You are on page 1of 12

Process Safety and Environmental Protection 163 (2022) 82–93

Contents lists available at ScienceDirect

Process Safety and Environmental Protection


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

Resilience assessment framework for fast response process systems


Bhushan Pawar, Mitchell Huffman, Faisal Khan, Qingsheng Wang

]]
]]]]]]
]]

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 cl e i nfo a bstr ac t

Article history: Resilience is essential to ensure safe and sustainable process operations. It plays a critical role in enabling
Received 29 March 2022 operations in the remote and extreme environments encountered during operations held offshore or in the
Received in revised form 28 April 2022 Arctic. Resilience is a property of the process or system, and considers three distinct characteristics: ab­
Accepted 6 May 2022
sorption, adaptation, and recovery. This work proposes a resilience assessment framework of process
Available online 13 May 2022
systems with fast responses, such as reaction systems. The three characteristics are modelled using system
design variables with covariate consideration. Subsequently, these three characteristics are integrated to
Keywords:
Resilience assess resilience. To enhance the resilience, design changes and operational interventions are explored. The
Reliability and maintainability proposed framework is explained using the assessment of an autocatalytic reactor as a case study. A thermal
Chemical process systems runaway reaction is modelled for resilience, and operational intervention strategy such as adding inhibition
Mathematical modeling is tested to enhance the resilience of the reactor. It was concluded that as the inhibitor injection time was
decreased from 5.8 min to 1 min, the value of the proposed resilience metric increased from 0.7 to 0.9. This
case study confirms the applicability and effectiveness of the proposed framework.
© 2022 Institution of Chemical Engineers. Published by Elsevier Ltd. All rights reserved.

1. Introduction construction, and electricity distribution (Aidoo et al., 2022;


Fernández-Hernández et al., 2019; Guo et al., 2020; Righi et al., 2015;
Resilience is essential to sustain the safe operation of process Wiig et al., 2020). Therefore, there are multiple definitions of resi­
systems in industry. The frequent disturbances and uncertainties in lience available in the literature (Jain et al., 2018b). For instance,
process systems due to natural disasters, human errors, or technical Woods (2015) illustrated the concept of resilience as the ability to
failures can lead to major economic and human losses. Therefore, rebound from trauma and achieve equilibrium, or the ability to
process industries are keen to make their systems resilient to all adapt to future surprises as the system conditions vary with time.
types of disruptions. By enhancing the resilience of process systems, Perrings (2006) defined the resilience of economic systems as the
losses can be minimized, and profitability can be maximized. ability to withstand market or environmental shocks without losing
Therefore, there has been an increasing focus on research seeking to the capacity to supply essential goods and services. Hollnagel et al.
enhance the resilience of process systems. This relatively new field is (2006) did pioneer work in defining resilience as an engineering
often referred to as resilience engineering. A resilient system should concept from a sociotechnical perspective, describing it as the ability
absorb, adapt to, and recover from disruptions. The term ‘resilience’ to resist change. These definitions were utilized by researchers to
is still not defined very clearly from an engineering perspective, conceptualize and assess resilience from an engineering perspective
however. Most of the work about resilience has been performed by (Azadeh et al., 2014, 2015; Jain et al., 2018a, 2018b, 2019; Moreno-
social scientists and urban planners. Because of this ambiguity, re­ Sader et al., 2019; Salehi et al., 2020; Zarei et al., 2021). These defi­
silience is very difficult to assess mathematically. There are two nitions were then extended to process systems to illustrate the re­
major challenges: proposing an appropriate definition for the term sponse to a disruption. For instance, Jackson (2009) defined
‘resilience’ in the context of the system under consideration and resilience as the ability of a system to adjust its functioning before,
performing the resilience assessment of a system using system during, and following disturbances, so that it can sustain the re­
parameters or variables. quired function under varying conditions. Similarly, Yarveisy et al.
These challenges have been addressed from various research (2020) characterized the resilience of a system based on three ca­
domains such as aviation, health care, railways, manufacturing, pacities: absorptive, adaptive, and restorative capacity. They pro­
posed these capacities based on the system’s behavior during the
intermediate period between disruption and restoration.

Corresponding author.
E-mail address: qwang@tamu.edu (Q. Wang).

https://doi.org/10.1016/j.psep.2022.05.016
0957-5820/© 2022 Institution of Chemical Engineers. Published by Elsevier Ltd. All rights reserved.
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

