (1,2,3). As soon as the relief valve opens vapour contained
in the freeboard volume of the pressure vessel will be released and the pressure falls
rapidly. The liquid phase cannot follow this rapid change of pressure with a prompt
change in temperature and the liquid becomes superheated. This leads to thermodynamic disequilibrium between the phases which is re-established after a short time
by vigorous re-evaporation of the liquid. During this period the high depressurisation
rate is reduced. Then a marked pressure recovery might occur when the vapour volume produced by evaporation exceeds the volume of the mixture which flows out of
the vessel. This recovery is very pronounced if a large and steep pressure decay has
occurred at the first stages of the transient. Vapour is still discharged through the
vent-line together with some droplets entrained from the interface separating the predominantly liquid and the predominantly vapour regions of the vessel. As soon as this
level reaches the vent-line a distinct two-phase mixture will be discharged with large
liquid content. Nevertheless, the evaporation processes continue and the thermodynamic disequilibrium is reduced. The interface level is gradually collapsing so t h a t the
vent-line will be no longer blocked. Then a predominantly vapour mixture will again
leave the vessel with several liquid droplets entrained. During this process the pressure
in the vessel falls continuously until a new state of equilibrium has been established
with the surroundings.
409
A. Benuzzi and J. M. Zaldivarfeds.), Safely of Chemical Batch Reactors and Storage Tanks, 409429.
1991 ECSC, EEC. EAEC. Brussels and Luxembourg. Printed in the Netherlands.
410
The magnitude and the interaction of the processes mentioned above depend on
a number of geometrical, operational and physico-chemical parameters. These parameters are identified in six examples which also demonstrate their importance. The
experimental d a t a reported here have been obtained at four test sites in five different vessels. The working fluids have been either water or refrigerant R114 and the
operating pressure are up to 7.2 MPa. The theoretical calculations have been carried
out with the codes RELAP, SAFIRE, RELIEF and D E E R S . Details about the input
options used by each code are described individually in each example.
2.
All the codes t h a t will be used in analysing the venting examples which follow have
been developed for different types of applications and although in principle solve similar sets of conservation equations for the mass, momentum and energy they all have
significantly different features for describing the phenomenology of the transient. Here
at the J.R.C. we have available the American codes RELAP, SAFIRE and D E E R S
together with the code RELIEF (originally called VESSEL) which has been developed
locally. Theoretical calculations from all of these codes are demonstrated in the next
section however, it is useful to describe here the key features of these codes and to
identify their modelling capabilities for the emergency venting of chemical reactors.
The main code features as used in the Benchmark Exercises (5) are shown in Table 1.
TABLE 1 : The main features of the codes
Name
of the
CODE
Nodes
in the
Vessel
Nodes
in the
Vent-line
No of
Chemical
Reactions
No of
Components
External
Heat
Fluxes
Two-phase
Flow
Models
RELAP5
Several
Several
Yes
According to
Flow Regime
SAFIRE
8 or 50
10
10
Yes
RELIEF
Several
10
10
Yes
Drift-flux
DEERS
Several
Several
Yes
Drift-flux
All the above mentioned codes employ different types of two-phase flow models.
This is a closure relationship which depends on the flow pattern of the liquid-vapour
mixture and correlates the two-phase parameters in the mass, momentum and energy
conservation equations. In principle, for the two-phase flow along a channel with or
without external heat input the following relationship is applicable
1 x
(Pa
ot
u>
1 a
411
This is directly derived from the definition of the quality, x, as the ratio of vapour to
the total mass-flow-rate and from the definition of the void fraction, a, as the ratio of
area occupied by the vapour to the total flow area. It is usually possible to express
x as a function of the local specific enthalpy from the energy conservation equation
together with the assumption of thermodynamic equilibrium between the two phases.
W h a t is still necessary in determining the other parameters included in equation 1
is either a relationship between x and a, or a relationship between the void fraction
and the ratio of the vapour to liquid average velocities. There is a whole range of
such relationships which are either of pure empirical origin or other more sophisticated
relationships which are based on the detailed phenomenology of the flow.
Most of the codes presented below, use a purely empirical relationship for the
velocity difference between the vapour and liquid phases (drift-flux model) or they
employ a typical drift-flux correlation. In the drift-flux terminology the relative velocity
between the two phases is directly related to the corresponding drift velocities ugj- and
u/_j according to
Ug-Ul
= Ugj - U(,j .
[2]
= a Ugj
[3]
= (1 - a) utj
[4]
412
Therefore, most of the code assessment work has been concentrated on separate effects
and integral test d a t a in the high pressure region.
Two-phase flow conditions in R E L A P are treated using separate mass and momentum equations for the individual phases (two-fluids model). Due to the assumption
t h a t the least massive phase is always saturated only one energy equation is needed
for the two-phase mixture. The approach selected can describe conditions of nonhomogeneous flow (different phase velocities), and thermal disequilibrium effects during the phase transition (evaporation, condensation). Mass, momentum and energy
exchange between the two phases are calculated by correlations which are flow regime
dependant. Specific models exist for the prediction of critical flow conditions (choking)
based on the sonic velocity in the single or two-phase fluid.
The R E L A P 5 - E U R / M F code includes the description of one-dimensional heat
conduction in the vessel wall and the heat transfer between the solid vessel wall and
the fluid. The heat transfer is calculated using the temperature difference between the
wall and the fluid and a heat transfer coefficient based on the concept of a "boiling
curve". The latter includes correlations for single phase natural and forced convection,
subcooled and saturated nucleate boiling, critical heat-flux, transition boiling, minim u m heat-flux, annular and dispersed film boiling. The heat transfer correlations are
automatically selected based on local flow parameters and wall surface temperatures.
