Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

Contents lists available at ScienceDirect

Journal of Loss Prevention in the Process Industries

journal homepage: www.elsevier.com/locate/jlp

Safety-based process plant layout using genetic algorithm

Antonio C. Caputo a, *, Pacico M. Pelagagge b, Mario Palumbo b, Paolo Salini b
a
b

Department of Engineering, University of Roma Tre, Via della Vasca Navale 79, 60100 Roma, Italy
Department of Industrial Engineering, Information and Economics, University of L'Aquila, Zona industriale di Pile, 67100 L'Aquila, Italy

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 19 October 2014
Received in revised form
17 January 2015
Accepted 18 January 2015
Available online 19 January 2015

This paper presents a method based on a genetic algorithm for optimizing process plant layout. The
relative location of main process units is determined to minimize an annual cost function including the
cost of material transfer between process units (piping and pumping costs), land cost, and the expected
annual loss resulting from damage to each secondary unit caused by primary accidents occurring in
nearby process units. This method is an improvement over previous attempts using genetic algorithms or
mathematical programming techniques to optimize plant layout, which neglected pumping costs and
included safety issues by evaluating the infringement of predened safety distances only. In this
approach the operating cost of material transfer is included and the likelihood of accidents is taken into
account thus providing good practical solutions to the plant layout problem incorporating more realistic
cost functions and constraints. In the paper, after discussing the structure of the annual cost function and
describing the working logic of the layout generating algorithm, a case study is described to demonstrate
the effectiveness of the proposed methodology.
2015 Elsevier Ltd. All rights reserved.

Keywords:
Process plant safety
Genetic algorithm
Plant layout
Risk reduction
Layout optimization

1. Introduction
The process plant layout problem consists in dening the
optimal spatial arrangement of a set of facilities and the required
connections among them within a plant site. The location of process units and plant facilities is chosen in order to reduce land use
and the costs of the piping interconnecting each units pair, as well
as to organize more efciently the production and increase the
plant safety. Such requirements frequently give rise to conicting
goals. If, on the one hand, material transfer cost and occupied area
are reduced by placing interrelated units at a short distance, on the
other hand safety concerns ask for minimum safety distances between process units to be maintained. Process plant layout design is
an activity usually carried out by human designers (Mecklenburgh,
1985). Computer-aided layout planning methods were instead
scarcely adopted in the past in this sector owing to the difculty in
minimizing a number of different objective functions simultaneously in a realistic manner. Nevertheless, there is a growing need
to develop computer-aided methods to support process plant engineers in the rapid generation of alternative chemical plants

* Corresponding author.
E-mail addresses: acaputo@uniroma3.it (A.C. Caputo), pacico.pelagagge@
univaq.it (P.M. Pelagagge), mario.palumbo@univaq.it (M. Palumbo), paolo.salini@
univaq.it (P. Salini).
http://dx.doi.org/10.1016/j.jlp.2015.01.021
0950-4230/ 2015 Elsevier Ltd. All rights reserved.

layouts. Heuristic techniques were suggested at rst (Amorese

et al., 1977; Gunn and Al-Asadi, 1987; Suzuki et al., 1991), offering
computational efciency but no guarantee on the optimality of the
solution. Jayakumar and Reklaitis (1994) in alternative formulated a
graph partitioning problem to allocate equipment to sections
created by aisles or corridors. These attempts were followed by
mathematical programming approaches aimed at reaching optimal
solutions. Mixed integer linear programming formulations were
developed by Georgiadis et al. (1999) to allocate items to candidate
locations onto a grid or, in the continuous domain, for single or
multiple-oor problems by Papageorgiou and Rotstein (1998),
Patsiatzis and Papageorgiou (2002a, 2003), Ozyruth and Realff
(1999), Barbosa-Povoa et al. (2001).
Attempts to incorporate safety issues in computer-aided layout
optimization procedures have been also carried out adopting
heuristic, mathematical programming and stochastic optimization
techniques. Fuchino et al. (1997) adopt a heuristic approach where
the equipment modules are divided into groups and then subarranged within groups according to safety requirements
following a permutation and partition method. Suzuki et al. (1991)
rely on a set of heuristics supporting a facilities interchange procedure based on a cost function calculated from unit separation
distances. Georgiadis and Macchietto (1997) analyze the layout of
multioor production facility of equally sized units. Penteado and
Ciric (1996) adopt a mixed integer non-linear programming

140

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

technique to optimize cost and safety of circular or elliptical process

units layout. In that approach, the cost of a layout is a function of
piping cost, land cost, nancial risk, and protection devices cost.
The nancial risk term captures the risk of unsafe plants and can be
expressed as the expected losses if major accidents happen (i.e.,
res or explosions). Patsiatzis and Papageorgiou (2002b) suggest a
MINLP model by adopting rectangular shapes and rectilinear distances, including risk related to accidents propagating from a
source to a target utilizing the equivalent TNT method. Patsiatzis
et al. (2004) extend previous continuous domain process plant
layout mathematical models to include safety aspects and economic loss by utilizing the Dow Fire and Explosion index. Castell
et al. (1998) consider a process layout including oriented rectangular facilities and adopt a non differentiable objective function,
utilizing Genetic Algorithms (GA) for its optimization. In their
approach safety is modeled by the Mond re and safety index
which species the preferred minimum distance between process
units. The extent to which these constraints are violated determines, although not explicitly, the safety cost of the layout.
Franceira et al. (2007) use a quadratic assignment problem
formulation to minimize a total cost function including land,
piping, pumping. In this case too, safety is not accounted for in an
explicit manner, but only as a penalty in case of violation of predetermined separation distances, neglecting actual physical effects
and equipment losses. Jung et al. (2011) adopt a MINLP approach
together with Probit method to estimate probability of damage to
equipment to be factored as a potential structural damage cost in
the overall cost function including land cost and interconnection
cost. Park et al. (2011) use again a MILP formulation to optimize
plant layout in a multi-oor facility including land and oors area
cost, piping and pumping costs, as well as expected loss caused by
equipment explosions in the objective function. They use Probit
function and TNT method to assess the probability of equipment

n et al., 2010; Garcadamage. Some authors (Vazquez-Roma
nchez et al., 2013; Vazquez-Roma