After a disruption, the performance of any process system tends 2. The proposed framework
to decrease with time. If there are no safeguards or intervention
actions taken to impede the fall in performance, then the system will A framework that models a process system’s performance after
not be functional after some time. However, a resilient system’s the onset of a disturbance during the three phases of resilience is
performance may decrease after a disturbance, but will recover to its proposed. Firstly, the operational variables that determine the safe
initial performance level after a certain time period. Typically, any operating condition of the process are identified. Later, these vari­
resilient system goes through three phases after a disruption ables are used to model the performance of the process system as a
(Azadeh et al., 2014, 2015; Abimbola and Khan, 2019; Salehi et al., dimensionless quantity whose magnitude indicates the level of
2020; Yarveisy et al., 2020). These phases are typically referred to as functionality of the system. Reliability and maintainability concepts
absorption, adaptation, and recovery. This classification of a system’s are used to define the performance function using covariate models.
performance into three phases is based on the system’s response to a Finally, this model is used to quantify resilience of the process
disruption. These phases are also dependent on socio-technical system. This framework is instrumental in determining the resi­
factors such as teamwork, management, and communication, and lience of a process system against a particular disturbance, shock, or
organizational factors such as equipment inspection, maintenance other changing condition. It can be used as a tool to test the resi­
backlog, and experience from past incidents (Baek et. al, 2015; lience of different system designs against various disturbances and
Bergström et. al, 2015; Patriarca et. al, 2018; Thomas et. al, 2019). In therefore make it easier to choose the most resilient design. All steps
prior literature, various frameworks have been proposed to quantify involved in the framework are illustrated in Fig. 1 and are elaborated
resilience using such socio-technical factors. These frameworks in the following sections.
usually implement surveys to gauge the impact of socio-technical
factors on system’s resilience and propose a resilience metric based 2.1. Identifying performance variables and defining failure
on the data obtained through these surveys. The surveys are often
conducted with professionals from the chemical industry to de­ The idea of a resilient system implies that, in the event of a
termine the important socio-technical factors pertaining to resi­ disruption, the system will not go into a complete failure state. To do
lience based on expert judgement (Azadeh et al., 2014, 2015; Yazdi so, an intervention action needs to be taken to continue the system
and Kabir, 2017; Jain et al., 2018a, 2018b, 2019; Moreno-Sader et al., operation at a lower performance level and later recover the system
2019; Rostamabadi et al., 2020; Salehi et al., 2020; Zarei et al., 2021). to its original performance level. The system’s performance is de­
After collecting the survey data, statistical techniques such as ana­ termined by certain key variables, which also dictate the safety of
lytical hierarchy process and fuzzy logic are typically performed to the system’s operation. In this framework, such variables are re­
determine the factors that highly influence the system’s resilience ferred to as performance variables.
(Ahmed and Kilic, 2019). This process helps in prioritization, ulti­ Performance variables are the operating variables that determine
mately improving intervention and restoration strategies. the safe operation of the system. Performance variables of a system
Although the survey methodology begins to identify the factors have limits beyond which the system becomes unsafe to continue
contributing to the system’s resilience, it tends to be subjective. This operation. The system is in failure state if one or more performance
is because the survey responses will depend on individual opinions variables have values which are beyond these limits. Therefore, a
and experiences, which are subject to change. Therefore, there is a failure condition is defined based on the values of performance
need for a general framework for resilience assessment that is variables. Consider a system with n performance variables denoted
strictly based on a system’s dynamics, properties, and historical data. by P1, P2, ....,Pn . Every performance variable has an upper and/or
In this study, a general framework for resilience assessment is pro­ lower limit for safe operation of the system. Therefore, for safe op­
posed by modeling the system’s performance during the absorption, eration of system,
adaptation, and recovery phases. Yarveisy et al. (2020) implemented
a similar approach by modeling the system’s performance using Let F correspond to an event where the system is put into a
reliability of the system to quantify the resilience capacities. Based failure state. F can be represented as,
on this idea, models for the absorption, adaptation, and recovery F : The event when Pi Pimin or Pi Pimax for any i .
performance of the system were proposed using the reliability and Let Ai be the event when Pi Pimin or Pi Pimax for any i , then
maintainability of the system.
F = {A1 A2 … An }
In this proposed framework, the performance models and resi­
lience assessment are solely dependent on system dynamics, safety The criteria for choosing performance variables will depend on a
constraints, and historical maintenance data. Therefore, this resi­ system’s dynamics and properties. For instance, chemical processes
lience assessment lacks the ambiguity that is a major issue with usually have constraints on the process variables such as maximum
survey methodology. The performances for the three phases are allowable pressure or temperature. Such constraints minimize the
modeled using reliability and maintainability of the system. probability of any incident and ensure that the operation can be
Covariate models are used to model the failure and recovery rates of performed safely. The next step after defining the failure condition is
the system as a function of the system’s operating variables. A re­ to model the system’s performance during the three phases.
silience metric is proposed to quantify the system’s resilience based
on the performance models. This framework can also be used to 2.2. Modeling system performance
assess the resilience of a process system against a disruption for
different design configurations. A system’s design plays an im­ After a disruption, a resilient system goes through three phases:
portant role in the intervention action of the system after a dis­ absorption, adaptation, and recovery, as shown in Fig. 2. The sys­
ruption. Using this framework, the resilience metric for different tem’s performance in each phase is distinct and contributes to the
design configurations can be analyzed by varying the design vari­ overall resilience of the system. It is therefore necessary to model
ables of the system, and the design with the maximum resilience can the performance of the system during each of the three phases for
be chosen. The applicability of the proposed framework is tested for resilience assessment. There have been other studies that quantify
a batch reactor system with a thermal runaway condition. absorption, adaptation, and recovery performance of a system, but

83
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

Fig. 1. The proposed framework for resilience assessment.

mathematical models for the system’s performance during all


phases, taking into consideration the theoretical definitions of ab­
sorption, adaptation, and recovery properties from previous lit­
erature.

2.2.1. Absorption performance modeling


Absorption has been associated with the property of resilience in
various research fields such as ecology, sociology, and civil infra­
structure (Henry and Emmanuel Ramirez-Marquez, 2012; Holling,
1973; Nan and Sansavini, 2017). It is defined as the ability of the
system to effectively absorb disturbances without affecting its per­
formance (Henry and Emmanuel Ramirez-Marquez, 2012; Pawar
et al., 2021). Ideally, a resilient system would absorb all disturbances
Fig. 2. A typical performance curve for a disrupted system. while maintaining a constant performance level without any need
for an intervention action. However, it is very difficult to design a
system that would absorb all disturbances. Therefore, a decreasing
they are highly dependent on expert judgement obtained through performance is observed for most systems during the absorption
surveys and questionnaires (Abimbola and Khan, 2019; Jain et al., phase, as shown in Fig. 2. The decrease in the system’s performance
2018a, 2018b, 2019; Moreno-Sader et al., 2019). In this paper, we until the time of intervention determines the absorption capability
have focused on resilience assessment of the system considering of the system (Yarveisy et al., 2020). A large reduction in the system’s
only the design and operational variables of the system. The system’s performance during the absorption phase implies that the absorp­
performance (Q ) is modeled as a dimensionless quantity with a tion capacity of the system is low. Therefore, system design should
magnitude between 0 and 1, as shown in Fig. 2. The value of Q prior include minimization of the drop in system performance after a
to disruption of the system is 1, which indicates that the system was disruption. In order to achieve this goal, it is important to quantify
performing its function as expected. Q then decreases after the the absorption capacity of the system. Yarveisy et al. (2020) pro­
disruption, indicating a drop in performance, and later increases posed simple and robust metrics for quantifying absorption capacity
during the recovery phase. The following sections propose using the concept of reliability. They defined the absorption capacity