Also, the loss coefficient due to abrupt area change at the exit of the vessel is calculated
internally by the code. The same is true for the friction between fluid and wall and
the friction between the phases.
The physical modelling of the code results in a system of first-order partial differential equations which are solved numerically by an efficient semi-implicit numerical
technique. The concept of free nodalisation gives to the code the capability to describe
a large variety of systems with different degrees of geometrical detail and makes it
capable of handling whole plant calculations. However, at the moment it does not
allow for chemical reactions during the venting process; thus its direct relevance to the
analysis of chemical reactors transients is rather limited.
2.2. THE SAFIRE CODE
The main feature of SAFIRE is its ability to handle up to ten simultaneous chemical
reactions with ten components. This code can only describe relatively simple vessel
geometries with nozzles or long pipes serving as venting devices and which can be
attached either at the top or at the bottom of the reactor vessel. SAFIRE solves the
one-dimensional mass, momentum, and energy conservation equations in the vent-line
but uses a single control volume for describing the vessel.
The flow regime inside the vessel is chosen by the user and once specified it does
not alter throughout the duration of the transient. The user may opt for either no
vapour/liquid disengagement coupled in the vent-line with either an homogeneous flow
regime or with a constant prespecified exit mass quality. Alternatively, the user may
select the partial vapour/liquid disengagement option with a "churn" or "bubble" flow
regime. These regimes are both typical drift-flux models in which the vapour drift-flux
and the bubble rise velocity are described by :
413
Bubbly flow
_
Jgi
UOQ =
a (1-Q)
Jgt
Uao =
f7,
l7J
(i _ a3)
1.18 [
Churn turbulent
g g
^ - ^ ] P*
2 5
[8]
U0
[10]
Radial variations of velocity and void fraction are taken into account via an empirical
distribution parameter C 0 . If C 0 is set to 1 there is not variation assumed in the radial
direction. Values between 1 and 2 should be set at the code input. The user has
also the possibility for setting the two-phase friction factor for the vent-line by either
selecting a value which is kept constant during the transient period or by activating
an option which calculates the friction factor automatically. When a two-phase flow
mixture is expected in the vent-line then this friction factor is calculated primarily
from single-phase flow relationships based only on the liquid phase viscosity.
The physical properties must be provided by the user in the input d a t a in terms
of the coefficients corresponding to the correlations included in SAFIRE. A constant
time step is also required at the input which is kept constant during the transient. The
size of this time step must be specified at the input.
To a certain degree the use of several input options in characterising the venting
process makes the code user dependant. For this reason in the examples described
below, two calculations from the SAFIRE code has been included which are contributed
by different users (11,12). Both calculations are still firmly based on the original
version of SAFIRE given to the J.R.C Ispra, however, in SAFIRE-2 calculations there
are included some structural alterations. These alterations are mainly addressing the
convergence problems encountered when the friction factor is calculated automatically
by the code. In these cases there are some differences in the way the coupling equation
between the vessel and the vent-line selects the right exit-quality when more than two
numerical solutions are found by the code.
The main comments which reflect our experience from the use of SAFIRE are
t h a t some alterations might be necessary for the numerical scheme to the code (12). A
two-phase viscosity should be used instead of the liquid viscosity which has been used
in the original version of the code (even if the mixture in the vent-line was containing
vapour up to 98%). However, it must be pointed out t h a t this code has been developed primarily for vent-sizing calculations and has been also used successfully for the
interpretation of the results from the DIERS project with chemically reacting fluids.
2.3. THE J.R.C. CODE "RELIEF"
The code RELIEF is under development at the J.R.C. Ispra. As used for the Benchmark Exercises (5) it has employed the opposite approach to SAFIRE. It vessel was
discretised in several control volumes but a single control volume was used for the
414
vent-line. The code has the capability to handle chemical reactions of arbitrary order
and can describe the vent line hydrodynamics if it is thought necessary. External heat
transfer through t h e vessel and the vent line can be also included.
The modelling in RELIEF is based on t h e following principles and assumptions
as described by Friz G. (13) and Duffield et al (14). Thermodynamic equilibrium has
been assumed between the phases. The pressure drop in t h e vessel due to friction,
and acceleration forces is assumed to be negligible and a drift-flux model has been
used as a closure relationship. The kinetic energy and axial conduction terms in the
energy equation are also assumed to be negligible. For t h e numerical solution of the
conservation equations a modified "donor-cell" technique which considers continuity
waves is adopted.
The drift-flux model is drafted in terms of the relative velocity between t h e phases
such t h a t ,
am(l-a)n
. .
U
-U>
= U>> J(1- M )
m
'
M
n
where otM is t h e void fraction which gives to the term a ( l a ) its maximum value.
The denominator is a scaling factor which ensures t h a t t h e maximum value of the slip
is given by Upooi irrespective of the value of void fraction a t which it occurs. The
coefficients m and n describe bubbly flow and droplet flow respectively and have been
fitted to experimental data. Upooi can be thought as a characteristic bubble velocity
which is closely related to t h e terminal velocity of the largest bubble in t h e system and
contains t h e physical property group
[goAp/pi)1'4.
In t h e Benchmark exercises t h e mass flow in t h e vent-line was modelled in a
simplified manner taking into account blowdown experiments performed a t the J.R.C.