n and Mannan, 2010) suggest
Sa
a MINLP layout cost minimization model associated to a probabilistic analysis performed by Monte Carlo simulation to account for
uncertainties in accident scenarios consequent to gas releases.
However, they focus on adding new equipment to an existing facility instead of building a layout from scratch. Moreover, they
minimize a cost function including piping cost, land cost and fatalities compensations cost. Therefore, they include safety issues
only with reference to fatalities caused by toxic chemicals release,
while no equipment loss is accounted for. Xu et al. (2013) extend

n et al. (2010) by using a hybrid
the approach of Vazquez-Roma
Genetic Algorithm to solve their original MINLP formulation.
Furthermore they account for safety issues as a violation of predetermined safety distances and as a violation of distances of safe
concentration levels from gas leakage computed resorting to a
PasquilleGifford dispersion model. However, in this case too only
piping cost, land cost and fatalities cost is included. Even other
Authors focus on layout optimization in case of toxic releases
attempting to locate process units respect inhabited buildings, but
do not consider equipment loss (Daz-Ovalle et al., 2010, 2013; Jung
et al., 2010), or design layout with the aim of reducing individual
risk by imposing additional constraints on safety distance in a
traditional MILP formulation (Han et al., 2013a, b).
Overall, the previous approaches, exception made for Park et al.
(2011), always neglect operating cost of the pipe network related to
overcoming friction losses, and in a large majority only include
safety costs either indirectly or, when explicitly, in a substantially
deterministic manner, such as in terms of violation of minimum
safety distances.
Therefore, it can be concluded that although in recent time
the problem of optimal process plant layout has attracted the

attention of a number of researchers, there is still a need for an

approach combining process plant layout planning and detailed
risk assessment. With this aim a genetic algorithm based optimization approach to the plant layout problem in the continuous
domain is presented in this work which includes in a realistic
manner all relevant costs, including land use, piping investment
and operating expenses, as well as economic loss from potential
accidents evaluated in a probabilistic manner. The choice of
adopting a genetic algorithm derives from the intrinsic exibility
that this approach allows in dening and computing a realistic
objective function (which can even be non-differentiable),
removing the constraints of linearity or the limitations of a
mathematical programming solution method. Moreover, it
should be pointed out that while GAs are an established layout
design method for manufacturing facilities (Liggett, 2000; Singh
and Sharma, 2006; Drira et al., 2007) they have almost never
been applied to process plant layout design problems, as most
scholars preferred to adopt mathematical programming formulations (i.e. MILP or MINLP). To the best of our knowledge, in fact,
only Castell et al. (1998) and Xu et al. (2013) declare utilization of
GAs. However, Castell et al. (1998) did not provide any GA detail,
while Xu et al. (2013) use GA to solve a MINLP model in a more
computationally efcient manner. In both cases the objective
function is not inclusive of all relevant terms discussed above,
and the manner in which safety issues are considered (i.e.
infringement of Mond safety distances, or expected fatalities
caused by toxic gas release neglecting equipment loss) may be
neither exhaustive nor realistic enough. Therefore, as an additional motivation gap, an opportunity arises to explore the
effectiveness of GAs in a more realistic risk-based layout planning
scenario.
In the paper, the layout problem is formulated at rst, and an
objective cost function is derived. Then the encoding of analytical
formulation into the genetic algorithm structure is described
as well as the solutions adopted to comply with the layout
problem constraints. An industrial case study concludes the
paper by exemplifying the method application and its
capabilities.
2. Layout problem formulation
According to Mecklenburgh (1985) a good layout is associated to
low pipework cost, a small plant area, and a safe design. To account
for these requirements layout optimization is here carried out by
minimizing the objective function of (Eq. (1)) representing the
layout-related total annual cost of the plant TC

TC PIC POC EAL LC

(1)

where PIC is the equivalent annual piping investment cost, POC is

the annual pumping-related operating cost for overcoming friction
losses in the piping, EAL is the expected annual loss from accidents,
and LC is the equivalent annual land cost. All costs are expressed as
annual costs (V/yr) and computed as indicated below.
2.1. Piping cost
To compute piping investment cost it is assumed that all pipes
run from the centre point of a process unit and that the pipe length
between two interconnected units i and j, having centre points
coordinates (xi, yi) and (xj, yj) respectively, is the rectangular
distance

Lij
xj  xi

yj  yi

(2)

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

It is also assumed that the cost per unit length CUPij (V/m) of the
pipe interconnecting units i and j is known from a preliminary
sizing which enabled the designer to set the technical specication
of each pipe line, including pipe diameter Dij, material, and wall
thickness.
The overall equivalent annual cost of piping investment is
therefore

PIC

t XN XN
d C L
j1;jsi ij UPij ij
2 i1

(3)

where factor avoids counting two times the same pipe line, N is
the number of facilities to be allocated, and t is the capital recovery
factor t [s(1 s)T]/[(1 s)T  1], being T the plant life span
(years) and s the interest rate (%/year), while dij represents the
connectivity matrix and is dened as follows: dij 1 when unit i
and j are connected, dij 0 when units are not connected. It should
be noted that when multiple pipe lines connect the same units, as
happens for instance in case of heat exchangers for cold and hot
stream, the cost per unit length has been multiplied by a factor
equal to the number of separate pipe lines connecting the process
units.
2.2. Pumping costs
The annual energy cost for overcoming friction losses is
computed as follows. At rst the pressure drop Dpij (Pa) in the pipe
connecting units i and j is computed knowing the uid density rij
(kg/m3) and the ow velocity vij (m/s)

vij

Qij 4

(4)

pD2ij

Dpij

rij v2ij Lij

f
2 Dij

(5)

where Qij is the ow rate (m3/s) between units i and j, while f is the
friction factor. The corresponding pumping power Pij (W) is

Pij

Qij Dpij
h

(6)

POC


CE H XN XN
dij Pij
i1
j1;jsi
2

(7)

being CE (V/Wh) the energy cost, and H the annual operating hours
(hr/yr).
2.3. Safety cost
The expected annual loss is computed as the summation of the
expected monetary loss related to each process unit or facility in
the plant

