Computers and Chemical Engineering 56 (2013) 218–227

Contents lists available at SciVerse ScienceDirect

Computers and Chemical Engineering

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

A model to optimize facility layouts with toxic releases and

mitigation systems
Christian Díaz-Ovalle a , Richart Vázquez-Román b,∗ ,
Julio de Lira-Flores b , M. Sam Mannan c
a
Instituto Tecnológico de Roque, Departamento de Ingenierías, Km 8 carretera, Celaya-Juventino Rosas, Gto., CP 38110, Mexico
b
Instituto Tecnológico de Celaya, Departamento de Ingeniería Química, Av. Tecnológico s/n, Celaya, Gto., CP 38010, Mexico
c
Mary Kay O’Connor Process Safety Center, Artie McFerrin Department of Chemical Engineering, Texas A&M University, College Station, TX 77843-3122,
USA

a r t i c l e i n f o a b s t r a c t

Article history: The possibility of including mitigation systems in layout models is explored in this work. The model is
Received 13 January 2012 based on a previous work by the authors to estimate toxic concentrations around each releasing facility
Received in revised form 8 March 2013 surrounded by a mitigation system. The mitigation systems considered here includes water, steam, and air
Accepted 21 May 2013
curtains and exponential decays are assumed for the concentrations shapes before and after the installed
Available online 11 June 2013
curtain. The selection of the mitigation system type to install is included as a variable to determine when
solving the proposed MINLP model. Additional constraints include the conventional non-overlapping and
Keywords:
risk estimations based on probit functions. The objective function includes occupied land costs, intercon-
Facility layout
Toxic releases
nection costs, risk damage costs, and mitigation costs. A software package called TROL has been developed
Mitigation systems to automatically interact with GAMS and ease the initial and final layout descriptions. Numerical results
Water, steam and air curtains indicate that the proposed model produces more practical and optimal layouts.
MINLP problem © 2013 Elsevier Ltd. All rights reserved.

1. Introduction
×Several accidents that include explosions, fires or toxic
releases have occurred in the chemical industry. Explosion and fire
accidents affect workers and property whereas accidents involv-
ing toxic releases may not produce any damage to equipment but
augment the vulnerability of personnel at the plant, and even civil-
ians in the surrounding areas. Some of the damage caused by the
accidents could have been reduced if minimal changes in the instal-
lation layout were made (CCPS, 2003). Facility layout represents
an effective option to reduce the risk of accidents in production
systems. The plant layout is considered a fundamental problem in
chemical plant design. Several strategies have been developed to
solve this problem based on practical experience and using differ-
ent computational tools or heuristic rules (Mecklenburgh, 1985).
An extended list of methods to solve layout problems include par-
tition algorithms, genetic algorithms, and evolutionary methods
(Drira, Pierreval, & Hajri-Gabouj, 2007).
∗ Corresponding author at: Instituto Tecnológico de Celaya, Departamento de
Ingeniería Química, Av. Tecnológico y A.G. Cubas s/n, Celaya, Gto. CP 38010, Mexico.
Tel.: +52 461 61 17575x153; fax: +52 461 61 17744.
E-mail addresses: richart@iqcelaya.itc.mx, richartvr@gmail.com
(R. Vázquez-Román).
Solving the plant layout based on risk analysis has initially
considered the possibility of explosion accidents where risk is
modeled as a function of the distance between the probable explo-
sion point and the affectation points (Penteado & Ciric, 1996;
Patsiatzis, Knight, & Papageorgiou, 2004). An old work has pre-
sented a graph-based algorithm to produce optimal partitioning
to allocate units in different sections where edges’ values reflect
safety costs (Jayakumar & Reklaitis, 1994). In the case of toxic
releases, it has been convenient to group some process units in
facilities and the concept has been extended to include other build-
ings, such as control rooms, so that a facility refers to the portion
of land surrounded by streets where units and people are located
(Vázquez-Román, Lee, Jung, & Mannan, 2010). The probability of
death is computed via probit functions, where the concentration is
typically used as an independent variable though the probability
of death inside a building tends to be inferior (Geeta, Tripathi, &
Narasimhan, 1993). The actual concentration values depend on the
dispersion phenomena so that dispersion models are required to
estimate the exposure concentration, which is subsequently con-
verted to the probability of death.
Dispersion modeling was initially approached by the statisti-
cal distribution of the concentration in the space. Requirements
for accurate predictions and advances in computational tools have
encouraged formulations to include thermal effects and momen-
tum considerations; see, for instance, the models SLAB (Zeman,
1982), FEM3 (Ermak, Chan, Morgan, & Morris, 1982), DEGADIS

0098-1354/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compchemeng.2013.05.017
 C. Díaz-Ovalle et al. / Computers and Chemical Engineering 56 (2013) 218–227 219