There t h e mass flow in both the super-critical and sub-critical region is calculated in
a similar way as in t h e gas dynamic theory using t h e upstream stagnation pressure
and t h e mixture density. The friction a t the entrance region and along t h e vent line is
considered and t h e critical pressure ratio, used for the transition to sub-critical flow, is
taken from t h e ideal gas relationship. With t h e assumption of homogeneous conditions
at t h e entrance of the vent-line the volumetric flow is obtained from t h e mass flow
using t h e mixture density taken from t h e cell which adjoins the vent-line.
2.4. THE DEERS CODE
The Design and Evaluation of Emergency Relief Systems code (DEERS) solves t h e
one-dimensional mass, momentum and energy conservation equations as described by
Klein (15). These equations are solved in the axial direction of t h e flow for t h e vessel
and for various configurations of the vent-line all t h e way up to t h e end of the catch
tank.
It can handle up to four chemical reactions with six components. Different temperature, pressure and concentration of these components are allowed along t h e axis
of t h e flow. Along the vessel t h e fluids are allowed to exist in different combinations
of liquid and vapour mixtures and chemical reactions can occur either in t h e liquid
o r / a n d in t h e vapour phase of the various components.
Nozzles or different combinations of long pipes can be taken into account in the
vent-line which can be a t t h e top a n d / o r a t t h e bottom of t h e reaction vessel and can
415
be vertical or inclined to the horizontal. In addition to these various rupture disks can
be placed at any control volume along the reactor vessel.
The two-phase model employed by this code relates the void fraction to the relative
velocity between the two phases according to :
ut
= sj 10 Ad [ apg + (1 - a)pi ] a
[121
where the empirical factor Aj, is referred to as the interface slip coefficient and has
dimensions of m Kg
s . This coefficient has been tested against experimental
d a t a from the M P M C test facility of the J.R.C. Ispra (7) for simple non-reacting fluids
and the following values were proposed:
A d = 8.0 X 10~ 4
Aj. = 1.5 X l 0 - 5
A j = 0.0
(Non-foamy liquids)
(Foamy liquids)
(Homogeneous mixture).
Similar to the SAFIRE and the RELIEF codes this model is not allowed to alter
throughout the duration of the transient and has been used for void fractions from
0 to 1 irrespective of the flow regimes. Comparisons have been made (16,17) with
several experimental d a t a obtained under steady state conditions in order to test the
accuracy of this model. For void fractions between 25% and 70% it has been found to
be accurate enough for several types of two-phase mixtures.
Heat conduction in the solid vessel wall and heat transfer between the wall and
the vessel contents can be also taken into account. Various external heat fluxes are
allowed along the axis flow thus, venting transients under complex heating or cooling
conditions can be analysed. The two-phase friction factor and the loss coefficients due
to an a b r u p t area change in the vent-line are calculated internally by the code. In
addition to this, the user has the possibility to supply externally different heat transfer
coefficients and different wall roughness factors in the various control volumes.
D E E R S is a general purpose code which can be used in the venting of a large
variety of systems. However, the use of a single two-phase model throughout the whole
transient restricts the accuracy of its predictions. Another disadvantage of this code
is t h a t it requires several hours of CPU time in an ordinary Personal Computer. Both
these have been modified at the Joint Research Centre by converting and transferring
the code to mainframe computers and by introducing several two-phase flow models
into the code.
3. V e n t i n g E x a m p l e s
The purpose of the examples shown here describe the phenomena and demonstrate the
importance of some of the parameters related with venting. Basis of the Examples are
two experiments carried out at the Multiphase Multicomponent (MPMC) test facility
of the Joint Research Centre together with two test cases with refrigerant R114 which
have been kindly contributed by Hoechst AG (8). The input conditions of these tests
are selected so as to be in line with two more depressurisation experiments reported in
the international literature (18,19). The dimensions of the vessel, the working fluids,
416
the starting conditions, the type and the dimensions of the vent-line are summarised
in Table 2.
The same parameters have been monitored for all the examples presented below
however, more emphasis is given to comparisons where experimental d a t a are available.
The parameters monitored are the pressure, the void-fraction and the liquid/vapour
interface in the reactor vessel. At the exit of the vessel the mass-flux and the static
exit-quality are also calculated. The latter is defined as the ratio of the vapour mass
to the total mass exiting the vessel at a particular time step. In the first two examples
with a long venting pipe attached at the top of the vessel the pressure history at a
particular location is also calculated by the participating codes.
Not all of the above mentioned parameters could be calculated for each example
due to the modelling assumptions taken by each code. For example, with the SAFIRE
code it isn't possible to calculate the interface level-swell in the vessel due to the
assumption t h a t the vessel is modelled with a single node. Similarly, R E L I E F did
not provide any d a t a for the pressure and static quality in specific locations along the
vent-pipe because the whole vent-pipe was considered as a single node.
Example
No.
Vessel
Volume
K)
1
50xl0-3
50xl0~
6.32X10-3
Vessel
Diam.
(m)
Type
of
Vent
Vent
Diam.
(cm)
Working
Fluid
Init.
Pressure
(Pa)
Init.
Void
(%)
0.40
L/D=90
2.0
Water
2.5x10s
0.40
L/D=90
2.0
Water
2.5X10
0.178
Nozzle
1.27
Water
2.9x10s
s
46
35
2
4.53
1.19
Bottom
Nozzle
5.4
Water
7.2 x 1 0
14.6X10-3
0.216
L/D=79
Orifice
1.9
1.5
Freon 114
7.0x10s
15
105 x l O - 3
0.40
L/D=79
1.9
Freon 114
8.3x10s
15
TABLE 2. The venting examples considered for different vessels and different fluids;
L/D refers to long pipes with the specified Length/Diameter ratio.
