Vous êtes sur la page 1sur 6

Optimum Operation of an Adaptive Strategy for pH Control

in Semi-batch Reactors
P K Sarkar, Associate Member
S J Sahu, Non-member
An adaptive control strategy has been proposed for handling a class of commercially important batch or
semi-batch processes, having dead time in dynamics and non-linearity in their process characteristics.
The scheme implemented a pulsed addition of buffer solution as the manipulated variable, and the
pulse width was modulated as a linear function of error. The simulation of the scheme produced results
that were in good agreement with the experimental data obtained from a laboratory reactor set up. A
study on the simulated reactor control system has been carried out to find the optimum controller
parameters, for achieving minimum value of ultimate error. The surface generated predicts the
optimum point of operation.
Keywords : Semi-batch reactor; pH control; Non-linearity; On-off control; Pulse width modulation; Gain scheduling

NOT
ATION
NOTA

CB

buffer concentration

ER

error = (SP reactor pH)

re-circulation rate

QB

buffer flow rate

TOF

off time period

TON

on time period

reactor volume

mixing model parameters

V1, V2, Q1,


Q 2, K 1, K 2,
M1, M 2, D
INTRODUCTION
Control of pH, whether in a continuous or semi-batch
reactor system, have drawn considerable attention from
researchers as well as practicing engineers. In the
synthesis of numerous chemical and biochemical
products, successful control of pH often determines the
product quality.
It has been recognized that control of pH is difficult when
applied to, practically, any acid-alkali pair. The more
important reason for problems in controlling pH may be
attributed to the inherent non-linear dependence of pH on
the alkali/acid mole ratio in the reaction system. This
P
K
Sarkar
is
with
the
Department
of
Chemical
Engineering, Jadavpur University
University,, Kolkata 700032 while S
J Sahu is with the Department of Chemical Engineering,
Haldia Institute of T
echnology
Technology
echnology,, Haldia 721657.
This paper (modified) was received on April 16, 2007. Written
discussion on this paper will be entertained upto November 30,
2007.

Vol 88, September 2007

produces a non-linear process gain characteristics.


Several workers15 have addressed this problem. Among
these works, several attempts were made to develop
special functions to represent the process non-linearity
and to synthesize an analytical form of the control
function48. Probably, the modification proposed by
Shinskey3,9 of the final control element, by using two
control valves of dissimilar sizes where the larger size
valve would be used for very high or very low pH range
(low process gain), and the smaller size valve to be used in
pH range close to the neutral point (high process gain), is
the more practical approach to the solution.
In practical application, the pH sensor is put into a part of
the system where the hydro-dynamic forces are relatively
low, to ensure the mechanical stability of the glass
electrode that has a bulb made of very thin wall thickness.
Often the sensor is installed in a re-circulation pipeline.
This arrangement includes a delay or dead time element
in the process dynamics.
A mechanically stirred batch reactor, in which a
neutralization reaction takes place by external addition of
buffer solution, should be called a 'semi-batch' reactor
system. The buffer addition, in fact is the manipulation or
control action executed by the pH controller. So by nature
of this semi-batch operation, any thing that goes into the
reactor stays within. Dynamically, this character adds an
integrator element into the process. In practical terms,
any wrong control action may induce an irreversible and
undesirable change of the product quality.
In face of these special process characteristics, even PID
control renders a poor closed loop performance. An
adaptive control algorithm has been developed that is
suitable for such control applications. As a class name,
this algorithm may be denoted as 'gain scheduling type
pulse width modulation mode' of control7,8,10.
21

MA
TERIALS AND METHODS
MATERIALS
A laboratory scale semi-batch reactor (2.47 l) set-up has
been fabricated with features like re-circulation line,
having conductivity or pH-sensor insertion facility,
mechanical stirrer, electrolytic tracer or buffer solution
reservoir (0.7 l) with delivery line and solenoid valve for
controlled addition.
The configuration of this set-up that has been used for
obtaining the reactor mixing dynamics is shown in
Figure 1(a). In this configuration, the tracer is a 0.24 g/ml
KCl solution.
The configuration of this set-up that has been used for
obtaining reactor control dynamics is shown in
Figure 1(b). In this configuration, the buffer is a 0.5 mol/l
HCl solution, the initial concentration of reactor fluid is
0.005 mol/l NaOH solution for each experimental run.