EAL

XN
i1

The expected value of production interruption is not included in

this model, as it is case specic and may even occur in case of
damage of a single equipment irrespective of layout design. In Eq.
(9) pCD,i is the cumulative annual probability that the considered
unit has of being damaged owing to loss of containment or mechanical failure, EVi is the economic property damage value of the ith process unit, CFi is a credit factor (usually set at 1, but can assume
any user specied value in the [0,1] range) which accounts for any
safety measure or protection device able to limit the actual property damage cost to the unit. The cost of such additional safety
measure can be included in equipment value EVi. NPi is the average
number of people standing by the equipment for operational and
maintenance purpose, PPi is the probability that people actually
stay near the equipment when the accident occurs (this depends on
work shift organization and the needs for local manned supervision), PCDPi is the operator's overall fatality probability given the
consequence intensity at location of the i-th process unit, while FL
is the economic value of a fatality. Usually this is assumed as about
2  106 $/fatality according to the statistical value of life approach
(Caputo et al., 2011).
In Eq. (9) the term pCD,i is evaluated as a probability complementary to that of not being damaged by any other process unit in
the plant, added with probability pA,i that the i-th process units is
damaged by an internal failure unrelated to accidents occurring to
other equipment

2
pCD;i

41  @

13

Y
1  pAD;ij A5 pA;i

EALi

(8)

The expected annual loss for the generic i-th process unit (EALi)
is computed as the sum of expected monetary loss due to damage
to the equipment, and the expected loss due to fatalities of people
involved in equipment operation and maintenance.

EALi pCD;i CFi EVi NPi PPi PCDPi FL

(9)

(10)

jsi

While equipment internal failure is not relevant as far as layout

design is concerned, it is included for completeness, as it contributes to the absolute value of expected annual loss. The probability
that the i-th unit has of being actually damaged by a generic other
unit j is

pAD; ij pAj pd; ij

(11)

where pAj is the probability of an accident occurring in unit j, while

the probability pd,ij of damage to unit i caused by unit j, is computed
resorting to a Probit model of the kind

Y k1 k2 lnD

being h the pump mechanical efciency. The total annual operating

cost for pumping losses POC is

141

(12)

where Probit coefcients k1, k2, and the dose D computed at the
actual rectilinear distance between units i and j, are dened according to the type of accidental event (Lees, 1996; Van den Bosh
and Weterings, 1997). From the Probit value Y then the corresponding probability of damage can be easily obtained. As an
example, in the case of damage caused by overpressure, Cozzani
and Salzano (2004) suggest the following Probit equations for
different kinds of target process units, where the dose is expressed
in terms of peak overpressure DP

atmospheric vessels : Y 18:96 2:44 lnDP

(13)

pressurised vessels : Y 42:44 4:33 lnDP

(14)

elongated equipment : Y 28:07 3:16 lnDP

(15)

small equipment Y 17:79 2:18 lnDP

(16)

If accident modes other than explosion may occur in the process

units (i.e. pool re, jet re, reball, ash re, mechanical explosion,
BLEVE, vapor cloud explosion) the same analysis is carried out each
time referring to a different class of initiating event. In this case for

142

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

each k-th class of initiating event occurring at the j-th unit and
affecting the i-th unit the damage probability pAD,ijk is computed as
above described, and the overall value of pCD,i follows as

pCD;i 41  @

Y Y
jsi

13

X
1  pAD;ijk A5
pA;ik

(17)

As far as the human fatality probability when the operator is

working nearby the i-th process unit is concerned (PCDPi), it can be
computed according to Eqs. (10), (11) and (17), but term pd,ij should
be computed using Probit equations representing injuries to
humans (Van den Bosh and Weterings, 1997) such as

death from lung hemorrhage Y 77:1 6:91 lnDP

(18)

If a more detailed analysis is required then the damage area

boundary can be computed by resorting to specic consequence
models (Lees, 1996; Van den Bosh and Weterings, 1999; CCPS,
2000) to evaluate the radius where a given threshold value of the
physical effect is obtained (Cozzani et al., 2006).
2.4. Land cost
The annual equivalent cost of the land, which is assumed to have
a rectangular shape enclosing all process units and a unit cost ULC
(V/m2) is

LC tULC



  


wj
wi 
l

yi i
 xj 
2 max
2 min
2 max
 

xi


lj
 yj 
2

min

(19)
being t the capital recovery factor, while wi (and wj) is the length of
a process unit side parallel to the x axis, and li (and lj) is the length of
the side parallel to the y axis.
The facilities locations, chosen randomly by the genetic algorithm, dene the distances Lij between process units as well as the
size of the land area, thus allowing to compute the objective
function value.
3. The genetic algorithm solution method
In discrete and combinatorial optimization problems, or when
non differentiable objective functions occur, stochastic optimization techniques such as Genetic Algorithms (GA) may be successfully utilized (Goldberg, 1989; Davis, 1991). GA have been already
applied to process plant layout problems, where they have generally proved to largely outperform other available heuristics and
metaheuristics. GA is a stochastic global search method that mimics
the process of natural biological evolution. It operates on a population of individuals, each one described as a string composed by
binary genes and representing a candidate solution to the problem,
and applies the principle of survival of the ttest to produce better
performing individuals in subsequent evolutionary generations of
the examined population. At each generation, individuals are
selected according to their level of tness and then are bred
together. This process leads to the creation of individuals better
suited to their environment than their parents. In practice a GA
operates starting with a population of random individuals. Thereafter each string, i.e. an individual of the population, is evaluated to
nd its tness value. Then individuals of the new population are
generated including: the best individual(s) copied from the previous generation (the so called Elite Count); new individuals obtained