The codes used here do not include a level tracking model for the interface in the
vessel between the predominantly liquid and the predominantly vapour regions. For
this reason R E L A P uses a rather simple model which assumes t h a t the mixture level
existed in the highest volume within a vertical channel where the liquid volume fraction
exceeds the value of 0.05. The exact location of the level is then found according to:
au - a
level
13
H + dh
where H is the bottom elevation of the volume identified, dh is the height of this volume,
a is the void fraction, OCT, and a y are the values for the void fraction in the volume below
417
or above respectively. For the codes RELIEF and D E E R S the liquid/vapour interface
is assumed to be at the place where the void fraction gradient gets its maximum value
along the vessel length.
Different nodalisation approaches have been followed in the various codes without
any a t t e m p t to optimise the size a n d / o r the number of control volumes used. In R E L A P
the vessel is subdivided into 10 volumes with the exception of Example-4 where 12
volumes were used in order to allow the correct position of the discharge line, and the
correct vessel geometry. In the cases with a long vent-pipe R E L A P used only 2 control
volumes. In the calculations with the two SAFIREs the vent-line was represented with
8 control volumes (SAFIRE-1) and with 50 control volumes (SAFIRE-2). R E L I E F in
all the cases used 100 control volumes for the vessel. As is already shown in Table 1
SAFIRE and RELIEF used a single control volume in the vessel and in the vent-line
respectively. In all the calculations shown here with the D E E R S code the vessel and
the vent-line have been modelled with 18 and 15 control volumes respectively.
3.1. EXAMPLE-1; T O P VENTING, LOW PRE-RELIEF VOID
This is a water blowdown case which has been carried out in the Multi-phase Multicomponent test facility of the J.R.C. Ispra with a long venting pipe attached at the top
of a cylindrical vessel. In this example the contents of a nearly full vessel under low
starting pressure (compared to the ambient pressure) are vented in a catch tank. The
experimental d a t a available for this exercise are shown in Fig.l together with other
parameters which are calculated by the participating codes.
The driving mechanism is the limited vapour production in the vessel which holds
the pressure close to the starting value while a predominantly liquid mixture is expelled
under high mass-fluxes. The whole process maintains the vessel pressure high (Fig.l)
for a relatively longer period than in the corresponding case with lower initial filling
(Example-2). The pressure oscillations will continue until enough vapour has been
produced and the liquid-vapour interface has fallen below the entrance of the vent-line.
The differences between the two SAFIRE calculations are due to the empirical
drift-flux distribution parameter C0 for the void fraction and due to the different ways
by which the two-phase friction factor is calculated in the vent-line. In SAFIRE-1 the
C0 has been taken as 1.5 together with a constant two-phase friction factor whereas,
in SAFIRE-2 C0 was equal to 1 and the two-phase friction factor has been calculated
internally according to the local conditions in each node along the vent-line. In both
codes the flow regime in the vessel has been assumed as churn turbulent with partial
liquid/vapour disengagement. The flow in the venting pipe has been modelled in
both SAFIRE calculations as a homogeneous mixture. Details about the two types of
calculations with SAFIRE can be found in (11) and (5,12) respectively.
All codes reproduce reasonably well the pressure curve inside the vessel with a
rather smooth parabolic type of curve. The higher mass-flux predicted by the code
R E L I E F during the first 6 seconds of the transient means t h a t enough liquid is expelled
from the vessel so t h a t bulk evaporation will begin earlier than for the other codes.
This is also reflected in the void-fractions comparisons (Fig.l) at least in the first few
seconds of the transient. These void-fractions in the vessel are then translated into
higher exit-qualities. Contrary to this the calculations by R E L A P and D E E R S show a
418
EXAMPLE
PRESSURE
IN
THE
VESSEL.
LU
1.80
<n
1.60
1.00
T I M E
(SEC)
FIG. la.
SAFIRE-1;
SAFIRE-2;
or (x)
RELIEF;
Experiment;
DEERS j .
419
EXAMPLE 1 MASS - FLUX AT THE EXIT OF THE VESSEL
1500.
T I M E
(SEC!
FIG. lb. Top Venting, 2.5 bar pre-relief pressure, 2% initiai void [
or fxj Experiment;
RELAP;
SAFIRE-1;
SAFIRE-2;
RELIEF;
DEERS j.
420
EXAMPLE 1 NORMALISED LEVEL SWELL IN THE VESSEL
;
:
0.45
0.40
0.35
0.30
0.25
:
X
0.20
0.15
0.10
0.05
/<
V/
/
0.00
40.0
T I M E
60.0
(SEC)
SAFIRE-1;
SAFIRE-2;
or (x)
RELIEF;
Experiment;
DEERS j .
421
period of mass-flux fluctuations while the exit quality remains very close to all-liquid
flow. During this period the pressure in the vessel is maintained at high levels; the latter
explains to a certain extent the pressure fluctuations observed in the vent-line. The
phenomenological differences between the RELIEF, R E L A P and the D E E R S codes are
also found in the comparisons of the exit-qualities calculated during the first 40 seconds
when the mass flux is rather unstable. The two SAFIRE calculations are similar to
those of RELAP.
The exit-qualities shown in Fig.l are in line with the calculations for the level swell
which is a fictitious interface between the predominantly liquid and the predominantly
vapour regions in the vessel. For the RELIEF code this level swell causes this interface
to be full contact with the top of the vessel during the first 58 seconds of the transient.
A similar level swell curve is also shown in the same figure for the R E L A P and for
the D E E R S codes. The differences between these codes reflect the variations of the
exit-qualities calculated although it must be also taken into account t h a t the level-swell
is calculated differently in RELAP.