Nomenclature
m mitigation systems
a parameter for exponential function of concentration k release facility
aM parameter for exponential function of concentration l affected facility
post-mitigation
AP probit parameter for each gas Greek letters
Ax length of occupied land in x-direction ˛ direction slice
Ay length of occupied land in y-direction
b parameter for exponential function of concentration
bM parameter for exponential function of concentration (Spicer, Havens, Tebean, & Key, 1986), and HEGADAS (Witlox,
post-mitigation 1994). Then, dispersion models added physical features from the
B binary variable to indicate the angular position of surroundings, like boundary conditions, to produce more accurate
the facility predictions; this is represented in computational fluid dynamics
BC binary variable for no-mitigation case – CFD-tools (Blocken, Carmeliet, & Stathopoulos, 2007; Hanna,
BM binary variable related to applying mitigation Hansen, Ichard, & Strimaitis, 2009). However, the weather is a
Bmit binary variable to select a mitigation system from a rather difficult variable in dispersion modeling because of its
set stochastic nature in parameters such as atmospheric condition,
BP probit parameter for each gas wind direction, wind speed, temperature, and humidity (Sullivan,
C selected concentration Holdsworth, & Hlinka, 2004; Marx & Cornwell, 2008).
CD cost willing to avoid a fatality The facility layout problem may use a stochastic formulation for
CL total cost land weather conditions, where calculation of random data is included
CL land cost per square length to get probability distribution of the damage in toxic releases
CM concentration of gas post-mitigation (Vázquez-Román et al., 2010). Moreover, dispersion parameters in
CM total mitigation cost specific conditions have shown risk reduction. This set of conditions
CMit cost of a mitigation system in a facility is defined as the worst credible scenario (Teles, Castro, & Matos,
CP total cost for piping 2012). The definition of the worst credible scenario is formed by
CP pipe cost per length a low wind speed, stable atmospheric conditions, and non-terrain
CR concentration for a release from a facility obstructions (Crowl & Louvar, 2002; Díaz-Ovalle, Vázquez-Román,
CT total cost & Mannan, 2009). The stochastic approach considers micromete-
D separation distance between the point of a release orological effects on the toxic dispersion phenomenon, whereas
and the center of a facility the deterministic approach is based on the worst-case scenario.
Da,b separation distance between facilities Optimal solutions to the facility layout problem may reduce risk to
DM distance from release point to the facility border acceptable levels (Díaz, Vázquez-Román, Jung, & Mannan, 2009).
Dmin,x minimum separation in x-direction between facili- However, both deterministic and stochastic approaches may sug-
ties gest large unpractical separation distances between releasing and
Dmin,y minimum separation in y-direction between facili- occupied facilities (Díaz-Ovalle, Vázquez-Román, & Mannan, 2010).
ties Distances between facilities might be reduced if the concentration
f frequency of occurrence in a toxic release accident is decreased, in which case a mitigation system should be used.
in a facility Mitigation systems are currently used in real process plants to
LX length of land decrease the concentration during toxic dispersions. The selection
Lx length of a facility of a mitigation system strongly depends on the mitigating fluid
LY width of land and the fluid to mitigate. The selection is formulated by considering
Ly width of a facility design parameters and a dispersion factor to achieve more efficient
Ma,b set of interconnectivity mitigation (Molag et al., 2001). Mitigation systems are also selected
m tangent of angle ˛ based on the type of accident: Foams are used for liquid leaks, air
Pe persons inside in a facility curtains or water curtains for gases presenting physic effects, dilu-
P probability of death in a release tion effects or absorption effects (Dimbour, Dandrieux, Gilbert, &
PL expected life time of the plant Dusserre, 2003). Three mitigation systems are incorporated here in
Px distance from the facility center to the source in x- the layout model to decrease separation distances between facili-
coordinate ties based on a previous work by us (Diaz-Ovalle, Vazquez-Roman,
Py distance from the facility center to the source in y- Lesso-Arroyo, & Mannan, 2012).
coordinate
R risk in terms of affected people/year for a release 2. Problem statement
Sx , Sy slice-vectors to define quadrant positions
st street length The problem here considers a set of new facilities to be accom-
t personal exposure time modated in a given land where other facilities may have been
x x-coordinate of a facility already installed. Facilities may interact among them so that the
y y-coordinate of a facility interconnectivity is given. Interconnectivity is an important factor
Y probit value in a release from a facility and its impact on cost is evident in this problem. The typical
approach, based on minimum distances between facilities, is
Subscripts
incorporated in the model and also assuming that there must be
i already existing facilities
a street around each facility (Vázquez-Román et al., 2010). The
s new facilities for sitting
problem is focused on risk constraints, where the estimations come
r release types
from modeling the toxic release of a dense gas using a mitigation
system. The risk is considered a result of the frequency times
220 C. Díaz-Ovalle et al. / Computers and Chemical Engineering 56 (2013) 218–227