Tracer
Solution

The adaptive controller has been indigenously made by


using CMOS integrated chips and a fixed frequency
crystal. The cylindrical buffer solution reservoir has a
diameter that is large enough, so the level difference
between the buffer solution and the reactor liquid surface,
changed less than 0.1% during the experimental run.
This fact ensures constant level in the buffer reservoir,
and constant buffer flow during any ON time period.
The controller is designed to produce a pulse train11,12.
The high and low state of the pulses correspond to fully
ON and OFF state of the final control element (FCE). The
FCE is a solenoid valve in this system, that is driven by
the controller through a current amplifier. During the ON
time period, valve is fully open, buffer from an overhead
storage flows down into the reactor, but during the OFF
period, no buffer addition takes place, and the continuous
stirring plus re-circulation gradually brings up
Buffer
Solution

Single shot pulse


generator

Solution Vlv

Set
point

Solution Vlv

Adaptive
controller
pH Snsr
trnsmtr

Cond.
snsr
Amplfr.

Recrdr
Reactor

Reactor

Pump

Pump
Experimental parameters
Reactor volume = 2.47 l
Re-circulation rate = 200 ml/s

Experimental parameters

Initial alkali concn. = 0.1 mol/l

Reactor volume = 2.47 l

Buffer flow rate = 38 ml/min

Re-circulation rate = 200 ml/s

Buffer concn. = 5 mol/l

Tracer concn. = 0.24 g/ml

Off time period, TOF = 25s

Tracer pulse width = 0.2s

Initial on time period = 10s

(a)
Figure 1 (a) The
reactor
control

22

set-up

configuration

(b)
for

the

study

of

mixing

dynamics,

(b)

the

set-up

configuration

for

study

of

IE(I) Journal-CH

homogeneity in concentration in the reactor. This feature


takes care of the dead-time and integrating character of
the system.

pH =

A pump drives the reactor liquid through the recirculation line. The pH sensor situated on the recirculation line sends the process variable signal to the
controller, where the error is calculated as the difference
between set point and the measured pH. The control
action is realized by reducing the period of next ON pulse.
The error calculation and ON time period manipulation
only happens after completion of each window time period
(Tw ), where Tw is defined as Tw = TON + TOF. Thus,
gradually the ON pulse widths are reduced so that lesser
amount of buffer is added in each ON period, as the
system comes closer to the set point. The required gain
scheduling is obtained by the modulation (reduction) of
ON pulse period as a function of the error value.

A + B ( MR )

(1)

1 + C ( MR ) + D ( MR )2

For, 0.1 < MR < 1.0

For, 1.0 < MR < 10.0

A = 2.3242683

A = 2.59139x108

B = 1.4620537

B = 55127196.0

C = 0.21593513

C = 40356205.0

D = 0.66249482

D = 438541.42

To represent the hydrodynamics of the reactor, several


experiments have been carried out with KCl solution as
an electrolytic tracer. Six number of non-ideal mixing
models from the literature13-20 have been tried to fit the
mixing dynamics data.
A simulation program MIXMOD in QBasic has been
developed for generating mixing dynamics with the above
model configurations (modified for semi-batch format) to
compare with the experimental data for similar
operational parameters. The best results have been
obtained for the Chollete-Cloutier14 model.