3.2. EXAMPLE-2; T O P VENTING, LARGE PRE-RELIEF VOID
This is a similar case as in Example-1 but at the beginning of the transient only
65% of the vessel volume was filled with water. This changed completely the venting
mechanisms and the vapour produced in the vessel was not sufficient to maintain a
high pressure for very long (Fig.2). For this reason the pressure curve in the vessel is
not of the same parabolic shape as in Fig.l. The amplitude of the pressure oscillations
observed in the corresponding figures for the previous case are significantly reduced as
is shown in Fig.2 and the pressure in the vessel falls rather fast in the first 15 seconds
of the transient. Also, the delay in boiling during the first few seconds of the transient
has a much more profound effect in this case rather than in first example where the
vessel was practically full.
All calculations carried out seem to reproduce rather well the experimental d a t a
for the pressure in the vessel. The calculations for the pressure in the vent-line are in
good agreement experimental values after the first 10 seconds of the transient. In this
example the two SAFIRE curves have been calculated with the same input options as
those used in Example-1.
The pressure curves corresponding to the two SAFIREs are strongly influenced
from the way the two-phase friction factor was calculated in the vent-line. This together
with the differences in the C0 value are the main reasons for the variations observed
in Fig.2 among the calculated values of the mass-flux at the exit of the vessel. What
is also worth observing from the same figure is t h a t the period of predicted two-phase
flow in the vent-pipe is rather short.
All calculations presented in Fig.2 for the exit-quality demonstrate t h a t after a
certain time the flow in the vent-line becomes all-vapour flow until the end of the
transient. This time is shorter for R E L A P and longer for the R E L I E F code. In
the same figure it is shown t h a t D E E R S still predicts oscillations for the exit-quality
and the mass-flux for the first 30 seconds of the transient. Contrary to the previous
example, the calculated level swell does not reach the vessel top as is shown in Fig.2.
The RELIEF and the D E E R S codes, predicted venting of a two-phase mixture during
422
EXAMPLE 2
2.00
<r>
1.60 .
L0
PRESSURE
20.0
40.0
T I M E
IN
THE
VESSEL.
60.0
80.0
(SEC) .
FIG. 2a. Top Venting, 2.5 bar pie-relief pressure, 35% initial void [
RELAP;
SAFIRE-1;
SAFIRE-2;
or (x)
RELIEF;
Experiment;
DEERS j .
423
EXAMPLE 2 MASS FLUX AT THE EXIT OF THE VESSEL
40.0
60.0
TIME
(SEC) .
0.90
0.80
0.70
0.60
till! I
-. ,. : ii/ H i
/ i l l ! ! Iil'U j i / >
b il
0.50
0.40
0.30
0.20
F' '
HI?
i;
|iih * inI
till I1 K! I I Ml U "
OK\ i y i
ni i * v
T I M E
C
(SE )
FIG. 2b. Top Venting, 2.5 bar pre-relief pressure, 35% initiai void [
RELAP;
F
SA IRE-1;
F
SA IRE-2;
or (x)
F
RELIE ;
Experiment;
DEERS j .
424
EXAMPLE 2 NORMALISED LEVEL SWELL IN THE VESSEL
1.20
i.O
20.0
40.0
TIME
EO.O
80.0
(SEC) .
FIG. 2c. Top Venting, 2.5 bar pre-relief pressure, 35% initial void [
RELAP;
SAFIRE-1;
SAFIRE-2;
or (x) Experiment;
RELIEF;
DEERS j .
425
the first 38 seconds. This is probably due to the method used in calculating the mixture
level which allows the local void fraction above this level to be different from one.
On the other hand, R E L A P does not show any contact of the calculated interface
with the vessel top, there are not any mass-flux fluctuations this time and after a
short period of mass-flux increase the flux follows an exponential decrease. This is in
agreement with the calculated exit-quality which is practically one (all-vapour flow)
during the whole transient. As for the void-fraction in the vessel, all codes seem to
agree on the shape of the curve calculated with small variations for the final vessel void
fraction.
The general conclusion drawn from this example is t h a t when the vessel is not
full, the depressurisation lasts for a shorter period and the vessel pressure falls rather
quickly in the beginning of the transient. The vapour production does not maintain
the initial pressure in the vessel, the period of two-phase flow in the vent-line lasts
shorter and the calculated mass-flux is smaller than in Example-1. Also, the pressure
histories in the vessel and in the vent-line are free from the high frequency oscillations
found in the previous example.
3.3. EXAMPLE-3; TOP VENTING THROUGH A NOZZLE
This is a water blowdown exercise carried out at the Thayer School of Engineering (18)
in a small vessel with a nozzle rather than a long vent-pipe attached at the top of the
vessel. The input conditions are similar to those of Example-1 thus, we can identify if
the phenomena described in the first example are independent of the vent type and of
the size of the reactor vessel. The experimental d a t a available for this exercise are the
pressure history at the b o t t o m of the vessel, the final void fraction in the vessel and
the interface level-swell d a t a measured by an electrical circuit between the nozzle and
the water in the vessel.
The experimental pressure in the vessel is shown with the continuous line in Fig.3.
The high frequency pressure fluctuations are present for about 8 seconds and the delay
in boiling is of the same magnitude as the fluctuations in the pressure curve. Similar to
the first example, bulk vapour production does not start immediately and the vapour
produced in the vessel maintains the pressure in the vessel at relatively high levels.
This lasts until enough void is created in the vessel and during this period the interface
level is up to the top as is shown in Fig.3. Then, the interface level fluctuates up to
the top of the vessel indicating that the exit-quality is varying during this period of
instability. This also affects the calculated mass-flux and the pressure in the vessel.