the severity of any given accident scenario (Modarres, 2006). The where s ∈ S, k ∈ I ∨ k ∈ S, k =
/ s, and
frequency estimation comes from historic data (Taylor, 2005) and Lxs + Lxk
min,x
its severity or impact comes from the amount of injured people Ds,k = + st, s ∈ S, k ∈ I ∨ k ∈ S, k=
/ s (4)
2
(Crowl & Louvar, 2002). Thus, the expected quantity of people in
each facility is a required datum. min,y Lys + Lyk
The following sets are used in the model: already existing Ds,k = + st, s ∈ S, k ∈ I ∨ k ∈ S, k=
/ s (5)
2
facilities, i ∈ I, new facilities for sitting, s ∈ S, release types, r ∈ R,
Modeling a disjunction is not possible in commercial opti-
mitigation systems, m ∈ M, intervals, ˛ ∈ A. In addition, the follow-
mization packages such as GAMS. Each disjunction has to be
ing subsets are considered: interconexions and intervals a,b ∈ Ma,b ,
reformulated as a MINLP (mixed integer non-linear programming).
facilities having a particular release, ri(i,r), facilities with mitiga-
Two methods have been applied to reformulate the disjunction:
tion systems, mi(i,m), existing facilities with personnel, pi(i), new
The convex hull method and the big-M (Lee & Grossmann, 2000).
facilities having a particular release, rs(s,r), new facilities with mit-
In particular, the non-overlapping constraint has been formulated
igation systems, ms(s,m), and new facilities with personnel, ps(s).
using the big-M method (Patsiatzis et al., 2004) and the convex hull
Parameters in the model include: length and width of land, LX and
(Vázquez-Román et al., 2010). In this work, the convex hull method
LY. length and width of new facilities, Lxs and Lys , length and width
is used to solve all disjunctions.
of existing facilities, Lxi and Lyi with a center point in xi and yi , a
distance from the facility center to the source in existing facilities,
3.2. Risk estimation
Pxi,r and Pyi,r , a distance from the facility center to the source in new
facilities, Pxs,r and Pys,r , persons inside in a installed and a new facil-
Risk estimations always refer to specific scenarios. In this case,
ity, Pei and Pes , frequency of occurrence in a toxic release accident
the scenario considers a toxic release where the consequence refers
in a existing and new facilities, fi and fs , expected life time of the
to the amount of injured people. The damage produced in these sce-
plant, PL ; street length, st, pipe cost, CP , land cost, CL , cost willing
narios depends on the toxic concentration, which is estimated here
to avoid a fatality, CD , and cost of a mitigation system in an exist-
via CFD simulations to fit an exponential decay function as indicated
ing facility, CMitS and in a new facility, CMiti . Also, the total costs
in (Vázquez-Román et al., 2010). However, the dispersion models
definition is for pipe, CP , land, C, damage, CD , and mitigation, CM.
are typically formulated as a system of ordinary differential equa-
Then the variables to determine are: Center position of each new
tions. Unfortunately, optimizers such as those in GAMS (Brooke,
facility, xs and ys , the occupied area, the selected mitigation system
Kendrick, Meeraus, & Raman, 1998) cannot include differential
for each facility with release, the final interconnection, land, risk,
equations in the model. To overcome this difficulty, the predicted
and mitigation costs associated with the optimal layout so that the
concentration is approximated with an exponential decay func-
total cost of facility layout, CT , is minimized.
tion, which is a function of the separation distance between the
release point and any surrounding point. The correlated function
3. Mathematical formulation becomes:
R bk,r D2
The facility layout problem can be formulated as an optimization Ck,r,l = ak,r e k,r,l

type where an objective function is subject to several constraints.

k ∈ ri (i, r) , l ∈ ps (s) , r∈R (6)
A set of geometrical constraints allows selecting feasible distri-
∨k ∈ rs (s, r) , l ∈ pi (i) , r∈R
butions for new facilities and risk constraints are included in the
model to consider the probability of death, which becomes a func- ∨k ∈ rs (s, r) , l ∈ ps (s) , k=
/ l, r∈R
tion of the separation distance among facilities. Mitigation systems R
where Ck,r,l is the concentration for a release r in the center of facil-
are applied here only in facilities with any toxic release.
ity k and affecting the facility l, ak,r and bk,r are parameters for the
exponential function, and Dk,r,l is the separation distance between
3.1. Geometrical constraints the point of a release r in a facility k affecting facility l. The Euclidian
distance has to be used to measure the distance since it is used to
Land issues consider a specific area where the facilities will be calculate concentrations due to toxic dispersions.
allocated: Damage to people by toxic dispersion is summarized by the
Lxs
 Lx  probability of death, and it depends on concentration, exposure
s
+ st ≤ xs ≤ LX − + st , s∈S (1)
2 2 time, and toxicity properties of the released gas. This relation is
described in a model developed by Haber (CCPS, 2000), which is
Lys
 Ly  correlated with a normal logarithmic probabilistic distribution to
s
+ st ≤ ys ≤ LY − + st , s∈S (2) describe the probability of occurrence for each event (Geeta et al.,
2 2
1993). This approach allows calculation of the probability of death
Facilities must never be stacked on top of each other in mul- based on probit functions:
tiple story buildings. The so called non-overlapping condition is
formulated considering a facility s as a reference to allocate another
facility k. Thus, facility k could be allocated in four regions, identi-
fied as right (R), left (L), above (A) or below (D). A disjunction has
been proposed to represent this constraint:
⎡ ⎤
“A”, “D”
 ⎢ ⎥
min,x
   ⎢ xs ≥ xk − Ds,k
⎥
“L” “R” ⎢ xs ≤ min,x
xk + Ds,k ⎥
∨ ∨⎢ ⎥ (3)
min,x
xs ≤ xk − Ds,k min,x
xs ≤ xk + Ds,k ⎢   ⎥
⎣ “L” “R” ⎦
min,y ∨ min,y
ys ≤ yk + Ds,k ys ≤ yk − Ds,k
 C. Díaz-Ovalle et al. / Computers and Chemical Engineering 56 (2013) 218–227 221

Table 1
20 m
Probit parameters for some toxic dense gases from guidelines for chemical process
quantitative risk analysis (CCPS, 2000).

Chemical A B n

Ammonia –35.90 1.85 2.00

Chlorine –8.29 0.92 2.00 Facility
Phosgene –19.27 3.686 1.00

nr

Yk,r,l = APr + BrP ln Ck,r,l tk,r

k ∈ ri (i, r) , l ∈ ps (s) , r∈R (7)

∨k ∈ rs (s, r) , l ∈ pi (i) , r∈R
∨k ∈ rs (s, r) , l ∈ ps (s) , k=
/ l, r∈R Fig. 1. Isoplete of 3 ppm after mitigation of a chlorine release allocated in (10,7).