84
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

as the ratio of a system’s reliability at the time when intervention n

begins to the reliability when disruption occurred. abs (t) = o (t) exp ai fi (Pi (t))
i=1 (6)
A similar approach to Yarveisy et al. (2020) has been adopted to
model the absorption performance. After a disruption, the perfor­ ai ’s are the regression coefficients that represent the impact of
mance variables deviate from their normal operating values or range each performance variable on the failure rate. There are simple
and approach their limits. As the performance variables approach linear and exponential models that can alternatively be used when
their limits, the probability of system failure increases, and the re­
abs is constant (Ebeling, 2004). These situations can be interpreted
liability of the system decreases. Ebeling (2004) defined reliability as as a special case scenario of Eq. (6).
the probability that a system will perform its function for a given
period when used under stated operating conditions. The function of
the system is to continue its operation safely without going into a 2.2.2. Adaptation performance modelling
failure state. Reliability incorporates the effects of disruption on the Adaptation is a crucial aspect of a resilient system (Francis and
system’s performance variables. Therefore, it can be used as a Bekera, 2014). A system undergoes an adaptation phase when it is
measure of system performance during the absorption phase. The unable to absorb all external disruptions. If the disruptions are se­
absorption performance (Q abs ) is defined as, vere and beyond the absorption capacity of the system, then the
Q abs (t) = R (t) t [t1, t2] (1) system adjusts itself to adapt to the disruptions and continue its
operation at a lower performance level. Smit and Wandel (2006)
R (t) is the reliability of the system after the disruption, t1 is the defined adaptation from a human systems perspective as the ability
time of disruption, and is the time at which intervention action is of the system to undergo a process or action to better cope with
taken to prevent system failure. If tF is the time when a system changing factors such as risk, hazard, or stress. Adger (2006) also
failure event occurs, then defined adaptation as the capacity of the system to self-organize and
R (t) = Pr {t tF } (2) adapt to changing circumstances. Therefore, in accordance with
these definitions, it can be concluded that adaptation is the ability of
Here, Pr {t tF } is the probability that the system will perform its the system to readjust itself according to external disruptions and
operation safely over the time period t . The reliability of a system continue its operation without going into failure state.
can also be defined as follows, using its failure rate or hazard rate The idea of readjustment implies that some intervention is
function ( ). needed to adapt to the disruption and stop the system from reaching
t the failure state. The intervention can be automated such as with a
R (t) = exp (t ) dt level controller or pressure controller, or it can be from a human.
0 (3)
When the system adapts to the disruption and operates at a lower
(t) gives an instantaneous rate of failure of the system at any performance level, it is implied that the performance variables are
time t (Ebeling, 2004). Using Eqs. (1) and (3), the absorption per­ no longer approaching their limiting values. Instead, the perfor­
formance of the system is represented as mance variables P1, P2, …, Pn of the system become constant after the
t system adaptation. However, the system is not restored to its ori­
Q abs (t) = exp abs (t ) dt t [t1, t2] ginal configuration during the adaptation phase and is therefore still
t1 (4)
at risk of failure. As the system operates in the adaptation phase at a
abs (t) is the failure rate of the system during the absorption lower performance level, the reliability of the system decreases and
phase. the probability of failure increases. However, the rate of probability
After defining the absorption performance, the next step is to of failure increase during the adaptation phase is much lower than
model the failure rate ( ). During the absorption phase, the failure during the absorption phase. This is because of the intervention
rate ( ) of the system can be modeled using the system’s perfor­ action, which helps to decrease the failure rate severely. Therefore,
mance variables. The failure rate of the system increases as the the failure rate during the adaptation phase ( adap ) is very low
system approaches its failure state. Failure state is defined as the compared to the failure rate during the absorption phase ( abs ).
condition where performance variables reach their limiting values
for safe operation. Therefore, the failure rate of the system will in­ adap abs (7)
crease as the performance variables approach their limiting values.
Covariate models are the most common technique used to model the Using Eq. (7), it can be concluded that the system’s reliability
failure rate of the system as a function of system’s covariates, or during the adaptation phase will decrease more gradually than
explanatory variables (Ebeling, 2004). These are variables, which during the absorption phase, where the reliability decreases sud­
have an obvious correlation with the failure rate of the system. In denly over a short period. The adaptation performance is modeled
this case, the covariates are the performance variables of the system. using reliability, similar to the absorption performance. A linear
Therefore, the failure rate of the system during absorption phase model for adaptation performance is proposed as shown below.
( abs ) can be modeled using the covariate approach.
R (t3) R (t2)
Q adap (t) = Q abs (t2) + (t t2) t [t2, t3]
abs (t) = f (P1 (t), P2 (t), …, Pn (t)) (5) t3 t2 (8)
The exact form of f (P1, P2, …, Pn) can be determined by per­
Because R (t) does not decrease drastically during the adaptation
forming experiments to analyze the relationship between the per­
phase, the adaptation performance curve has been approximated as
formance variables P1, P2, …, Pn and the failure rate abs . However, if
a straight line with a slope equal to the average rate of change of
the relationship is unknown, there are simple models such as a
proportional hazards model that can be used to model abs (Barker R (t). The slope ( R (t 3)
t3
R (t 2)
t2 ) determines the adaptability of the
and Baroud, 2014). The proportional hazards model states that the system. The magnitude of the slope should be close to zero for highly
failure rate ( abs ) is a function of the time driven baseline hazard adaptable system. This can be achieved by taking the recovery action
function ( o ) and the exponential summation of the covariates as quickly as possible, so that R (t3) R (t2). Eq. (8) can also be ex­
(functions of performance variables), leading to Eq. (6). pressed in terms of adap and abs .