Apart from the parabolic shape of the pressure curve, the calculations with REL A P and D E E R S shown in Fig.3, give a better insight to the phenomena involved
by being the only codes which account for the fluctuations in the interface level-swell.
These fluctuations are probably the result of the variations observed in the mass-flux
and the static quality at the exit of the vessel. In the two SAFIRE calculations the same
equilibrium rate model is used to describe the flow through nozzles. The differences
between those calculations are due to the empirical drift-flux distribution parameter
C0 for the void fraction. Contrary to the Examples 1 and 2, the mass-fluxes calculated
from SAFIRE 1 and 2 are comparable. Thus, for the SAFIRE code, the mass-flux is
more sensitive to the two-phase friction factor than to the value of C0.
426
EXAMPLE 3
UJ
1.50
1.00
PRESSURE
T I M E
IN
THE
VESSEL.
(SEC) .
6000.
5000.
:-*--
FIG. 3a. Top Venting, 2.9 bar pre-relief pressure, 2% initial void [
RELAP;
SAFIRE-1;
SAFIRE-2;
RELIEF;
Experiment;
DEERS j .
427
EXAMPLE 3
TIME
(SEC)
T I M E
(SEC)
FIG. 3b. Top Venting, 2.9 bar pre-reJief pressure, 2% initial void [
RELAP;
SAFIRE-1;
SAFIRE-2;
RELIEF;
Experiment;
DEERS ,
428
The exit-qualities shown in Fig.3 are in line with the calculations for the interface
level swell. The level-swell calculated by the RELIEF code (Fig.3) is in full contact
with the top of the vessel during the first 18 seconds. After this period this interface
level falls below the top of the vessel and all-vapour flow starts. The corresponding
interface calculations from the codes RELAP, RELIEF and D E E R S are shown in the
same figure and are in good agreement with the experimental data.
3.4. EXAMPLE-4; BOTTOM VENTING
This is a bottom venting exercise carried out in the large vessel of General Electric
(19). This vessel is 1.19 m in diameter, 4.07 m in height and the venting nozzle
has been placed 0.7 m from the bottom of the vessel. The the starting pressure was
72 bar thus, the capabilities of the codes are tested under high operating pressures
together with their abilities for scaling up from laboratory size vessels. The parameters
measured in this example are shown in Fig.4. These parameters are the pressure and
the liquid/vapour interface in the vessel, as well as the mass-flux and the static quality
at the exit of the vessel.
As is shown in Fig.4, the pressure in the vessel decreases slowly during the first 20
seconds of the transient this is then followed by a period of rapid pressure reduction
until a new state of equilibrium is reached. During the first 20 seconds the vapour
produced in the vessel holds up the pressure in the vessel whilst a predominantly
liquid mixture is expelled from the vessel. This is confirmed also by the low exitquality during the first 16 seconds of the transient until the liquid/vapour interface
reaches the venting nozzle. Then, two-phase mixture is expelled from the vessel and
during this period the pressure in the vessel falls rapidly.
In the two SAFIRE calculations the radial variation of the void fraction was
deactivated (C0 1) and a churn turbulent flow regime in the vessel has been assumed
with no partial liquid/vapour disengagement. The quality at the entrance of the venting
nozzle in both codes is assumed to be the average quality of the vessel. However, in
SAFIRE-1 the Equilibrium rate model has been employed for the two-phase nozzle flow
whereas, in SAFIRE-2 the Homogeneous Equilibrium Model (HEM). The differences in
the flow model are not very significant for the pressure and the mass-flux but affected
the calculated exit-qualities especially close to the end of the transient. It must be
pointed out t h a t in SAFIRE the position of the venting nozzle can be only placed
at the b o t t o m of the vessel (due to the fact t h a t the vessel is modelled with a single
control volume). Therefore in SAFIRE the transient is terminated with 100% void in
the vessel. This was not the case for the other codes which discretised the vessel with
several control volumes.
Better representation has been achieved with the R E L I E F and R E L A P codes
which modelled the vessel with several nodes. In fact the shape of the pressure curve
calculated by R E L A P is very similar to the Experimental curve even with a small number of nodes in the vessel. Close to the end of the transient, due to liquid entrainment
in the vent the measured mass-flux is higher than the value predicted by R E L I E F and
RELAP. These small differences in the mass-flux are translated into large deviations
from the measured exit-qualities due to two-dimensional effects which could not be
taken into account in one-dimensional codes. The calculations with the D E E R S code
429
EXAMPLE 4
PRESSURE
IN
THE
i
VESSEL .
>
'
50.0
35000. 1,
-J
15000.
RELAP;
430
EXAMPLE 4 STATIC QUALITY AT THE EXIT OF THE VESSEL
\
\
0. so
0.40
j
J
i
\ \
0.20
0.10
\ ~\
0.30
NN.
:
-
0.00
T I M E
(SEC)
RELAP;
431
are also in excellent agreement with the experimental d a t a for the exit-quality. The
shape of the other curves is reproduced equally well however with some significant
differences for the vessel pressure and for the interface level swell. These differences
are due to the interface slip coefficient used in the calculations. The value suitable for
non-foamy liquids has been selected which was the same as in all the other top venting
cases with no a t t e m p t to optimise it for bottom venting.
The venting mechanism noticed from this bottom venting example is t h a t the
vapour produced during the first part of the transient cannot easily escape from the
vent. This process lasts until the level swell in the vessel drops significantly and until
two-phase flow starts. Then the change of the flow regime creates a sharp reduction of
the pressure in the vessel. In reality this change is not so sharp due to twodimensional
effects. Better representation is achieved when the vessel is analysed with several nodes.