by crossover recombination of couples of selected individuals of the

previous generation, where the probability of an individual of being
picked for breeding depends from its tness level; new individuals
randomly generated; mutant individuals obtained by randomly
changing some of their genes; migrant individuals from past generations. The process stops when a convergence criterion establishes that the ttest individual has been found or a maximum
number of iterations has been reached. The minimization problem
in this paper is solved resorting to a GA coded by the authors that,
given a set of N facilities to be allocated onto a plant layout, generates the physical location and orientation of each unit until an
optimal layout is found corresponding to a minimum of the
objective function TC (Eq. (1)). In the genetic representation of this
problem each individual of the population represents a candidate
plant layout arrangement and is coded as a binary string made of N
substrings, each one coding the location of a single process unit in
terms of centre point coordinates (xi, yi) and an additional bit Oi
representing unit orientation. If the orientation bit is 0 the unit has
the longest side parallel to the y axis, if it is 1 the longest side is
parallel to the x axis. The tness level of each individual is inversely
proportional to the total cost associated to the layout represented
by the individual. A major problem when solving layout problems
with GA is that of generating feasible solutions in the population
during the stochastic evolution process and to ensure that the
offspring of two parents representing feasible solutions still is a
feasible solution. In fact, neglecting geometrical constraints
resulting from facilities overlapping or from the assignment of
more than one equipment to the same physical location, the GA
would inevitably generate infeasible individuals and the number of
feasible solutions in a randomly generated population would
rapidly go to zero preventing the solution of the layout problem.
Therefore, infeasible solutions can not be allowed in the population,
as discussed for example by Suresh et al. (1995) and new genetic
operators or repair operators may be required to ensure feasibility of the generated individuals (Al-Hakim, 2000).
In the implemented model the GA chooses the coordinate of
equipment centre points (xi, yi) on a grid of predened width. In
case the grid pitch is chosen smaller than the maximum side length
of the largest process unit it may happen that in a candidate layout
two or more units overlap. In this case the GA discards the unfeasible individual. The same happens when the GA attempts to locate
two or more units in the same grid location.
Overall, to solve the layout problem one must know the units
connection matrix dij, the size of process units (expressed as length
of the two sides li and wi of their rectangular footprint), the probabilities of accidents of each process unit pAj, the economic property damage value of each process unit EVi, the values of Qij, CUPij,
Dij,rij, f, for each pipe connecting two process units, the values of
constants T, s, CE, H, ULC, FL, PPi, NPi. The type and amount of hazardous material inventory at each process unit is also required to
compute accident effects.
After obtaining a new generation, for each candidate individual
the values of distances Lij are computed, the values of Dpij, Pij, PCDi
are computed, the cost functions PIC, POC, LC, EAL are computed,
the value of the objective function TC is computed. Then the individuals for the successive generation are created and the process
is repeated until termination. When the stopping rule is reached
the best t solution (i.e. minimum TC) is chosen which corresponds
to the searched optimal layout expressed as a string of coordinates
of process units centre points and their orientation in the plane.
4. Case study
In order to show the application of the method a case study is
examined. Reference is made to a nitric acid production plant as

described by Ray and Johnston (1989). The plant is intended to

produce 280 t/day of 60% by weight solution of nitric acid. It should
be pointed out that even if this specic plant may not be representative of current process technology, it only acts as an application example for the proposed method which has a general
applicability. The reference plant has been chosen because it is
described in detail in a publicly and easy to nd printed source,
allowing the reader to replicate the case study.
The process scheme is shown in Fig. 1, while plant technical
details can be found in the referenced textbook. A brief description
of the process follows referring to stream numbers and equipment
as indicated in Fig. 1.
The process starts with vaporization (E06) of ammonia (2) using
process heat (70). Steam (65) is then used to superheat (E07)
ammonia (4). Filtered air (21) is compressed in a two-stage axial
compressor (LP-HP). Part of the air (24) is diverted (27) for acid
bleaching, and the balance (25) for ammonia oxidation is circulated
through a jacket on the inner wall of the reactor (R). This preheated
air (26) and the ammonia vapor (5) are then mixed (M) and passed
through the platinum/rhodium catalyst gauze. Ammonia is rapidly
oxidized and the heat of reaction raises the reaction-gas temperature. Reaction gas (6) ows through a series of heat exchangers
(E08, E09, E10, E11) for recovery of energy as either high-pressure
superheated steam (63), or as shaft horsepower from the expansion of hot tail gas (16). This heat exchanger train cools the reaction
gases (6,7). Further cooling is followed by separation of approximately one third acid product (10). The gas is then combined with
bleached air (16) containing additional nitrogen peroxide. This
mixture further oxidizes using additional oxygen supplied from the
secondary air stream (29). It passes through an empty oxidation
vessel (O) and then (8) through the secondary cooler (E12). In
cooling, the gas provides heat to a recirculating warm-water loop
(72). This loop, in turn, provides (70) the energy to evaporate the
liquid ammonia feed (2). The reaction gas mixture (9) entering the
absorber (T1) is almost completely oxidized to nitrogen peroxide. In
the absorber, deionized water (78) is added to the top tray and
weak acid (11) from the low-pressure condenser (E11) is added to a
tray corresponding to its acid strength. Down-owing acid and upowing gases alternately mix as the chemical steps of action formation and nitric oxide oxidation take place with the accompanying release of heat. There are three operational zones in the
absorber: the lower zone cooled with deionized cooling water (54);
the middlezone cooled with chilled water (75); and the upper zone
which is essentially adiabatic. Highly efcient heat removal in the
middle and lower zones is particularly important due to its effect
on the oxidation and dimerization reactions. For this design, chilled
cooling water and make-up water are used. The chilled water (75)
for the absorber is supplied by an absorption refrigeration unit
(E14). Acid from the bottom of the absorber (12) is bleached (T2)
using partially cooled, compressed air (28). The bleach air (29),
containing nitrogen peroxide stripped from the acid, is then added
to the main reaction gas stream at the oxidation vessel (O). The cold
tail gas (15) leaving the absorber (T1) is warmed by heat exchange
(E13) with the hot secondary air stream (27) on route to the
bleaching column. The tail gas (16) is then heated to the expander
inlet temperature by passage through two further exchangers (E10,
E05) in the recovery train. The expander (TGE) recovers part of the
required compressor power. Expanded tail gas (19) then ows
through an economizer (E04, E03) providing heat to the highpressure boiler-feed deionized water (45). Subsequently, tail gas
(20) is exhausted to the atmosphere. The cooling water circuit
provides river water (32) to the steam condenser (E01) and
compressor intercooler (E02).
This plant processes several hazardous substances, namely
ammonia, reaction gases, tail gas and reaction acid. While hazard

143

Fig. 1. Nitric acid process plant scheme.