85
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

t3 t
R (t3) = exp (t ) dt M (t) = 1 exp µrec (t ) dt
t1 t3 (11)
t2 t3
= exp abs (t ) dt adap (t ) dt µrec is the recovery rate of the system, which is analogous to the
t1 t2
repair rate. M (t) demonstrates the proximity of the system’s current
t2
state to the complete restoration state. In other words, it represents
R (t2) = exp abs (t ) dt the efforts and resources needed to recover the system to its initial
t1
state. The recovery performance should demonstrate the ease with
Q adap (t) which the system can be restored to the state prior to the disruption
(Abimbola and Khan, 2019). Therefore, the recovery performance has
been modeled as a linear function of M (t) as shown below.
= Q abs (t2)
Q rec (t) = c1M (t) + c2
+
exp ( t2
t1 abs (t ) dt
t3
t2 adap (t ) dt ) exp ( t2
t1 abs (t ) dt ) (12)

t3 t2 The constants c1 and c2 can be determined using the continuity


criteria:
(t t2) t [t2, t3] (9)
• At t = t3, Q rec (t3) = Q adap (t3) , and
The adaptation phase may not be observed, or it may only occur • As t , Q rec (t) 1, because if the recovery action is performed
for a short period of time, in some systems which are highly reactive for infinite time, then eventually the system will be completely
to intervention. This is because a system will enter the recovery restored to its initial state.
phase immediately after intervention instead of adapting to the
disruption if a system is highly reactive to intervention. Using the above conditions,
t
2.2.3. Recovery performance modelling Q rec (t) = Q adap (t3) + (1 Q adap (t3)) 1 exp µrec (t ) dt
t3
Recovery is the process of restoring a system’s performance to its
original levels. For a resilient system, the final goal is to recover the t [t3, ) (13)
system after a disruption. The system transitions to recovery phase Eq. (13) represents the model for the recovery performance of the
after it has absorbed maximum possible disruptions during the ab­ system, but it only reaches its initial performance value of ‘1’ after
sorption phase and has successfully adapted to them during the infinite time. This result is expected as it is not possible to fully re­
adaptation phase. There are many research studies, which focus on store or repair any system to its initial state before the disruption.
the mathematical modeling of the recovery performance of civil Therefore, this model represents how close the system is to its initial
infrastructures such as buildings or bridges when they are disrupted state after the recovery action.
by extreme natural events such as earthquakes, floods, and hurri­ Recovery rate is similar to the repair rate of the system. However,
canes (Gardoni et al., 2002; Sharma et al., 2018; Nocera et al., 2019). the repair rate is the rate at which a system is repaired or restored
Probabilistic models are most common method used to define the from a failure state to its initial state, whereas the recovery rate is
recovery curve based on the historical data of recovery activities the rate at which a system is restored to its initial state from a low
such as time needed for damage inspection, mobilization, and re­ performance state. A proportional hazards model can also be used to
pairs (Sharma et al., 2018). Cimellaro et al. (2006) proposed prob­ model recovery rate, similar to the failure rate (Barker and Baroud,
abilistic recovery models for the seismic failure probability of civil 2014). The covariates may or may not include performance variables,
infrastructures. They proposed linear, trigonometric, and ex­ however, because the recovery actions usually depend on socio­
ponential models for the recovery curve based on the time required technical factors such as the number of workers and their experience
to repair the structures after seismic damages. Barker and Baroud or skills. The impact of such factors can be analyzed by using them as
(2014) took a similar approach to model infrastructure recovery covariates for modeling the recovery rate as shown below.
from power outages caused by thunderstorms across the United
n
States. They used maintainability as a measure of recovery, and they
µrec (t) = µo (t) exp bi Ci (t)
implemented a proportional hazards model to define the repair rate (14)
i=1
by considering covariates such as number of customers, loss, and
geographical location. µo is a baseline recovery rate function which is used as a re­
This framework proposes a similar approach for modeling system ference, bi ’s are the regression coefficients, and Ci ’s are the covariates
recovery performance. Maintainability is defined as the probability that can be functions of performance variables Pi ’s and/or socio­
that a failed system can be repaired or restored to a specified state technical factors.
within the period of time in which maintenance activities are per­
formed (Ebeling, 2004). In the case of a resilient system, the system 2.3. Resilience assessment
is recovering from a lower performance state to the initial perfor­
mance state. During this phase, the performance variables There are many studies where resilience assessment of a system
P1, P2, …, Pn are restored to their initial values from before the dis­ is performed using the area integral of the performance curve Q (t)
ruption. Therefore, recovery is proposed as a quantity that is ana­ over the time interval from disruption to restoration (Bruneau and
logous to the maintainability of the system. First, the probability that Reinhorn, 2007; Bonstrom and Corotis, 2016; Sharma et al., 2018).
the system operating at a lower performance level can be repaired or According to these studies, the resilience metric is defined as
restored to its initial performance level within time t is denoted t4
as M (t) . t1
Q (t) dt
Resilience = t4
M (t) = Pr {tR t} (10) t1
dt (15)

is the time needed to repair or recover the system to its initial In Eq. (15), resilience is quantified as the ratio of the area under
state. By the definition of maintainability given by Ebeling (2004) the actual performance curve Q (t) to the area under the ideal per­
and Barker and Baroud (2014), M (t) can also be defined as formance curve over the time interval from disruption to restoration.