EXPERIMENTS AND SIMULA


TION
SIMULATION
The pH values corresponding to alkali / acid mole ratio
(MR) within the range 0.1 to 10 are experimentally
obtained. The dependence of pH on the mole ratio (MR)
has been expressed by the developed two rational function
type empirical equations for the alkaline (10 < MR < 1.0)
and acidic (1.0 < MR < 0.1) regions, by using non-linear
regression analysis through the software CVEXP 1.2.
Standard deviation obtained for this curve fit is around
0.1. The result is given by the following equation.

The configurations of the original model and the modified


form of this model for semi-batch application are shown in
Figure 2(a) and 2(b), respectively. In the semi-batch form
of the model, the dead time element has become a part of
the re-circulation path.

Stirred pot 1

Delay

Stirred pot 2
(a)

The

Chollette'-Cloutier

model

for

non-ideally

Parameter values used in


simulation programmes
Reactor volume = v = 2.47 l
Recircln. rate = Q = 200 ml/s

pot

Tracer/buffer
application
M2 > M1

K1 = 0.3; K2 = 1K1
V1 = K1V; V2 = K2V

K1

Q1 = K2Q; Q2 = K1Q

Q2 < Q1
M1
Ideal mixer 1

Q1

M1 = V1/Q1 = 5.37 s

M2 = V2/Q2 = 29.17 s
D = 10.0 s

mixed

K2
Q
D

(b)

M2
Ideal mixer 2

Q2

Re-circulation line configures the delay


Reconfigured

Chollette'-Cloutier

model

for

semi-batch

process

Figure 2 The non-ideal mixing model for dynamic simulation of the semi-batch reactor

Vol 88, September 2007

23

A sample of the experimental and simulated concentration


traces have been shown in Figure 3. The modified
configuration of Figure 2(b) has been used for simulation
of mixing as well as reactor modelling.

The experimental and simulation results for nine


different parameter combinations have been presented in
Table 1.
To search this area more closely to obtain the parameter
combination for any lesser error value than obtained in
Table 1, a program BCHRCT-2 in QBasic has been
developed for generating the error response surface
against the two parameters TOF and CB. The search has
been carried out in the similar range as in Table 1, that is,
for TOF from 20s to 30s, and for CB from 0.4 mol/l to 0.6 mol/
l with 20 grid divisions in each parameter axes. The error
response has been shown in Figure 5, and this 3-D surface
shows a minimum error EMIN = 2.862103,
corresponding to TOF = 23 s and CB = 0.56 mol/l.

A simulation program BCHRCT-1 in QBasic has been


developed for generating closed loop control dynamics in
which the modified configuration of the Chollete-Cloutier
model has been used to represent the reactor
hydrodynamics with the re-circulation path for semibatch system format. The empirical equation (1) has been
used in this program to compute pH from the MR value
obtained at the end of re-circulation line at any instant. A
sample of the closed loop dynamics for controlled variable
(pH), set point (SP = 7), and controller output as the pulse
train, demonstrating the constant OFF-periods, and
modulated ON-time periods has been shown in Figure 4.

DISCUSSION AND CONCLUSION


Any measurements within the window time would convey
high fluctuations of pH during the non-homogenous state
of reactor concentration. Such measurements, if used for
control action determination will result in fluctuating
values of ON time and deterioration of control quality.
Also, near small values of error, such a strategy may
produce rapid opening and closing of the solenoid valve
(Chatter), which is highly undesirable.

Concentration,

g/ml

A number of experimental runs have been conducted for


different combination of TOF and CB values. Some of the
values have been compiled, around the parameter
combination for which minimum error has been obtained.
1.2

Experimental data Experimental parameters


Simulation result Reactor volume = 2.47 l
Recirculn. rate = 200 ml/s
Tracer concn. = 0.24 g/ml
Tracer pulse width = 0.2s

0.08
0.04

Figure
mixing

0
0

Tracer injection

25
Time, s

3 The experimental
dynamics

Experimental
parameters

and

It is not surprising that the mixing model proposed by


Cholette and Cloutier produced a satisfactory agreement
with the experimental result, with respect to dynamic
traces as well as the mixing time ( 25s), as this model
has proved its usefulness for stirred vessel hydrodynamic
applications13,14,20.