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

144

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

from release of toxic gas occurs, here we are mainly focused on

damage to equipment. Thus, risk analysis will be limited to chemical and physical explosions only. Effects are estimated according to
the TNT model in terms of overpressure, and the equivalent TNT
mass associated to each process unit, according to the operating
conditions, type of material and contained inventory is shown in
Table 1.
The equivalent TNT mass is computed as

MTNTeq

h  Mmat  DHmat
DHTNT

(20)

assuming an explosion efciency of 10%, and TNT high heating

value DHTNT 4686 kJ/kg, while DHmat and Mmat is the heating
value and mass of the material contained in the process units.
Following a preliminary sizing of process units their economic
value has been estimated as shown in Table 1. The Table also shows
other related data of interest. As far as equipment and piping
construction material is concerned, mild steel is suitable for the
construction of all the process units up to the cooler/condenser,
however, at this stage it is possible for nitric acid to be present in
either liquid or vapor form in the reaction mixture. Stainless steel
grade 304L is used for the cooler/condenser and all subsequent
equipment. With reference to material costs, it was assumed an allcomprehensive cost (material and installation) of 5.00 V/kg for
mild steel and 25.00 V/kg for stainless steel. Pipe internal diameter
has been optimized for each pipe line and ranged from 21 mm to
686, with cost per meter ranging from 18 V/m to 1080 V/m. Please
note that to simplify computations and to allow the GA to converge
to realistic layout solutions, all equipment that for operational or
connection requirements need to be placed close to each other (i.e.

two subsequent compression stages or an equipment and its

ltration unit, a pump and its process unit etc.) have been grouped
in a single compound unit as shown in Table 1.
The following parameters value were also assumed: plant life 10
years; the yearly discount rate s 5%; energy unitary cost
CE 0.1 V/kWh; unitary land cost ULC 15 V/m2, economic value
of a fatality FL 2  106 V/fatality. Plant operation time was
assumed as 6000 h/year, the friction coefcient f 0.02, while
pump efciency as h 0.9. It was assumed that in average one
operator works nearby any process unit for a period of 10e60 min
per day, so that the instantaneous individual probability of being
caught in an accident ranges from 0.7% to 4.2%.
The genetic algorithm has been coded with a 5 5 1 bits
substring structure, thus each individual is coded resorting to
(N  11) bits. Therefore, both the x and y centre points coordinate of
each process unit can assume 25 32 different values. This means
that the layout is basically built on a 32  32 grid and that the
location of a process unit can assume 1024 values (or 2048
including equipment orientation). The grid spacing has been
assumed 10 m. Considering that in the present example 20 major
process units are considered each chromosome is coded by a 220
bits string.
The GA parameters were set as follows: maximum number of
iterations N_MAX 500, crossover probability 0.7, mutation
probability of 0.1, population including 100 individuals. The tness
of each individual is dened as the ratio of the average layout cost
of the entire population to the layout cost of the individual. Fitness
values are then normalized over the [0,1] range for ranking individuals. In the following GA runs the population of 100 individuals at each generation is obtained as follows. The best 4
individuals of a generation are transmitted to the next generation,

Table 1
Process units data.
Equipment
footprint
(length  width
or diameter D)
(m)

Process unit

Layout
equipment
identication
code

Process scheme ID

Equipment
economic
value EV
(V)

Accident
probability
pA (yr1)

Damage
radius
(m)

Inventory
(kg)

Energy
released
in explosion
due to
mechanical
failure (kcal)

TNT
equivalent
mass (kg)

Ammonia Tank
Ammonia Vaporizer
and Superheater
Mixer e Reactor
Expansion-Compression
-Intercooling Unit
Cooling Water
Source/Well
Ion Exchange Unit
Deionized Water Tank
Steam Condenser
Economizer-Deaerator
Unit
Fin Fan Cooler
Steam Superheater
Liquid e Vapor Separator
Tail Gas Warmer
Reaction Gases Secondary
Cooler
Waste Heat Boiler e Tail
Gas Preheater e Cooler
e Condenser e Oxidizer
Unit
Refrigerating Unit
Absorber
Bleaching Column
Nitric Acid Solution Tank
Stack
Superheated Steam Outlet

AT
AVS

TK1 P1
E06 F1 E07 P10

1,460,000
162,000

0.001
0.002

3077
821

898,560
4000

e
181,978

16,866.7
320.4

D 8.00
22

MR
ECI

314,000
6,700,000

0.005
0.005

451
201

500
36

83,623
21,891

53.0
4.7

22
66

CWSW

M R E05
SE TGE LP
E02 HP
P5

0.001

22

IEU
DWT
SC
ED

P4 IEU P6
TK3 P7
E01 P9
P8 E04 D E03

75,000
228,000
60,000
45,000

0.002
0.001
0.002
0.002

e
e
e
549

e
e
e
e

e
e
e
447,669

e
e
e
95.5

42
94
24
25

FFC
SS
LVS
TGW
RGSC

E15
E08
LVS
E13
E12

35,000
35,000
65,000
115,000
396,000

0.005
0.002
0.002
0.002
0.002

68
213
513
240
230

e
e
e
e
e

857
26,361
366,350
37,978
32,195

0.2
5.6
78.2
8.1
6.9

26
1  2.50
32
14
14

BPCO

E09 F E10
E11 O P2

704,000

0.002

395

164,319

35.1

39

R
A
BC
NAST
S
SSO

E14 P11 P12

T1 P3
T2
TK2

259,000
259,000
118,000
1,932,000
25,000
e

0.005
0.002
0.002
0.001
e
e

e
490
210
e
e
e

e
e
e
e
e
e

e
323,563
25,273
e
e
e

e
69.0
5.4
e
e
e

23
D 2.00
D 1.00
D 8.00
D 1.00
11

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

145

Fig. 2. Plot of piping equivalent annual cost.

Fig. 4. Plot of annual pumping cost.

the following best 76 individuals are bred with a 70% probability, 20

new individuals are obtained from a mutation of randomly picked
individuals from the previous generation. The remaining 4 individuals are discarded and substituted with the best 4 individuals
from the tenth previous generation.
In order to show the working process of the GA, a set of simulations will show how the prospective layout changes as the GA
runs. In general, cost reduction was obtained, as expected, through
a progressive reduction of the distance between process units and
recombination of units positions in order to avoid an excessive
increase of expected accident cost, as shown in the sequence of
layouts generated by the GA. Layout is generated over a predened
330  330 m maximum area indicated in light green (in the web
version) background in the following gures. Blue lines (in the web
version) represent pipes interconnecting equipments. Land area
occupied by any layout is represented in light blue background.
However, in order to better appreciate the GA operation, at rst,
each cost item will be examined separately showing how the
optimization process progresses through successive GA
generations.
Fig. 2 shows the progressive reduction of equivalent annual
piping costs during a sample run of the GA. Cost reduction was
obtained through a progressive reduction of the distance between
process units as shown in the sample sequence of layouts generated
by the GA in Fig. 3 for a few successive generations.