86
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

The goal of resilience assessment is to propose a mathematical the reactor. This condition leads to a rapid and uncontrolled rise in
quantity that represents the ability of a system to deliver perfor­ the reactor temperature. Such conditions can also result in an ex­
mance as close as possible to ideal performance. Ideally, the per­ plosion because of the sudden increase in temperature over a short
formance value of the system would remain one at any time. period of time. Therefore, it is necessary to design this system to be
However, most systems will experience a decrease in their perfor­ resilient to disruptions such as a loss of control loop. One way to
mance after a disruption as shown in Fig. 2, and an eventual increase achieve this goal is to modify the batch reactor design to include a
after intervention and recovery actions. A resilience metric is needed reaction inhibitor injection mechanism. Reaction inhibition slows
to represent the proximity of the actual performance curve to the the rapidly increasing reactor temperature, preventing a complete
ideal performance curve. The area integrals of the actual and ideal failure condition due to explosion, and creating an opportunity to
performance curves can demonstrate how close the actual system is regain safe operation. Then, the batch reactor can be cleaned, and
to an ideal resilient system. If the ratio in Eq. (15) is high (close to 1), the control loop can be repaired or reinstalled to restore the system
then it can be concluded that the system’s behavior is very close to to original performance level. However, there are multiple design
ideal. Therefore, the system’s resilience to the disruption will also factors that can impact the resilience of the system. For instance, the
be high. time of injection of the inhibitors tinj plays an important role in
Eq. (15) also implies that to maximize the resilience of the enhancing the system’s resilience. Earlier injection of inhibitors
system, the area under the performance curve Q (t) should be should lead to a lower risk of explosion and higher resilience. This
maximized. This equation is also instrumental in analyzing the ef­ hypothesis is verified by applying the proposed resilience assess­
fects of various operating and design variables on a system’s resi­ ment framework to this batch reactor system and thereby proving its
lience. Once the impact of these variables on resilience is analyzed, applicability to process systems.
the system’s design can be modified to maximize the resilience, and Ethyl acetate is used as a solvent and azobis-isobutyronitrile is
the intervention and recovery actions can be planned accordingly. used as an initiator (I) for the polymerization reaction. The initiation,
propagation, and termination kinetics of the uninhibited poly­
merization of MMA are described by reactions I – V.
3. Application of the framework: case study of batch reactor
Initiation
Thermal runaway incidents occur predominantly in batch reactor kd
I 2R· (I)
operations. Liu and Wilhite (2019) developed a model for transient,
non-isothermal, well-mixed batch reactors for a free radical poly­ ki
R· + M P1· (II)
merization of methyl methacrylate (MMA). Within this study, they
investigated reaction inhibitor injection as a procedure for thermal Propagation
runaway mitigation. The impact of injection time on the mitigation kp
procedure was then analyzed. This batch reactor model will be used Pn · + M Pn + 1· (III)
as a case study to prove the applicability of our resilience assessment Termination
framework.
ktc
Pm · + Pn · Dm + n (IV)
3.1. Process description ktd
Pm · + Pn · Dm + Dn (V)
This system consists of a batch reactor with a cooling jacket, as
Here, M, R, P, and D represent monomers, initiator radicals,
shown in Fig. 3. The MMA polymerization reaction is the process
polymer radicals, and dead polymer products respectively. m and n
being analyzed. The system is set up in such a way that there is
are positive integers denoting chain length. The reaction mechanism
insufficient cooling which leads to a thermal runaway scenario. It is
involves free radical initiation, propagation by live polymer radicals,
assumed that there is a control loop failure, which results in the
and termination by combination of two radicals to form dead
reaction heat generation rate being greater than the cooling rate of
polymer products. All side reactions are neglected to investigate the
thermal runaway phenomenon (Liu and Wilhite, 2019). The rate
expressions for the uninhibited polymerization reaction are ob­
tained by using the mass balance. The definitions of the variables
and parameters are shown in Table 1.
d [M]
= k p [M] o
dt (16)

d [I]
= kd [I]
dt (17)

2f [I] kd
o =
kt (18)
These expressions are obtained using the quasi-steady state as­
sumption for the polymer radical concentration. o is the zeroth
moment of free radical concentration, which represents the level of
reactivity inside the reactor. The mathematical model for reactor
temperature is obtained by using the energy balance as follows.
dT Ua ( Hr)
= (Tc T) + kp [M] o
dt Cp V Cp V (19)
After the addition of inhibitors, the rate expressions and the re­
Fig. 3. A batch reactor system for MMA polymerization. actor model get modified slightly because of inhibition action,

87
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

Table 1 Table 2
Process variables/parameters and their definitions. Parameter values for the batch reactor simulation.

Variables/Parameters Definitions Parameters Values

T Reactor temperature Tc 300 K


To Initial reactor temperature [X]o 0.55 mol/L
Tc Temperature of cooling fluid [I]o 0.055 mol/L
tinj Time of injection of inhibitor [M]o 10 mol/L
[X] Inhibitor concentration kpo 4.41 × 105 s-1
[X]o Initial inhibitor concentration kto 2.6 × 108 L/mol·s
[I] Initiator concentration kdo 2 × 1015 s-1
[I]o Initial initiator concentration Ep 4348 J/mol
[M] Methyl methacrylate (MMA) concentration Ed 129,615 J/mol
[M]o Initial MMA concentration Et 6393 J/mol
kp , kx , kt , kd , kpo, kto, kdo Reaction rate constants f 0.5
Ep, Ed, Et Reaction activation energies ρ 0.895 g/cm3
f Efficiency factor of initiator Cp 1950 J/kg·K
ρ Density U 490 W/m2·K
Cp Specific heat capacity of reactor contents a 0.063 m2
U Overall heat transfer coefficient V 1L
a Heat transfer area Hr 5.858 × 104 J/mol
V Reactor volume
Hr Heat of polymerization reaction