where APr , BrP and nr are probit parameters for each gas r, tk,r is the
personnel exposure time to a given release (min), Ck,r,l is concen-
tration of gas (ppm), and Yk,r,l refers to probit units. Some values releasing facility. The following disjunction is used to detect this
of parameters are defined for several gases in Table 1. Thus, the slice:
probability of death is calculated by
⎡ ⎤
“˛-slice”
 Yk,r,l −5 u2 ⎢ y
  ⎥
⎢ S˛ yl − yk + Pyk,r ≥0 ⎥
Pk,r,l = √
1 −
e 2 du ∨ ⎢   ⎥
⎢ S˛x xl − xk + Pxk,r ≥0
2 −∞
˛A ⎢ y
  x   ⎥⎥
k ∈ ri (i, r) , l ∈ ps (s) , r∈R ⎣ S˛ yl − yk + Pyk,r ≤ S˛ m˛ xl − xk + Pxk,r ⎦
(8) y
    (10)
S˛ yl − yk + Pyk,r ≥ S˛x m˛−1 xl − xk + Pxk,r
∨k ∈ rs (s, r) , l ∈ pi (i) , r∈R
k ∈ ri (i, r) , l ∈ ps (s) , r∈R

∨k ∈ rs (s, r) , l ∈ ps (s) , k=
/ l, r∈R ∨k ∈ rs (s, r) , l ∈ pi (i) , r∈R
∨k ∈ rs (s, r) , l ∈ ps (s) , k=
/ l, r∈R

where u is a function to compute a Gaussian distribution of proba-

bility Pk,r,l . y
where S˛ and S˛x are vectors with elements either 1 or −1 and
The consequences of accidents are measured by the amount of they determine the allocation of facility l with respect to facility k
personnel involved, which, in this case, is based on the quadrants, and m˛ is the tangent of angle ˛. The values
of the vectors are specified in each quadrant. Thus, S˛x takes values
 y
Rk,r,l = fk,r Pk,r,l Pel of 1, −1, −1, 1 and S˛ takes values of 1, 1, −1, 1 (Vázquez-Román
et al., 2010). The number of slices must be 4n, n being an integer,
k r l
k ∈ ri (i, r) , l ∈ ps (s) , r∈R (9) because of the quadrant-type of divisions.
A complete formulation of curtain-based mitigation systems has
∨k ∈ rs (s, r) , l ∈ pi (i) , r∈R
been described in (Diaz-Ovalle et al., 2012). This model is used to
∨k ∈ rs (s, r) , l ∈ ps (s) , k=
/ l, r∈R estimate the concentration before and after applying a curtain mit-
igation system. It is then assumed that the concentration beyond
where Rk,r,l is the risk in terms of affected people/year for the release the curtain effect follows an exponential decay:
r, in a facility k, affecting facility l.

3.3. Mitigation constrains M

Dk,r
The harm effect of toxic releases depends on the gas concen-
tration. Let us consider a rectangular facility that is 25 m long and Dk,r,l _ DMk,r
20 m wide, where a chlorine release occurs inside the facility in the
point (10 m, 7 m), ((0,0) representing the center point of the facil-
ity). Then, a mitigated isoplete of 3 ppm would look as indicated in
Dk,r,l
Fig. 1, with the mark indicating the releasing point. It clearly shows
an angular variation on the mitigated concentration.
Assuming that the mitigation system is located around the
releasing facility, the gas is initially dispersed at a distance from the Facility k Facility l
releasing point to the facility border, Dk,r M , and the post-mitigation
Release point
dispersion starts from the allocation of the mitigation system to the Facility centerpoint
affected facility, Dk,r,l , Fig. 2. Thus, it is clearly necessary to iden-
tify the slice where a receptive point is located with respect to the Fig. 2. Separation distances from the releasing point to affected points.
222 C. Díaz-Ovalle et al. / Computers and Chemical Engineering 56 (2013) 218–227


M bM Dk,r,l −DM where Cs,r,I is the selected concentration, BsC is the binary variable
Ck,r,l,m = aM e k,r,m k,r
k,r,m for no-mitigation case, BsM is the binary variable related to apply-
k ∈ ri (i, r) , l ∈ ps (s) , r ∈ R, m∈M (11) ing mitigation, and Bmits , is a binary variable to select a mitigation
∨k ∈ rs (s, r) , l ∈ pi (i) , r ∈ R, m∈M system. Binary variables are restricted to
∨k ∈ rs (s, r) , l ∈ ps (s) , k=
/ l, r ∈ R, m∈M M
Bmits,m · Cs,r,i,m (16)
M
where, Ck,r,l,m is the concentration after mitigation for an r emis-
sion with a mitigation system m in the facility k affecting facility
l, and aM
k,r,m
and bM
k,r,m
are parameters obtained by regression from 
the exponential function to represent the post-mitigated cloud. All Bmits,m = BsM , ∀s ∈ rs (s, r) , m∈M (17)
possible 360◦ directions around the releasing facility are discred- m∈ms(s,m)
ited and parameters from Eq. (11) are correlated for each slice. The
The last equation is related to a new facility, k, with a release, r,
convex hull method is also used here to convert the disjunctions
affecting another new facility, s, with personnel:
into a MINLP.
The angular facility allocation also allows selecting the mitiga-

M
Ck,r,s = BkC · Ck,r,s
R + BkM · Bmitk,m · Ck,r,s,m
tion distance and concentration. A binary variable Bi,s,˛ , is used to (18)
m∈ms(s,m)
indicate the angular position of the facility and the selected miti-
∀k ∈ rs (s, r) , s ∈ ps (s) , k=
/ s, r ∈ R, m∈M
gation concentration is defined by
  where Ck,r,s is the selected concentration.
bM Di,r,s −DM
M
Ci,r,s,m = Bi,s,˛ × aM
i,r,m,˛
e i,r,m,˛ i,r,˛