It is clear how the GA tries to minimize the inter-equipment

distance in order to reduce piping investment. Moreover, the GA
tries to place close together the units linked by the highest cost per
unit length of piping, while only the units connected by low cost
pipe lines are held at greater distances. Therefore, from a certain
point onwards a further distance reduction does not contribute
signicantly to an additional cost reduction.
Figs. 4 and 5, instead, show the cost reduction and layout evolution when only pumping costs are considered. In this case too,
similar considerations to the previous case hold.
Passing to land cost, optimization results are shown in Figs. 6
and 7. In this case the effectiveness of the optimization process
can be observed in the same clear manner as a progressive reduction of the occupied land area is witnessed during the evolutionary
process.
When, instead, the expected annual loss only is considered, the
results of Figs. 8 and 9 are obtained. In Fig. 9 the circles represent
the damage areas of the process units. In particular, concentric
circles represent, for each process unit, the boundaries where
threshold overpressure values are obtained. In particular the inner
circle represents the area where overpressure is 75 kPa, leading to
probable total destruction of buildings; the second circle represents
the area where overpressure is 50 kPa, leading to nearly complete
destruction of houses, while the external circle represents the area

Fig. 3. Sample layout evolution during GA run (piping investment cost only).

146

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

Fig. 5. Sample layout evolution during GA run (pumping cost only).

Fig. 6. Plot of land cost.

where overpressure is 25 kPa, i.e. where frameless, self-framing

steel panel buildings are demolished and rupture of oil storage
tanks occurs. This means that for a process unit to be safe it must be
located outside the external damage circle of any neighboring unit.

People fatality probability is computed referring to a Probit equation representing death from lung hemorrhage.
In this case two main ndings are relevant. The rst, referring
to Fig. 8, is that a very small number of generations (i.e. the
evaluation of few hundreds of layouts) is enough to readily obtain
a good safety-based layout, thus conrming the method effectiveness. In particular, the damage loss rapidly converges towards
the theoretical minimum asymptotic value represented by the
sum of the expected annual loss of each unit implied by its own
accident probability (46.353 V/year), without any interaction
among units.
The second observation is that, as expected, the GA progressively attempts to place the process units as far as possible to increase the inter-equipment distance so that the mutual damage
probability is reduced. In the end a layout is obtained where
virtually no unit is located within the damage areas of the other
units. In particular the unit with the largest damage area (i.e. the
ammonia storage tank) is located at one corner while the remaining
units are placed in the farthest opposite corner. However, this
behavior would likely increase all the other cost items, thus asking
for a trade off solution.

Fig. 7. Sample layout evolution during GA run (land investment cost only).

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

147

Fig. 8. Plot of safety cost.

Fig. 10. Plot of total cost and single cost items.

Finally, Figs. 10 and 11 show the results of the overall optimization process where the objective function to be minimized includes all of the above cost items. In this case one observes a
moderate overlap of damage areas for some units which is dictated
by the fact that in this peculiar example safety costs were offset by a
suitable reduction of other cost items by getting equipment closer.
When all cost items are included the optimization leads to an
annual equivalent piping cost of about 30.600 V, a pumping cost
about 21.000 V, an equivalent land cost of 17.700 V and an expected
loss of about 59.200 V. The GA, therefore, within the prescribed 500
generations determines a minimum cost solution which increases
the annual equivalent piping cost by about 1.000 V/year (respect
the solution minimizing piping cost only), increases pumping cost
of 1.600 V/year (respect the solution minimizing pumping cost
only), increases land cost of about 5.900 V/year (respect land cost
minimization only), increases expected annual loss of 12.800 V/
year respect the solution minimizing safety-related cost only.
Nevertheless, this is a minimal overall cost solution with an annual
value of about 128.500 V/year. When minimizing the overall cost
and comparing the best individual of the rst and the last generation, we observed that piping cost passed from 234.100 V/year to
30.600 V/year, pumping cost reduces from 153.700 V/year to
21.000 V/year, land cost from 180.400 V/year to 17.700 V/year.

However, expected loss grows from 50.700 V/year to 59.200 V/

year.
From this latter consideration and from observation of Fig. 10 it
appears that while land, piping and pumping cost markedly
decrease, which means that the GA is progressively reducing the
distances between pieces of equipment while improved layouts are
being generated, expected loss remains substantially stable or even
increases slightly. This could erroneously lead to the supercial
interpretation that safety-related cost is a minor cost item which is
not appreciably inuenced by layout conguration and, consequently, can not contribute to drive the layout towards a minimum
cost solution. This is a wrong interpretation. In fact while pieces of
equipment are driven closer in order to minimize land, piping and
pumping costs, expected damage cost would naturally increase
owing to the higher value of physical consequences of an accident
and the resulting increased probability of damage propagation
between nearby units. Nevertheless, this does not happen in
appreciably manner because the GA is able to continuously relocate
pieces of equipment while they are being moved closer together in
order to maintain stable instead of increasing the expected damage
loss. This is a demonstration of the effectiveness of the proposed
method. The proof that the layout optimization is effective in stabilizing damage loss is shown in Fig. 12 where the cost

Fig. 9. Sample layout evolution during GA run (expected annual loss only).

148

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

Fig. 11. Layout evolution during GA run (total cost and single cost items).

minimization is based on an objective function including only land,

piping and pumping cost while neglecting damage loss. The expected loss corresponding to the generated layout is shown instead
as a separate plot in the same Figure, showing that when layout is
generated without taking into account damage loss this latter cost
increases signicantly as soon as equipment distances are reduced,
up to becoming much higher than the sum of all other cost items.
Fig. 13 shows the resulting optimal layout neglecting the damage
loss. In this example the sum of piping, pumping and land equivalent annual cost is 77.424 V while the expected annual damage
loss is 138.977 V, so that the actual annual cost for this optimal
layout is 216.401 V/year instead of 128.500 V/year previously obtained for the optimal solution including damage loss in the
objective function.
5. Model discussion
A few concluding remarks about the hypotheses made when
building the model can be useful to assess its advantages and
limitations.
As far as safety issues are concerned, it should be pointed out
that the proposed model adopts the so called performance-based
design approach, already widely accepted in re engineering

Fig. 12. Total cost evolution during GA run neglecting damage loss vs expected damage
loss.

context, respect the traditional prescriptive practice based of the

enforcement of predened safety distances. Here the damage
probability is computed based on actual physical effects and interequipment distances, and is not related to the respect of a strict
safety distance. Nevertheless, when predened safety distances are
a prerequisite, such as in order to comply with authorization procedures by Fire Corps or for insurance requirements, then the
respect of prescribed safety distances can be easily met by adding a
penalty term in the objective function, which charges a very high
cost when a minimum safety distance between equipment pairs is
not respected. In this manner the GA will discard solutions breaking
the minimum safety distances constraint.
The expression of the objective function as a linear combination
of four distinct terms, neglecting any cross-correlation, is justied
considering that the total cost is made of those four items which in
general can be considered independent from each other and only
consequent to the location chosen for process equipment. This is
true for land and safety cost, while a degree of correlation may exist
between piping investment and pumping cost. In fact, while piping
capital investment basically depends on pipe length, diameter and
construction material, pipe diameter can be inuenced by pipe
length in order to reduce pressure drop and pumping cost. This is

Fig. 13. Optimal layout obtained neglecting damage loss.