leaks. Therefore, according to the definition of performance vari­


ables, reactor temperature is used as a performance variable in this
except for Eq. (16). An inhibition reaction is involved along with the case study. The system failure condition is defined as the point when
reactions described in I – V. The kinetics of the inhibition reaction reactor temperature exceeds the critical temperature of MMA;
can be expressed as follows. therefore, the critical temperature is the upper limit of safe reactor
kx
operating temperature. This is because at the critical temperature,
Pn · + X Dn (VI) all reactor contents are in the gaseous phase, and therefore the risk
The rate expressions and reactor model are obtained using the of an explosion or leak is extremely high. The critical temperature of
mass and energy balance equations. For the inhibited condition, the MMA is 567.2 K and the failure condition is described as follows.
zeroth moment of inhibitor concentration ( oinh ) is different from the
T 567.2 K (25)
uninhibited condition. It was obtained using the same quasi steady
state assumption (Liu and Wilhite, 2019). Fig. 4 shows the reactor temperature profile for uninhibited re­
d [M] inh
action and for when inhibitors are injected at time tinj = 5.8min . The
= k p [M] o temperature profile for the inhibited reaction reaches the failure
dt (20)
condition at t = 5.8min when the reactor temperature T = 567.2K , as
d [X] inh shown in Fig. 4. This implies that the intervention action (inhibitor
= k x [X] o
dt (21) injection) should be taken before t = 5.8min to prevent the tem­
perature from reaching its upper limit. Therefore, to ensure the re­
inh k x [X] + (k x [X])2 + 8f [I] k d kt silience of the system during the thermal runaway situation,
o =
2kt (22)
tinj < 5.8 min (26)
dT Ua ( Hr) inh
= (Tc T) + k p [M] o
dt Cp V Cp V (23)
Liu and Wilhite (2019) analyzed the reactor model for multiple
scenarios by varying the kinetic parameters of the inhibition reac­
tion. They analyzed the temperature profile of the reactor for dif­
ferent combinations of k x and kp . In this case study, however, only
one scenario of inhibition reaction is analyzed. Therefore, the rate
constant for the inhibition reaction here is assumed to be
k x = 0.01k p (24)
The values of the system’s parameters were obtained from Liu
and Wilhite’s work. The initial conditions for solving differential Eqs.
(16) – (24) and the values of the system’s parameters are shown in
Table 2.

3.2. Defining the failure condition

According to the framework presented, the first step is to identify


the performance variables and define the failure condition of the
system. Batch reactor systems have a predefined maximum allow­
able temperature (Westerterp and Molga, 2006; Stoessel, 2009; and
Chen et al., 2021). If the reactor temperature increases beyond the Fig. 4. Reactor temperature for uninhibited reaction and reaction with inhibitors
maximum allowable value, then there is a high risk of explosion or at tinj = 5.8min .

88
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

3.3. Modeling the performance tinj


Q abs (t) = exp abs (t ) dt t [0, tinj]
0 (27)
A batch reactor system is highly reactive to intervention actions
The failure rate of the system abs (t) is modeled using the pro­
such as injection of inhibitors. The reactor temperature decreases
portional hazards model because the data for the failure rate of a
immediately when a sufficient quantity of inhibitors is injected at
batch reactor system with MMA polymerization reaction in a dis­
the appropriate time. This is not true for all quantities of inhibitors
rupted state was not available in prior literature. Therefore, covari­
and injection times. Liu and Wilhite (2019) demonstrated that the
ates that characterize the thermal runaway condition of a batch
thermal runaway condition can be avoided only by using a certain
reactor were identified. Thermal runaway scenarios are character­
range of values of inhibitor concentrations and injection times. They
ized by a rapid and uncontrolled rise in reactor temperature
proved that the reactor temperature would decrease immediately
(Westerterp and Molga, 2006; Stoessel, 2009; Wang, 2010; and
after the inhibitor injection, but that if the amount of inhibitors is
Kummer et al., 2020). Ni et al. (2016) proposed that a positive rate of
not enough to stop the polymerization reaction, then the reactor
temperature would increase again after some time and a thermal
change of reactor temperature ( ) can be used as a criterion for
dT
dt
dT
runaway condition would still occur. Similarly, if the injection time is identifying thermal runaway. A high value of implies that the
dt
too high, the reactor temperature will increase eventually, and the reactor temperature increases rapidly in a very short period. If dT is
dt
reactor will still enter a thermal runaway situation. In this case high, then the time needed for the reactor temperature to reach its
study, appropriate parameter values have been chosen from Liu and upper limit is low. Therefore, as dT increases, the failure rate of the
Wilhite’s work to model the reactor temperature during a thermal dt
system also increases. Hence, a proportional hazards model is used
runaway and inhibition condition.
to model the system failure rate using dT .
The reactor temperature profile before and after time tinj was dt

obtained by solving Eqs. (16) – (19) and (20) – (24) respectively in dT (t)
MATLAB, as shown by the dashed curve in Fig. 4. After obtaining the dt
abs (t) = o (t) exp
temperature profile, the performance curve of the system was ob­ ( )dT
dt baseline (28)
tained according to the proposed framework. In this case, only ab­
dT (t)
sorption and recovery phases were observed because the reactor dt
The covariate used in Eq. (28) is . A baseline case is used
temperature decreases immediately after inhibitor injection. After dT
dt baseline
an intervention action was taken, the performance variable (T ) of the
as reference for the proportional hazard model. An assumption is
system stopped approaching its upper limiting value and started
made for the baseline case that the temperature increases at a