˛∈A
(12)
3.4. Objective function
∀i ∈ ri (i, r) , s ∈ ps (s) , r ∈ R, ˛∈A
M
where Ci,r,s,m is the mitigated concentration in a mitigation system The total cost of the installation must be minimized and includes
M land costs, interconnection costs, mitigation costs and damage
m, Di,r,˛ is mitigation distance in facility i with release r in direction
costs. The land costs depend on the total occupied land, which
˛, aM
i,r,m,˛
and bM
i,r,m,˛
are mitigation parameters given in the facility is considered here as the minimum rectangular space where all
i with a release r, and the mitigation system m in direction ˛. facilities can be grouped. The occupied area is the product of two
Formulation of mitigation systems in the layout problem is variables, Ax and Ay , computed by
based on a list of valid mitigation systems that reduce the risk pro-
duced by toxic releases. The optimization includes a selection of Lxs
Ax ≥ xs + , ∀s ∈ S (19)
the mitigation system to produce an optimal risk reduction. The 2
selection of the mitigation system is achieved through binary vari-
ables. When a release, r, in an installed facility, i, affects a new Lys
Ay ≥ y + , ∀s ∈ S (20)
facility, s, with personnel, then the selection of a mitigation system 2
is formulated as: Since the occupied area is minimized, the above constraints

Ci,r,s = BiC · Ci,r,s
R + M
Bmiti,m · Ci,r,s,m oblige Ax and Ay to take the appropriate values. The cost
(13) of the occupied area, using a parameter for the land cost,
m∈mi(i,m)
becomes
∀i ∈ ri (i, r) , s ∈ ps (s) , r ∈ R, m∈M

where Ci,r,s is the selected concentration, BiC is the binary variable C L = CL Ax Ay (21)
for no-mitigation case, and Bmiti,m is a binary variable to select a The interconnectivity cost is estimated through a parameter for
mitigation system from a set. The binary variables are constrained the unit pipe cost multiplied by the total length of required pipe.
by For the sake of simplicity, a single value is used for the unit cost
 regardless of the size and type of the facilities and the separation
BiC + Bmiti,m = 1, ∀i ∈ ri (i, r) (14)
distance measured between the center of the involved facilities.
m∈mi(i,m)
Thus, the piping cost is
The case of a new facility, s, with a release, r, affecting an installed 
facility, i, occupied by people is formulated by C P = CP Da,b (22)
 (a,b)∈Ma,b
R M
Cs,r,i = BsC · Cs,r,i + BsM · Bmits,m · Cs,r,i,m
m∈ms(s,m) (15) where Da,b is the Euclidian separation distance between facilities a
∀s ∈ rs (s, r) , i ∈ pi (i) , r ∈ R, m∈M and b:
2
Da,b = (xa − xb )2 + (ya − yb )2 , ∀ (a, b) ∈ Ma,b (23)

The injury cost is based on the cost the company is willing to

pay to avoid a fatality and the risk. In this work the occurrence of
scenarios are considered non simultaneous so that
⎛ ⎞
⎜ ⎟
⎜       ⎟
⎜ ⎟
C = CD PL ⎜
D
fs,r Ps,r,k Pek + fs,r Ps,r,i Pei + fi,r Pi,r,s Pes ⎟ , ∀r ∈ R (24)
⎜ ⎟
⎝s∈rs(s,r) k ∈ ps (s) s∈rs(s,r)i∈pi(i) i∈ri(i,r)s∈ps(s) ⎠
k=
/ s
 C. Díaz-Ovalle et al. / Computers and Chemical Engineering 56 (2013) 218–227
The total mitigation cost is
 
CM = Cmits + Cmiti
s∈rs(s,r) i∈ri(i,r)
Thus the objective function contains the total cost defined by the
sum of costs indicated by Eqs. (21), (22), (24) and (25). The above
mode has been solved using GAMS. To easy the model application
and the layout description, a code has been written in Visual Stu-
dio C++. The software, called TROL (toxic release in optimal facility
layout), is briefly described in next section.
4. TROL
The facility layout is a problem whose solution can be
graphically described in a computer. A pioneer computational
software called ALDEPT (automated layout design program) con-
sidered prioritization of the neighboring facilities (Sheehof &
Evans, 1967). Some packages have included other sophisti-
cated methodologies such as diffuse algorithms in FLEXPERT
(Alif & Badiru, 1996), and genetic algorithms in HOPE (Kochhar,
Foster, & Heragu, 1998). Unfortunately, these packages are not
able to solve the facility layout problem with safety con-
straints.
TROL is an interface developed in Microsoft Visual Studio 2010
based in the Visual C++ language and to run in current Windows
Fig. 3. The TROL interface.
223
platform, Fig. 3. The target in TROL is to show the allocation of a
set of facilities, where some of them have releases, personnel and
(25) mitigation systems, as already described in this work. Users define
the features of facilities, properties of releases and mitigation
systems. Before solving the facility problem, TROL obtains the
dispersion and mitigation parameters by solving the equations
indicated in our previous work (Diaz-Ovalle et al., 2012), and
these parameters are storage in text type files. Then, TROL creates
a program in GAMS that runs automatically to get the optimal
allocation of all new facilities. The solution is also automati-
cally incorporated and the layout is graphically described in the
pane-window. Two case studies are given in the next section
where TROL has been used to find and describe the optimal
layout.
The way to interact TROL and GAMS is via files. The user must
indicate the directory where the gams.exe file has been installed. In
addition, the user must also indicate the directory and a name for
the gams-file, extension gms, which TROL will generate to write the
model. This gams-file is easily generated by conventional C++ code.
Required parameters not only for the process facilities, e.g. dimen-
sions, but also for the numerical analysis, e.g. numerical solver, are
incorporated through appropriate dialogs. Options to perform the
optimization in GAMS are also incorporated via a windows-dialog
to write the opt-file required in particular solvers. Once the gams
file is written, the gams.exe is executed using the following C++
instructions (see http://www.gams.com/dd/docs/api/):
224 C. Díaz-Ovalle et al. / Computers and Chemical Engineering 56 (2013) 218–227