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

neglected in this model as the choice of pipe diameter is based on

industrial rules of thumbs in order to guarantee a proper uid velocity given the required ow rate, irrespective of the pipe length.
Cost of valves and ttings is neglected as in parametric costing
methods it is usually estimated as a prescribed percentage of pipe
investment cost. Moreover, the number and type of valves and
ttings depends more from ow control requirements than on pipe
length or actual layout of pipe route. Some of the cost items depends linearly from distances, such as land cost, while other costs,
i.e. pumping cost and pipe cost have been assumed to have a linear
dependence with distance. This may not be strictly true, but is
consistent with the detail requirements of a preliminary layout
design and the adopted level of model complexity. Moreover this
simplifying assumption is shared with earlier literature on this
subject. One may object that actual cost of a pipe line depends not
only from the distance between starting and ending points but
even on the actual route including bends, crossing inter-section
roads with racks and so on. This is true. Nevertheless, this level of
detail can not be accounted for in the stage of preliminary layout
design but can be considered only when nal pipe layout is dened.
The same applies to pumping costs, which are affected by the actual
pipe route owing to the type and number of concentrated head
losses, which can not be accounted for in the preliminary design
phase. Safety costs, instead, are non-linear with distance but their
amount is correctly determined by the adopted computation
method.
Finally, the proposed approach is aimed at providing a preliminary layout based on total cost minimization. It can be objected
that cost reduction is not the only goal to be achieved when
designing a process plant layout, and that specic requirements,
often related to safety distances and process-related constraints or
even to maintenance accessibility, detailed site topography, construction issues, internal roads networks and so on often play a
relevant role. This is true, but the proposed model is not aimed to
detailed layout design. It is, instead, mainly aimed at providing a
quick preliminary layout, by automatically comparing a great
number of alternatives, in order to dene the preferable relative
location of process units factoring in all relevant costs. When
passing to the detailed layout design phase any further adjustment
and renement, needed to account for specic requirements, can
be easily made starting from a good preliminary solution.
6. Conclusions
In this paper a novel methodology for optimizing the preliminary layout of unequal area rectangular-shaped facilities
considering safety issues is presented. The proposed methodology
adopts a genetic algorithm formulation and is specically tailored
for generating layouts for process plants. It is based on the minimization of the total annual cost explicitly including land cost, investment and operating expenses for the piping network
connecting the process units and safety costs related to the potential loss from accidents in process units. Respect previous attempts to include safety aspects in the optimal layout problem, this
method explicitly adopts a probabilistic approach to model expected annual cost of accidents. Furthermore, it includes operational expenses in the form of pumping cost which are generally
neglected by other authors, and adopts a genetic algorithm procedure instead of a mixed integer linear programming formulation
as happens in most of the related literature concerning process
plant layout. In fact, the adoption of a genetic algorithm enables to
consider any degree of complexity in the denition of the objective
function and in the estimation of safety risk, which is usually prevented in more traditional mathematical programming approaches. This allows a more realistic solution of the layout

149

optimization problem. In particular the proposed algorithm has

demonstrated the ability to signicantly reduce layout cost and has
shown the important role played by expected loss when designing
a plant layout. The algorithm has also shown its effectiveness in
stabilizing accident-related loss when progressively reducing interequipment distance, thus avoiding a rapid increase of expected loss
when traditional minimization of land, piping and pumping cost is
pursued.

References
Al-Hakim, L., 2000. On solving facility layout problems using genetic algorithms.
Int. J. Prod. Res. 38 (11), 2573e2582.
Amorese, L., Cena, V., Mustacchi, C., 1977. A heuristic for the compact location of
process components. Chem. Eng. Sci. 32 (2), 119e124.
Barbosa-Povoa, A.P., Mateus, R., Novais, A.Q., 2001. Optimal two-dimensional layout
of industrial facilities. Int. J. Prod. Res. 39 (12), 2567e2593.
Caputo, A.C., Pelagagge, P.M., Salini, P., 2011. Impact of accidents risk on hydrogen
road transportation cost. Int. J. Energy Sect. Manag. 5 (2), 215e241.
Castell, C.M.L., Lakshmanan, R., Skilling, J.M., Banares-Alcantara, R., 1998. Optimisation of process plant layout using genetic algorithms. Comput. Chem. Eng. 22
(Suppl. 1), S993eS996.
CCPS, 2000. Guidelines for Chemical Process Quantitative Risk Analysis. Center for
Chemical Process Safety American Institute of Chemical Engineers.
Cozzani, V., Salzano, E., 2004. The quantitative assessment of domino effects caused
by overpressure. Part I. Probit models. J. Hazard. Mater. A107, 67e80.
Cozzani, V., Gubinelli, G., Salzano, E., 2006. Escalation thresholds in the assessment
of domino accidental events. J. Hazard. Mater. A129, 1e21.
Davis, L., 1991. Handbook of Genetic Algorithms. Van Nostrand Reinhold, New York.
Daz-Ovalle, C., Vazquez-Rom

an, R., Mannan, M.S., 2010. An approach to solve the
facility layout problem based on the worst-case scenario. J. Loss Prev. Process
Ind. 23, 385e392.

n, R., de Lira-Flores, J., Mannan, M.S., 2013. A model
Daz-Ovalle, C., Vazquez-Roma
to optimize facility layouts with toxic releases and mitigation systems. Comput.
Chem. Eng. 56, 218e227.
Drira, A., Pierreval, H., Hajiri-Gabouj, S., 2007. Facility layout problems: a survey.
Annu. Rev. Control 31, 255e267.
Franceira, S.N., de Almeida, S.S., Guirardello, R., 2007. Optimization of process plant
layout using a quadratic assignment problem model. In: Proc. ICheaP-8 Conference, 24e27 June, Ischia, Italy.
Fuchino, T., Itoh, T., Muraki, M., 1997. Arrangement of process equipment modules
with consideration of plant safety. J. Chem. Eng. Jpn. 30 (5), 896e901.
nchez, L.N., Vazquez-Rom