( ) and the corresponding baseline


decreasing. This implies that the system was highly reactive to the
intervention action and went into recovery phase immediately after constant rate constant ( ) dT
dt baseline
the intervention action. When the temperature was no longer in­
creasing, the batch reactor was cleaned, the control loop was re­
failure rate o (t) is also constant. o (t) and ( )
dT
dt baseline
are calculated
from Fig. 4 using the uninhibited reactor temperature profile. As the
paired or reinstalled, and the operating conditions were restored to
data for the failure rate is not available in the literature, an as­
their initial values. Thus, after completing all recovery actions, the
sumption is made that the mean time when the reactor temperature
system is restored to its normal operating condition as it was before
reaches its critical value of 567.2 K is approximately 5.8 min.
the disruption.
Therefore, the mean time to failure (MTTF) for the baseline case is
It should be noted that all batch reactor operations are unsteady
5.8 min.
state operations. This implies that the reactor temperature will al­
ways rise and then decrease as the reaction progresses for each 1 1 1
o (t ) = = = 0.17 min
operation cycle when operated at normal conditions. However, this MTTF 5.8 (29)
does not mean that the performance Q (t) of the system also de­
creases and increases according to the changes in temperature. This ( )
dT
dt baseline
is also approximated by assuming that the rise in
is because the system is still working at normal operating condi­ temperature will be essentially linear. Therefore, ( )
dT
dt baseline
is cal­
tions, and therefore it is known that the safety limits of the oper­ culated from Fig. 4 as the average slope of the increasing temperate
ating variables will not be exceeded. This means that operation is curve.
safe and at all times, Q (t) = 1 when the system is operated at normal
operating conditions without any disruptions. dT
= 35 K / min
However, if the batch reactor system experiences disruptions, dt baseline (30)
then its reliability will decrease with time immediately after the
disruption. This implies that the system’s performance Q (t) will also
3.3.2. Modeling recovery performance (Q rec)
decrease during the absorption phase. Therefore, Q (t) 1 when the
After intervention at time t = tinj , the reactor temperature is no
system experiences disruptions such as insufficient cooling. Thus,
longer increasing. Therefore, recovery actions can be commenced
the batch reactor system’s performance is not dependent on the
after t = tinj . As mentioned in the previous section, recovery actions
reactor temperature when the system is operated at normal oper­
can include activities such as cleaning the batch reactor to get rid of
ating conditions without any disruptions. Temperature is only used
the inhibition reaction products, repairing or reinstalling the cooling
as a tool to model the absorption performance or the reliability of
system control loop, and loading the reactor with reactants. The
the system to represent its proximity to the failure condition after a
recovery performance of the system can be given by specifying Eq.
disruption has occurred.
(13) as follows.

3.3.1. Modeling absorption performance (Q abs) t


Q rec (t) = Q abs (tinj) + (1 Q abs (tinj)) 1 exp µ rec (t ) dt t
The loss of control loop occurs at time t = 0s , therefore, the tinj

system will enter the absorption phase at t = 0 s. The intervention [tinj, ) (31)
action being taken is the injection of inhibitors at t = tinj . Thus, the
system will remain in the absorption phase from t = 0s to t = tinj . The recovery rate will depend on sociotechnical factors such as
The absorption performance can be modeled by specifying Eq. (4) to the number of workers and their experience or skills. These factors
create Eq. (27) as follows. will be unique for every system and can be obtained from the

89
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

Fig. 5. Temperature profiles and their corresponding performance for (a) tinj = 5.8min , (b) tinj = 4min , (c) tinj = 2min , and (d) tinj = 1min .

90
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

historical maintenance data of the system. However, models of the


recovery rate of a batch reactor for MMA polymerization reaction are
not available in literature. Therefore, this case study assumes that
the system’s recovery rate µrec (t) is constant. The mean time to re­
pair (MTTR) of the system is assumed to be 100 min. It is also as­
sumed that the recovery actions were completed at time t = 500
minutes. Therefore,
1 1 1
µrec (t) = = = 0.01 min
MTTR 100 (32)

3.4. Resilience assessment of the batch reactor system

After modeling the performance curve Q (t), the resilience of the


system can be calculated using Eq. (14) as follows.
t4
0
Q (t) dt
Resilience = t4
0
dt (33)
t4 is the time when the recovery actions are completed, and the
system is restored to its normal operating condition. Eqs. (27) – (33)
were used to calculate the resilience metric for different inhibitor
injection times.

3.5. Results and discussion

The system’s performance Q (t) was plotted for different inhibitor


injection times (tinj ). Fig. 5 shows the reactor temperature profiles
and their corresponding performance plots. The temperature pro­
files for uninhibited and inhibited conditions are shown to compare
the system’s response to different values of tinj .
Disruption occurs at t = 0 and the system goes into absorption
phase. According to Eq. (27), the absorption performance is defined
by the reliability of the system. When tinj = 5.8 min, the reactor
temperature reaches its limiting value of 567.2 K and therefore, the
absorption performance of the system decreases to nearly zero. This
is because dT increases tremendously at t = 5.8 min, which means
dt
that the failure rate also increases, and the reliability of the system
decreases. As tinj is decreased from 5.8 min to 1 min, the absorption
performance of the system improves, decreasing the drop in sys­
tem’s performance from t = 0 to t = tinj . For instance, at tinj = 1 min, Fig. 6. Summary of (a) reactor temperature profiles, and (b) performance curves at
tinj = 5.8min, 4min, 2min, and 1min .
Q (t) drops from 1 to 0.82, whereas at tinj = 5.8 min, Q (t) drops from
1 to 0.001. Therefore, the area under the Q (t) curve increases as tinj
decreases. This implies that the system becomes more resilient Table 3
Resilience assessment for different tinj values.
when the inhibitors are injected earlier. This can also be observed in
Fig. 5. The system’s response to different values of tinj can be com­ Design variable, tinj (min) Resilience
pared and analyzed from Fig. 6. The performance curve Q (t) ap­ 5.8 0.7
proaches the ideal performance curve as tinj decreases from 5.8 min 4 0.8
to 1 min. Also, the reactor temperature profiles were analyzed for 2 0.9
tinj = 5.8, 4, 2, and 1 min. It can be observed that the maximum re­ 1 0.9

actor temperature decreases when the inhibitors are injected earlier.