LPWSTR CmdLineLPSTR= (LPWSTR) CmdLine.GetBuffer();

LPCWSTR aux1= (LPCWSTR) E xeName.GetBuffer();
LPCWSTR aux2= (LPCWSTR) CurrentDir.GetBuffer();

if (!CreateProcess(
aux1, // pointer to name of executable module
CmdLineLPSTR, // pointer to command line string
NULL, // process security attributes
NULL, // thread security attributes
TRUE, // handle inheritance flag
CREATE_NEW_CONSOLE, // creation flags
NULL, // pointer to new environment block
aux2, // pointer to current directory name
&si, // pointer to STARTUPINFO
&pi // pointer to PROCES S_INFORMATION
)) {
DisplayError( "CreateProcess" );
free(CmdLineLPSTR);
return FALSE;
}

where ExeName is a string containing the gams.exe pathway and

Current Dir is a string containing the model-gams-file to run. The
code displays a window from where the file can be selected to run.
The evolution of running GAMS is visualized in this window. When
the execution stops, the results produced by GAMS are extracted
from the corresponding lst-file to verify the status. When the solu-
tion is achieved then the corresponding layout is produced in the
appropriate window. Otherwise, the user is notified in order to
modify parameters to make it run. Fig. 4 gives a graphical descrip-
tion of the interaction TROL-GAMS.
5. Case study
This section presents two examples to show the advantages of
the proposed model as well as the representation in the developed
code. These cases consider several releases allocated in either new
or already installed facilities. A set of valid mitigation systems is also
defined for each facility with a release. Any required information is
defined for each case study. The solution for these cases is obtained

Define option solver

code in a
.opt file
-------
-------
TROL: Main routine
-------
-------
-------
Results file as
Read information for .lst
option file Routine to run -------
GAMS -------
-------
Write cod e to .gms file -------
-------

-------
-------
-------

Update results
TROL modifies
facility allocation

Mitigation parameters
storage at .txt files

Fig. 4. The TROL interface.

 C. Díaz-Ovalle et al. / Computers and Chemical Engineering 56 (2013) 218–227 225

Table 2 Table 5
Installed facilities for case study 1. Installed facilities for case study 2.

Facility X Y Lx Ly Persons Facility X Y Lx Ly Persons

Facility A 30 20 50 30 20 Facility A 57.9 31.7 25 20 0

Facility B 35 75 60 70 2 Facility B 21.5 30 30 50 1
Facility C 97 20.5 45 30 5
Facility D 88.7 49.5 30 20 0
Table 3
New facilities for case study 1.