50
simulated

traces

of

The most significant parameters with respect to the


practical implementation of this control scheme have been
Reactor volume = 2.47 l

Initial alkali concn. = 0.005 mol/l

Re-circulation rate = 200 ml/s

Buffer flow rate = 38 ml/min

Buffer concentration = 0.5 mol/l

Off time period, TOF = 25 s

Initial on time period = 10 s


Results

Reaction time = 508.02 s; final pH = 6.975; error = 0.025

Manipulation

Controlled variable
(reactor liquid pH)
Set point (pH = 7)

500

1000

Time, s
Figure 4 Closed loop dynamics of the adaptive pH-control system

24

IE(I) Journal-CH

Table 1 Experimental
semi-batch
laboratory
Run No

and simulation
reactor

results

on

the

CB

TOF

TR

Set

pH

ER Remarks

mol/l

point

(final)

0.4

20.0

693.6

7.0

0.4

20.0

773.4

7.0

0.4

25.0

657.0

7.0

6.9

0.1

Exptl.

0.4

25.0

823

7.0

6.93

0.07

Simln.

0.4

30.0

722.2

7.0

6.8

0.2

Exptl.

0.4

30.0

703.6

7.0

6.84

0.16

Simln.

0.5

20.0

590.6

7.0

6.65

0.35

Exptl.

0.5

20.0

638.5

7.0

0.5

25.0

520.8

7.0

0.5

25.0

664.4

7.0

6.975 0.025

Simln.

0.5

30.0

550.0

7.0

6.69

0.31

Exptl.

0.5

30.0

538.6

7.0

6.552 0.448

Simln.

0.6

20.0

520.6

7.0

0.6

20.0

568.9

7.0

0.6

25.0

442.4

7.0

0.6

25.0

598.7

7.0

0.6

30.0

480.8

7.0

0.6

30.0

438.0

7.0

6.8

0.2

6.755 0.245

Exptl.
Simln.

6.5577 0.4423 Simln.


6.9

0.1

6.8

0.2

6.4012 0.5988
6.85

0.15

6.8848 0.1152
6.62

0.38

Exptl.

Exptl.
Simln
Exptl.
Simln.
Exptl.

6.6315 0.3685 Simln.

ER
Search range TOF
: 20s to 30s CB :
0.4mol/l to 0.6 mol/l

Results
Minimum error value : 0.002862
For TOF=23.0s, and CB=0.56 mol/l

be the manifestation of the slight difference between


simulated and actual mixing models, as they are allowed
to approach equilibrium at higher TOF values.
For the remaining runs, experimental TR obtained with
lower TOF values, are lower than the simulation result.
This may be attributed to the fact that the
electromagnetic solenoid valve has failed to respond for
very small ON time pulses due to its slower speed of
response. It may be expected that simulated responses
would show lower error ER values. For low TOF values,
the process variable fluctuations still present in the
system may lead to an extra addition of buffer, leading to
higher error. (run number 1, 4, 7).
Also hydrodynamic forces may affect the ionic dissociation
phenomena21 that would give rise to a noise in the
measured pH values. This fluctuation would increase in
amplitude near neutral point (pH = 7) due to the highest
values of the process gain in the vicinity of pH = 7, and
may result in an anomalous error value (run number 6).
The parameter values corresponding to the minimum
error point obtained from the smaller interval grid search
are in fair agreement with those obtained for run number 5
in Table 1, which corroborates with the mixing time
obtained in the mixing dynamics experiment ( 25 s) and
agrees with the shape of the surface along TOF axis.
However, the surface in Figure 5, shows a plato extended
in CB axis at least towards lower values. This may
represent the underfed condition that produces no
appreciable reduction in ER, at the expense of higher TR
values.
REFERENCES
1. H H Y Chin and R Aris. 'An Adaptive Control of the Batch
Reactor', Automatica, vol 2, no 41, 1964.