3.5. EXAMPLE-5; TOP VENTING, SMALL VESSEL
This example is a depressurisation exercise which has been carried out at Hoechst
with refrigerant R114 as the working fluid. The venting pipe includes an orifice which
has been placed 0.70 m from the exit of the vessel. The parameters measured during
this example are shown in Fig.5. For reasons of simplicity only one calculation from
SAFIRE is presented here. This calculation has been carried out with churn turbulent
flow, partial liquid/vapour disengagement, and with the drift-flux distribution parameter C0 equal to 1. The orifice plate was neglected in SAFIRE and the diameter of the
venting pipe was taken as 1.9 cm. R E L A P and D E E R S have modelled the orifice by
suitably adjusting the flow area between two control volumes used for the vent-pipe.
In the RELIEF code the discharge flow-rate model has been adjusted according to the
physical properties of the refrigerant R114. Heat-transfer phenomena from the vessel
wall have been taken into account in RELAP.
W h a t is significantly different in this example in comparison with Example-2,
is t h a t the delay in boiling during the first second which does not seem to affect the
pressure curve so much as in the other venting cases. The predictions with the RELIEF
code are in excellent agreement with the measured values. As is shown in Fig.5, all
codes seem to overpredict the mass-flux even though the calculated qualities at the
exit of the vessel are in relatively good agreement with measured values. It should be
mentioned here t h a t both the mass-flux and the static vent-quality have been measured
1.5 m downstream from the exit of the vessel.
The static quality for refrigerant R114 seems to fall rapidly to low values (close
to all-liquid flow) followed soon afterwards with a sharp increase to all-vapour venting.
The exit-qualities calculated by the RELIEF and the D E E R S codes showed t h a t allvapour flow has been eventually achieved after the first 30 seconds. The calculations for
the void-fraction in the vessel (Fig.5) reflect the differences between the calculated and
measured mass-fluxes so the actual voids are much lower than the calculated values.
3.6. EXAMPLE-6; TOP VENTING, LARGE VESSEL
In the present venting example the dimensions of the cylindrical vessel has been nearly
doubled so t h a t the vessel volume is now 105 litres instead of the 14.6 litres used in
432
EXAMPLE 5 PRESSURE
T i
IN
TH E
i
VESSEL.
1 i
! \ \*
"
\ \VV \
UJ
4.00
:
<n
"
3.00
2. 00
\\x
N
\ "
.
;
^ x
\ N\
\ \ \ \ \\ >
X*
\
^ 0^s ,\ X x\ * x
\\
\
~^C^- x * * < x
:
;'.
X w
*5T.
10.0
20.0
T I M E
30.0
10
(SEC) .
"
\ \
\ \X
fc*
f
It
xx
;
>^x
X
.
''^
^x^Tx
20.0
T I M E
I SEC)
FIG. 5a. Top Venting, 7 bar pre-relief pressure, 15% initial void [
RELAP;
F
SA IRE;
F
RELIE ;
or (x)
DEERS j .
Experiment;
433
EXAMPLE 5 QUALITY AT THE EXIT OF THE VESSEL
10.0
'
20.0
T I M E
FIG. 5b. Top Venting, 7 bar pie-relief pressure, 15% initial void [
RELAP;
SAFIRE;
RELIEF;
or (x)
DEERS }.
Experiment;
434
EXAMPLE 6 PRESSURE
IN
TH E
VESSEL.
J I I I T J
3.00
0.00 L .
0.0
_1_
10.0
20.0
30.0
TIME
(SEC) .
- /".
; \
:/
\
-i
\
LI V
'
\
jtV*
K,
\
N
>
#[
"x
\
N
x
*"% w
^
'
H
20.0
T I M E
30.0
I SEC) .
FIG. 6a. Top Venting, 8.3 bar pre-relief pressure, 15% initial void [
RELAP;
F
SA IRE;
F
RELIE ;
DEERS j .
or (x)
Experiment;
435
EXAMPLE 6 QUALITY AT THE EXIT OF THE VESSEL
0.30
FIG. 6b. Top Venting, 8.3 bar pre-relief pressure, 15% initial void [
RELAP;
SAFIRE;
RELIEF;
DEERS j .
or (x)
Experiment;
436
the previous example. The dimensions of the venting pipe which is again attached at
the top of the vessel are the same as in Example-5. However, contrary to the previous
example no orifice was present and the starting pressure was now 8.3 instead of 7 bar.
The experimental d a t a available from Hoechst AG are the pressure in the vessel,
the mass-flux and the quality 1.5 m from the exit of the vessel (Fig.6). Thus, the
measurements must be influenced both by the change in the size of the vessel as well as
from the change in the type of vent used. The shape of the pressure curve in the vessel
shown in Fig.6, is significantly different from the corresponding case in Fig.5. The
transient now lasts much longer due to the larger amount of liquid initially contained
in the vessel. The pressure does not fall as fast as in the previous example and the initial
delay in boiling with the pressure recovery afterwards not as large as in Example-5 or
as in similar water depressurisations. Temperature measurements before the opening
of the valve has confirmed t h a t the liquid was at saturation at 74 C. The period over
which this delay in boiling takes place is much longer than in the smaller vessels. This
is probably due to the larger amount of liquid contained in the vessel prior to initiation
of the transient. The best agreement in the pressure curve is predicted by RELIEF
after the first 10 seconds.
Similar to the previous example only one SAFIRE calculation is presented in
Fig.6 with the same input options for the flow regime in the vessel and in the vent-line.
Heat-transfer phenomena from the vessel wall have been also taken into account in the
calculations by RELAP.