Facility Lx Ly
Control room 20 15
New A 60 70
New B 110 80
New C 40 100
through TROL using a combination of the solvers CONOPT-DICOP.
The code was solved on an Intel® CorelTM 2 Duo T9600 2.8 GHz
processor PC.
5.1. Case study 1
In this case study, a distribution of facilities is sought on a square
piece of land of 1100 m in x and y coordinates. Features of the new
and installed facilities are described in Tables 2 and 3, respectively.
Interconnection exists between the following facilities: Facility
B–New A, Facility B–New C, and New A–New C. Costs are considered
20 $/m2 for land, 200 $/m for interconnections, and $8,000,000 for
each fatality. Distribution also includes a 5 m wide street between
facilities and the profitable life of plant is 50 years. Dispersions
assume atmospheric conditions at 1.5/F, 45% air humidity, a rough-
ness equivalent to lawn of 1 cm, and a temperature of 290 K.
Facility B includes a chlorine release at its center (0,0) with a
frequency of 2.5E-4 yr−1 , flow rate 420 g/s from a 1 m diameter pool,
density of 3.21 kg/m3 , and temperature of 245 K. New C presents
a phosgene release allocated in (–5,0) with a frequency of 1.5E-
6 yr−1 , flow rate 80 g/s through a 1 in. pipe, density of 4.35 kg/m3
and temperature of 260 K.
Mitigation systems included in this problem were modeled in
TROL and used the model proposed in Part I of this paper. Facility B
considers steam and water curtains as mitigation systems, whose
costs are $23,000 and $18,000, respectively. The steam curtain has
a pressure of 5 bar running through 0.5 in holes 50 cm apart. The
water curtain is at 298 K, with 20 gpm in volumetric rate through ¾
in holes 90 cm apart. The water curtain presents a droplet average
diameter of 300 ␮m with an opening angle of 120◦ from the nozzle.
Finally, New C has air and steam valid curtains were considered, and
are $20,000 and $23,000, respectively. The air curtains are 280 K,
45% humidity, and 5 bar through 1 in. holes 45 cm apart. The steam
curtain is 5 bar through 0.5 in. holes 50 cm apart.
The solution for this case study indicates a total cost of 1,054,715
and the selected mitigation systems are the steam curtain for Facil-
ity B and the air curtain for New C. The same case study was solved
assuming no mitigation, and the total cost drops to 1,997,610.
The center points for each facility are given in Table 4 and their
Table 4
Allocation and costs for case study 1.
Persons
graphically described layouts are given in Fig. 5. GAMS solved the
problem in 7.7 s.
10
0
2 5.2. Case study 2
0
In this case study, facilities on an available square piece of land
have 700 m in each side. Features of new and installed facilities are
described in Tables 5 and 6, respectively. There are interconnec-
tions between facilities: New A–New B and New B–New C. Costs are
defined as 16.0 $/m2 for land, 60.8 $/m for interconnections and
$8,000,000 for each fatality. Also, the distribution includes a 5 m
wide street between facilities and the profitable life of plant is 50
years. It is assumed that atmospheric conditions prevailing during
dispersions are 1.5/F, 25% air humidity, roughness equivalent to
lawn of 1 cm, and a temperature of 290 K.
Facility C includes a chlorine release in (10,−7) with a fre-
quency of 5E−4 yr−1 , flow rate 114 g/s from a 1 in. pipe, density of
3.21 kg/m3 , and temperature of 245 K. Also, a H2 S release occurs
in (0,0) with a frequency of 1E−3 yr−1 , flow rate 50 g/s, density
of 1.36 kg/m3 through a 1 in. pipe, and temperature of 222 K. New
C presents a similar chlorine release in (0,0) with a frequency of
2.5E−3. New D has a similar H2 S release in (0,0) with a frequency
of 1E−3.
Facility A considers air and water curtains as valid mitigation
systems whose costs are $18,000 and $20,000, respectively. The air
curtain has a pressure of 3 bar, 280 K and 40% humidity through
0.51 in. holes 35 cm apart. The water curtain is 298 K, with a vol-
umetric rate 14 gpm through 3/4 in. holes 20 cm apart, and this
curtain creates droplets with an average diameter of 300 ␮m with
an opening angle 90◦ from the nozzle. On the other hand, New
C has an air, water, and steam curtains whose cost are $22,000,
$23,000, and $24,500, respectively. The air curtain is 280 K, 40%
humidity, and 3 bar through 0.5 in. holes 35 cm apart. The water
curtain has holes 30 cm apart and opening angle of 120◦ , the rest
is the same as aforementioned. The steam curtain is 5 bar through
1 in. holes 35 cm apart. Finally, New D has air, water, and steam cur-
tains, whose cost are $22,000, $23,000 and $24,500, respectively.
The air and steam curtains are identical to New C curtains. The water
curtain has holes 50 cm apart and an opening angle of 90◦ , the rest
is similar to water curtain in New C.
The results indicate that the water curtain is the best mitiga-
tion system in all the facilities. The same case study is also solved
assuming no mitigation system is valid in the layout. The results for
both with and without mitigation systems are indicated in Table 7
and Fig. 6. Results clearly indicate the advantages in the layouts
when mitigation systems are implemented. Layouts are safe but
the mitigation model yields more practical layouts.

Without mitigation With mitigation Table 6

New facilities for case study 2.
x Y x y
Facility Lx Ly Persons
Control room 894.04 37.5 356 75
New A 145 66.5 100 47.5 Control room 15 15 10
New B 814.04 68 286 75 New A 25 45 2
New C 255 58 206 75 New B 60 60 0
Total land (m2 ) 97,636.5 45,742 – – New C 35 20 0
Total cost ($) 1,997,610 1,054,715 – – New D 35 20 0
226 C. Díaz-Ovalle et al. / Computers and Chemical Engineering 56 (2013) 218–227

Fig. 5. Results for case of study 1: (a) Without mitigation: total cost($) = 1,997,610 and (b) with mitigation: total cost($) = 1,054,715.

Table 7
Allocation and costs for case study 2.