Garca-Sa
an, R., Diaz-Ovalle, C., Mannan, M.S., 2013.
A multiobjective-driven approach to reduce risk in process layouts. Chem. Eng.
Trans. ISSN: 1974-9791 31. ISBN 978-88-95608-22-8.
Georgiadis, M.C., Macchietto, S., 1997. Layout of process plants: a novel approach.
Comput. Chem. Eng. 21, 337e342.
Georgiadis, M.C., Schilling, G., Rotstein, G.E., Macchietto, S., 1999. A general mathematical programming approach for process plant layout. Comput. Chem. Eng.
23 (7), 823e840.
Goldberg, D.E., 1989. Genetic Algorithms in Search, Optimization and Machine
Learning. Addison-Wesley, Reading, MA.
Gunn, D.J., Al-Asadi, H.D., 1987. Computer aided layout of chemical plant: a
computational method and case study. Comput. Aided Des. 19 (3), 131e140.
Han, K., Cho, S., Park, K., Yoon, E.S., 2013a. Layout optimization of chemical process
based on individual risk. In: Proc. 6th Int. Conf. on Process Systems Engineering,
25e27 June, Kuala Lumpur.
Han, K., Cho, S., Park, K., Yoon, E.S., 2013b. Optimal layout of a chemical process
plant to minimize the risk to humans. Proced. Comput. Sci. 22, 1146e1155.
Jayakumar, S., Reklaitis, G.V., 1994. Chemical plant layout via graph partitioning. I.
Single level. Comput. Chem. Eng. 18 (5), 441e458.
Jung, S., Ng, D., Lee, J.-H., Vazquez-Rom

an, R., Mannan, M.S., 2010. An approach for
risk reduction (methodology) based on optimizing the facility layout and siting
in toxic gas release scenarios. J. Loss Prev. Process Ind. 23, 139e148.

n, R., Mannan, M.S., 2011. New
Jung, S., Ng, D., Diaz-Ovalle, C., Vazquez-Roma
approach to optimizing the facility siting and layout for re and explosion
scenarios. Ind. Eng. Chem. Res. 50, 3928e3937.
Lees, F.P., 1996. Loss Prevention in the Process Industries, second ed. ButterworthHeinemann, Oxford, UK.
Liggett, R.S., 2000. Automated facilities layout: past, present and future. Autom.
Constr. 9, 197e215.
Mecklenburgh, J.C., 1985. Process Plant Layout. Wiley, New York.
Ozyruth, D.B., Realff, M.J., 1999. Geographic and process information for chemical
plant layout problems. AICHE J. 45 (10), 2161e2174.
Papageorgiou, L.G., Rotstein, G.E., 1998. Continuous-domain mathematical models
for optimal process plant layout. Ind. Eng. Chem. Res. 37, 3631e3639.
Park, K., Koo, J., Shin, D., Lee, C.J., Yoon, E.S., 2011. Optimal multi-oor plant layout
with consideration of safety distance based on mathematical programming and
modied consequence analysis. Korean J. Chem. Eng. 28 (4), 1009e1018.
Patsiatzis, D.I., Papageorgiou, L.G., 2002a. Optimal multi-oor process plant layout.
Comput. Chem. Eng. 26 (4e5), 575e583.

150

A.C. Caputo et al. / Journal of Loss Prevention in the Process Industries 34 (2015) 139e150

Patsiatzis, D.I., Papageorgiou, L.G., 2002b. Safe process plant layout using mathematical programming. In: Grievink, J., Schijndel, J.V. (Eds.), Computer Aided
Chemical Engineering, vol. 10. Elsevier, The Netherlands, pp. 295e300.
Patsiatzis, D.I., Papageorgiou, L.G., 2003. Efcient solution approaches for multioor process plant layout. Ind. Eng. Chem. Res. 42 (4), 811e824.
Patsiatzis, D.I., Knight, G., Papageorgiou, L.G., 2004. An MILP approach to safe process
plant layout. Trans. IChemE Part A Chem. Eng. Res. Des. 82 (A5), 579e586.
Penteado, F.D., Ciric, A.R., 1996. An MINLP approach for safe process plant layout.
Ind. Eng. Chem. Res. 35, 1354e1361.
Ray, M.S., Johnston, D.W., 1989. Chemical Engineering Design Project. A Case Study
Approach. Gordon and Breach Science Publishers.
Singh, S.P., Sharma, R.R.K., 2006. A review of different approaches to the facility
layout problems. Int. J. Adv. Manuf. Technol. 30, 425e433.
Suresh, G., Vinod, V.V., Sahu, S., 1995. A genetic algorithm for facility layout. Int. J.
Prod. Res. 33 (12), 3411e3423.
Suzuki, A., Fuchino, T., Muraki, M., 1991. An evolutionary method of arranging the
plot plan for process plant layout. J. Chem. Eng. Jpn. 24 (2), 226e231.

Van den Bosh, C.J.H., Weterings, R.A.P.M., 1997. Methods for Determining Possible
Damage, Green Book, third ed. Committee for the Prevention of Disasters, The
Hague, Netherlands.
Van den Bosh, C.J.H., Weterings, R.A.P.M., 1999. Methods for the Calculation of
Physical Effects, Yellow Book. Committee for the Prevention of Disasters, The
Hague, Netherlands.

n, R., Lee, J.H., Jung, S., Mannan, M.S., 2010. Optimal facility layout
Vazquez-Roma
under toxic release in process facilities: a stochastic approach. Comput. Eng.
Chem. 34, 122e133.
Vazquez-Rom

an, R., Mannan, M.S., 2010. A new trend in designing plant layouts
for the process industry. In: Cacaj, S. (Ed.), Modeling Simulation and
Optimization e Tolerance and Optimal Control. InTech, ISBN 978-953-307056-8.
Xu, Y., Wang, Z., Zhu, Q., 2013. An improved hybrid genetic algorithm for chemical
plant layout optimization with novel non-overlapping and toxic gas dispersion
constraints. Chin. J. Chem. Eng. 21 (4), 412e419.