For all codes the mass-flux calculated at the exit of the vessel is much lower than
the actual values measured 1.5 m downstream in the vent-line. The calculations from
R E L I E F are in closer agreement with the experimental d a t a shown in Fig.6. The
maximum value of the mass-flux measured in this exercise is slightly lower than the
corresponding value in Fig.5 when an orifice was reducing the flow area of the ventpipe by 38%. These differences in the mass-flux are expected to be also present in the
exit-quality. This is not the case though as is shown in Fig.6. The experimental d a t a
in this figure correspond to the static-quality measured 1.5 m downstream in the ventline and could be compared with the calculated quality at the exit of the vessel only
when the liquid entrained in the vent-line is assumed not to evaporate in this length.
The exit-qualities calculated with the SAFIRE and the D E E R S codes are in excellent
agreement with the experimental values with the corresponding R E L A P calculations
also reasonably close. A comparison between the exit-qualities measured in Examples
5 and 6 shows t h a t the quality of the mixture is above 50% and all-vapour-flow starts
later than when there was an orifice (Fig.5). In the latter case the exit-quality gets
close to all-liquid flow but for a shorter period of time.
This example demonstrated t h a t the evaporation process depends on the size of
the vessel even if the initial void-fraction is the same. On the other hand, the size of the
vent employed changes the duration of the two-phase release as well as the composition
of the two phase mixture. These two parameters might have cancelling effects which
cannot be easily identified when only the pressure in the vessel is monitored.
437
4. Conclusions
The present work focuses on the phenomenology of the depressurisation of reactor
vessels through six experimental examples performed in different test facilities. These
examples have been analysed in order to demonstrate the importance of the geometrical
and operational parameters.
The depressurisation processes depend on the size of the vessel and on the size of
the vent used. For example, the level swell oscillations observed in small reactor vessels
does not necessarily mean t h a t they will remain the same when the dimensions of the
vessel and the vent-line are increased. Also, the position of the vent can cause changes
in the depressurisation processes. Thus, during b o t t o m venting, the liquid discharge
causes a rather small depressurisation rate during the first seconds of the transient;
this rate is significantly increased later when two-phase flow starts.
Comparisons between different top-venting examples carried out with a nearly
full vessel and when the vessel was half-empty revealed two different depressurisation
processes depending on the initial fillings. The first process occurs in an almost full
vessel during the first seconds of the transient and is characterised by vapour production
together with an expulsion of a predominantly liquid mixture under varying massfluxes. In the second process a predominantly vapour mixture is expelled with larger
depressurisation rate than in the first process.
The characteristic delay in boiling which occurs during the top venting cases after
the venting valve has been opened, is practically shadowed by the pressure oscillations
when the vessel has been initially full. Differences in the importance of the delay
in boiling have been also observed in the larger vessels even with similar pre-relief
conditions. These differences have been attributed to the large mass of liquid contained
in the large size vessels.
Four very different numerical codes have been used for the theoretical calculations.
These codes have been either one-dimensional in the vessel (RELIEF) or in the ventline (SAFIRE) or both in the vessel and in the vent-line (RELAP & D E E R S ) . It has
been evident in all the examples t h a t for realistic representation of the the hydrodynamic processes, at least the vessel should be modelled with several control volumes.
Better representation of the parameters referring to the vent-line has been achieved
from one-dimensional calculations which, in addition to the vessel, discretised also the
vent-line in the axial direction. This unfortunately increases the computational time
and the complexity of the code. Thus, the answer to the dilemma of employing a
detailed one-dimensional code which uses greater computational time, against the use
of a fast and simple code with a certain number of assumptions, lies on how accurate
one needs to represent the phenomena occurring during an emergency relief.
Although good representation of the depressurisation processes in the vessel has
been achieved with the one-dimensional calculations shown here, the use of the same
two-phase flow model for the vessel as well as for the vent-line and for the whole range
of void fractions from 0 (all liquid flow) to 1 (all vapour flow) may not always be a
good approach. The results in the bottom venting example as well as for fluids other
than water demonstrated t h a t a more phenonomenological representation of the flow
might be necessary. Unfortunately, these models have the disadvantage of introducing
empirical parameters which then require a certain degree of tuning. Eventually, this
438
makes the numerical mode] user-dependant and the overall predictive capabilities of
the code will always remain doubtful.
Finally, from the examples shown here it can be concluded t h a t by measuring only
the pressure and the remaining mass in the vessel, not all of the occurring phenomena
can be identified. The mass-flux and the vent-line quality disclose some important
information whenever a detailed understanding of the depressurisation processes is
necessary. This detailed understanding seems to be necessary for scaling to larger size
units and for an accurate input (source term) to problems of atmospheric dispersion.
ACKNO WLEDGEMENTS
The author thanks Messrs. A.Benuzzi, S.Duffield, G.Franchello, G.Friz, H.Staedtke
and K.Bell who provided the calculations with RELAP, SAFIRE and RELIEF and
contributed significantly to the evaluations of the present examples.
REFERENCES
1
3
4
5
9
10
439
11
12
13
14
15
16
17
18
19
Nomenclature
Ad
D
dh
H
9
J
Re
C/oo
U pool
U
X
a
p
a
9
J
L
M
m
n
U
2p
void fraction
dynamic viscosity (Pa s)
density (Kg m - 3 )
surface tension (N m _ 1 )
Sub/Superscript
vapour phase
drift quantity
Node below current node
liquid phase
Maximum value
0.4 eqn.[ll]
0.3 eqn.jll]
Node above current node
two phase
EURR1
ISBN 0-7923-1233-3