Without mitigation With mitigation

x y X Y

Control room 213.7 12.5 127.5 59.2

New A 298.9 35 82.5 99.4
New B 298.9 80 35 92
New C 311.4 105 117.5 110.5
New D 188.7 27.5 117.5 85.3
Total land (m2 ) 37,823 16,458.2 – –
Total cost ($) 608,698 340,241 – –
 C. Díaz-Ovalle et al. / Computers and Chemical Engineering 56 (2013) 218–227
Fig. 6. Results for case study 2: (a) Without mitigation: total cost($) = 608,698 and (b) with mitigation: total cost($) = 340,241.
6. Conclusions
A model to optimize layouts at facilities that are installed or
are going to be installed and may release toxic materials has been
presented in this paper. To produce practical layouts, a selec-
tion of mitigation systems has been included in the optimization
scheme. This approach supports multi-mitigation systems where
the appropriate, cheap, safe, system is selected for each facility
for release. Mitigation systems are even able to reduce the total
layout cost by reducing the risk cost. However, an increase in facil-
ities and the number of mitigation systems increases the number
of equations and binary variables so that achieving the solution
may take too much time. This feature will be analyzed in a future
work.
Acknowledgements
The authors would like to acknowledge the funding from CONA-
CyT and additional support by PROMEP and DGEST.
References
Alif, A., & Badiru, A. B. (1996). FLEXPERT: facility layout expert system using fuzzy
linguistic relationships codes. IIE Transactions, 28(4), 295–308.
Blocken, B., Carmeliet, J., & Stathopoulos, T. (2007). CFD evaluation of wind speed
conditions in passages between parallel buildings-effect of wall-function rough-
ness modifications for the atmospherci boundary layer flow. Journal of Wind
Engineering and Industrial Aerodynamics, 95, 941–962.
Brooke, A., Kendrick, D., Meeraus, A., & Raman, R. (1998). GAMS–a user guide. Wash-
ington, DC: GAMS Development Corporation.
CCPS (2000). Guidelines for chemical process quantitative risk analysis. New York:
Center for Chemical Process Safety of the AIChE.
CCPS (2003). Guidelines for facility siting and layout. Center for Chemical Process
Safety, AIChE.
Crowl, D. A., & Louvar, J. F. (2002). Chemical process safety, fundamentals with
applications. In Prentice-Hall International Series in the physical and chemical
engineering sciences. Englewood Cliffs, NJ: Prentice Hall.
Diaz-Ovalle, C., Vazquez-Roman, R., Lesso-Arroyo, R., & Mannan, M. S. (2012). A
simplified steady-state model for air, water, and steam curtains. Journal of Loss
Prevention in the Process Industries, 25(6), 974–981.
Díaz-Ovalle, C., Vázquez-Román, R., & Mannan, M. S. (2009). Determinación de los
factores del peor escenario en la emisión de gases tóxicos. Información Tecnológ-
ica, 20(1), 3–10.
Díaz-Ovalle, C., Vázquez-Román, R., & Mannan, M. S. (2010). An approach to solve
the facility layout problem based on the worst-case scenario. Journal of Loss
Prevention in the Process Industries, 23(3), 385–392.
Díaz, C. O., Vázquez-Román, R., Jung, S., & Mannan, M. S. (2009). A com-
parison of deterministic and stochastic approaches to solve the facility
layout problem with toxic releases. Computer Aided Chemical Engineering, 25,
93–98.
227
Dimbour, J. P., Dandrieux, A., Gilbert, D., & Dusserre, G. (2003). The use of water
sprays for mitigating chlorine gaseous releases escaping from a storage shed.
Journal of Loss Prevention in the Process Industries, 16(4), 259–269.
Drira, A., Pierreval, H., & Hajri-Gabouj, S. (2007). Facility layout problems: a survey.
Annual Reviews in Control, 31(2), 255–267.
Ermak, D. L., Chan, S. T., Morgan, D. L., & Morris, L. K. (1982). A comparison of dense
gas dispersion model simulations with burro series LNG spill test results. Journal
of Hazardous Materials, 6(1-2), 129–160.
Geeta, B., Tripathi, A., & Narasimhan, S. (1993). Analytical expressions for estimat-
ing damage area in toxic gas releases. Journal of Loss Prevention in the Process
Industries, 6(2), 125–131.
Hanna, S. R., Hansen, O. R., Ichard, M., & Strimaitis, D. (2009). CFD model simula-
tion of dispersion from chlorine railcar releases in industrial and urban areas.
Atmospheric Environment, 43(2), 262–270.
Jayakumar, S., & Reklaitis, G. V. (1994). Chemical plant layout via graph
partitioning—1. Single level. Computers and Chemical Engineering, 18(5),
441–458.
Kochhar, J. S., Foster, B. T., & Heragu, S. S. (1998). Hope: a genetic algorithm for
the unequal area facility layout problem. Computers and Operations Research,
25(7-8), 583–594.
Lee, S., & Grossmann, I. E. (2000). New algorithms for nonlinear generalized disjunc-
tive programming. Computers and Chemical Engineering, 24, 2125–2141.
Marx, J. D., & Cornwell, J. B. (2008). The importance of weather variations in a quanti-
tative risk analysis. College Station, TX: Making Safety Second Nature & WCOGI.
Mecklenburgh, J. C. (1985). Process plant layout. New York.: Halsted Press.
Modarres, M. (2006). Risk analysis in engineering: techniques, tools and trends. United
States: Taylor and Francis Group.
Molag, M. Schoten, H., & Powell-Price, M. (2001). Effective applications of fluid curtains
to mitigate incidental gas releases. Loss Prevention and Safety Promotion in the
Process Industries, pp. 1051–1058.
Patsiatzis, D. I., Knight, G., & Papageorgiou, L. G. (2004). An MILP approach to safe pro-
cess plant layout. Trans IChemE Part A: Chemical Engineering and Design, 82(A5),
579–586.
Penteado, F. D., & Ciric, A. R. (1996). An MINLP approach for safe process plant layout.
Industrial and Engineering Chemistry Research, 35(4), 1354–1361.
Sheehof, J. M., & Evans, W. O. (1967). Automated layout design program. The Journal
of Industrial Engineering, 18(12), 690–695.
Spicer, T. O., Havens, J. A., Tebean, P. A., & Key, L. E. (1986). DEGADIS: a heavier-than-
air gas atmospheric dispersion model developed for the US Coast Guard. Proc.
Ann. Conf., Am. Poll. Control Ass.,
Sullivan, D. A., Holdsworth, M. T., & Hlinka, D. J. (2004). Monte Carlo-based dis-
persion modeling of off-gassing releases from the fumigant metam-sodium for
determining distances to exposure endpoints. Atmospheric Environment, 38(16),
2471–2481.
Taylor, J. R. (2005). Hazardous materials release and accident frequencies for process
plant. Taylor Associates ApS. I: 4. 5-4.10.
Teles, J. P., Castro, P. M., & Matos, H. A. (2012). Global optimization of water networks
design using multiparametric disaggregation. Computers and Chemical Engineer-
ing, 40(0), 132–147.
Vázquez-Román, R., Lee, J.-H., Jung, S., & Mannan, M. S. (2010). Optimal facility layout
under toxic release in existing plants. Computers and Chemical Engineering, 34(1),
122–133.
Witlox, H. W. M. (1994). The HEGADAS model for ground-level heavy-
gas dispersion II. Time-dependent model. Atmospheric Environment, 28(18),
2933–2946.
Zeman, O. (1982). The dynamics and modeling of heavier than-air cold gas releases.
Atmospheric Environment, 16(4), 741–751.