CB

TOF

2. R Aris and N Blakemore. 'Studies in Optimization the BangBang Control of a Batch Reactor'. Chem. Engg. Sci., vol 17, 1962.
3. F G Shinsky. 'Control for Non-linear Processes'. Chem. Engg.,
vol 19, March, 1962.
Figure 5
controller

The error response


parameters

surface

to

find

optimum

recognized as, the TOF and mass flow rate of buffer


material. Both of the parameters can influence the total
reaction time or batch time, which determines the
utilization efficiency or the production capacity of the
process equipment. In an industry, the later parameter is
usually manipulated by adjusting the buffer flow rate, by
selecting the line and solenoid valve size of buffer flow line.
In the laboratory, to study this effect, the buffer solution
concentration has been changed, for convenience. Thus,
an experimental routine has been conducted to investigate
the influence of these two parameters on error and
reaction time.
Run number 3, 6 and 9 show a higher experimental
reaction time (TR) than the simulated values. This may

Vol 88, September 2007

4. A J F Hodgson and D W Clarke. 'Self Tuning Control Applied to


Batch Reactors'. IEEE Conference on Applied Adaptive
Multivariable Control, Hull, England, 1982, p 146.
5. G D Cawthon and K S Knaebel. 'Optimization of Semi-batch
Polymerization Reactions'. Comp. & Chem. Engg., vol 13, 1989.
6. L Lapidus and N R Amundson. 'Chemical Reactor Theory- A
Review'. Prentice Hall Publishing Co Ltd, New York, 1984.
7. F Buchholt and M Kummel. 'Self Tuning Control of a pH
Neutralising Process'. Automatica, vol 15, 1979.
8. S Bergmann and K H Lachmann. 'Digital Parameter Adaptive
Control of a pH Process'. Proceedings of the JACC, San Fransisco,
1980, Paper TA2-E.
9. F G Shinsky. 'Process Control Systems'. McGraw Hill, New
York, 2nd Ed. 1979.
10. O L R Jacobs, et al. 'On Controlling pH Process Control'. IEE,
London 1985, p 643.

25

11. P K Sarkar, D Mukhopadhay and B Rakshit. 'Development of


an Adaptive Control for Batch Processes'. Proceedings of the
National
Symposium
on
Current
Trends
in
Process
Instrumentation and Control '. Kalpakkam, Tamil Nadu, October
1988.
12. P K Sarkar, D Mukhopadhay, B Rakshit and T Sen. 'Studies on
an Adaptive Controller Module for Semi-Batch Nonlinear Process
with Dead Time'. Journal of the Institution of (India), vol 72,
October, 1991.

15. R J Adler and R B Hovorka. 2nd JACC, Prep 3, Denver, 1961.


16. Van de Vusse. Chemical Engineering Science, vol 17, 1962.
17. R S C Rogers and R P Gardner. AIChEJ, vol 25, 1979.
18. M Rubinovitch and U Mann. Chemical Engineering Science,
vol 34, 1979.
19. D M Himmelblau. 'Process Analysis by Statistical Methods'.
John Wiley, NY, 1970.

13. C Y Wen and L T Fan. 'Models for Flow Systems and Chemical
Reactors'. Marcel Dekker, N Y, 1975.

20. P K Sarkar, D Mukhopadhay and B Rakshit. 'Optimization


Routine for Dynamic Simulation of Mixing Models'. DechemaMonographs, vol 116 - VCH, Verlagsgesellschaft, 1989.

14. A Chollette and L Cloutier, Canadian Journal of Chemical


Engineers, vol 37, 1959.

21. S Glasstone. 'Textbook of Physical Chemistry'. 2nd Ed.


Macmillan and Co. Ltd, London, 1962.

26

IE(I) Journal-CH

Vous aimerez peut-être aussi