For instance, at tinj = 5.8 min, the maximum reactor temperature
value was 567.2 K (the critical temperature of MMA), whereas at the system becomes more resilient. Therefore, to maximize the re­
tinj = 1 min, the maximum reactor temperature value was 333.3 K. silience of the system, tinj should be minimized as much as possible.
This also implies that the reactor operation becomes safer when the From Fig. 6, it can also be observed that the recovery perfor­
inhibitors are injected earlier. Therefore, it can be concluded from mance curve does not overlap with ideal performance curve exactly.
the reactor temperature profiles and their corresponding Q (t) curves This implies that the value of recovery performance is 1 only at in­
that the system becomes more resilient when the intervention ac­ finite time. This is accurate, as in application it is extremely difficult
tion is taken earlier. to restore any system to its exact initial state. Therefore, the pro­
Only two phases, namely, absorption and recovery, were ob­ posed maintainability-based model of the recovery performance
served for this case study. Because the system is highly reactive to gives an accurate representation of the progress from the recovery
intervention action, the adaptation phase is very short lived and actions with time. However, in some cases, it is possible to restore a
occurs at t = tinj . The inhibitor injection time tinj was treated as a system to its initial state depending on the system and recovery
design variable in this case study. The resilience metrics for actions. If the damaged components of the system are replaced with
tinj = 5.8, 4, 2, and 1 min were calculated and shown in Table 3. The new and more efficient components, then the system might be re­
calculated resilience metrics further prove that as tinj is decreased, stored to its initial state or even better state. Recovery also depends

91
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

on financial aspects such as available budget for restoration ex­ Azadeh, A., Motevali Haghighi, S., Salehi, V., 2015. Identification of managerial shaping
penses. In most cases, the restoration of the system to initial or even factors in a petrochemical plant by resilience engineering and data envelopment
analysis. J. Loss Prev. Process Ind. 36, 158–166. https://doi.org/10.1016/j.jlp.2015.
better state is very expensive. For instance, critical infrastructure 06.002
systems such as electric power systems usually come under the Azadeh, A., Salehi, V., Ashjari, B., Saberi, M., 2014. Performance evaluation of in­
jurisdictions of governments and therefore, they can afford to make tegrated resilience engineering factors by data envelopment analysis: the case of
a petrochemical plant. Process Saf. Environ. Prot. 92, 231–241. https://doi.org/10.
huge investments in the recovery actions for such systems (Yodo and 1016/j.psep.2013.03.002
Wang, 2018; Sarker and Lester, 2019). However, if the restoration Baek, J.S., Meroni, A., Manzini, E., 2015. A socio-technical approach to design for
costs are more than the available budget, then the system will have community resilience: a framework for analysis and design goal forming. Des.
Stud. 40, 60–84. https://doi.org/10.1016/j.destud.2015.06.004
to be restored to a performance level that is lower than its initial Barker, K., Baroud, H., 2014. Proportional hazards models of infrastructure system
state. recovery. Reliab. Eng. Syst. Saf. 124, 201–206. https://doi.org/10.1016/j.ress.2013.
12.004
Bergström, J., Van Winsen, R., Henriqson, E., 2015. On the rationale of resilience in the
4. Conclusions
domain of safety: a literature review. Reliab. Eng. Syst. Saf. 141, 131–141. https://
doi.org/10.1016/j.ress.2015.03.008
In this work, a framework for resilience assessment was devel­ Bonstrom, H., Corotis, R.B., 2016. First-order reliability approach to quantify and im­
oped by integrating reliability and maintainability concepts. First, prove building portfolio resilience. J. Struct. Eng. 142, 1–12. https://doi.org/10.
1061/(asce)st.1943-541x.0001213
the key operating variables that determine the safety of the system’s Bruneau, M., Reinhorn, A., 2007. Exploring the concept of seismic resilience for acute
operation were identified. These are performance variables with care facilities. Earthq. Spectra 23, 41–62. https://doi.org/10.1193/1.2431396
limiting values beyond which the system’s operation becomes un­ Chen, Q., Ni, L., Jiang, J., Wang, Q., 2021. Modeling of runaway inhibition in batch
reactors using encapsulated phase change materials. Renew. Energy 170,
safe. Second, a failure condition was defined based on the limiting 387–399. https://doi.org/10.1016/j.renene.2021.01.132
values of the performance variables. Finally, mathematical models Cimellaro, G.P., Reinhorn, A.M., Bruneau, M., 2006. Quantification of seismic resilience.
were proposed for the absorption, adaptation, and recovery perfor­ 8th US Natl. Conf. Earthq. Eng. 6, 3208–3217.
Ebeling, C.E., 2004. An Introduction to Reliability and Maintainability Engineering.
mance of a system in the event of a disruption. Covariate models Tata McGraw-Hill Education.
were used to analyze the effect of performance variables Pi on the Fernández-Hernández, I., Walter, T., Alexander, K., Clark, B., Châtre, E., Hegarty, C.,
system’s performance Q (t), and a resilience metric was defined Appel, M., Meurer, M., 2019. Increasing International Civil Aviation Resilience: A
Proposal for Nomenclature, Categorization and Treatment of New Interference
using the area integral of Q (t). A case study was performed to apply
Threats. Proceedings of the 2019 International Technical Meeting of The Institute
the framework on a batch reactor system with an MMA poly­ of Navigation, Reston, Virginia, pp. 389–407.
merization reaction to analyze its resilience in the event of a thermal Francis, R., Bekera, B., 2014. A metric and frameworks for resilience analysis of en­
gineered and infrastructure systems. Reliab. Eng. Syst. Saf. 121, 90–103. https://
runaway. The performance Q (t) of the batch reactor system was
doi.org/10.1016/j.ress.2013.07.004
modeled using reactor temperature as a performance variable. Gardoni, P., Der Kiureghian, A., Mosalam, K.M., 2002. Probabilistic capacity models
Inhibitor injection was used as an intervention to prevent the system and fragility estimates for reinforced concrete columns based on experimental
from reaching the failure condition. Therefore, inhibitor injection observations. J. Eng. Mech. 128, 1024–1038. https://doi.org/10.1061/(asce)0733-
9399(2002)128:10(1024)
time tinj was used as a design variable. It was found that the resi­ Guo, Q., Amin, S., Hao, Q., Haas, O., 2020. Resilience assessment of safety system at
lience of the system increased when tinj was decreased. This was also subway construction sites applying analytic network process and extension cloud
verified by obtaining reactor temperature profiles at decreasing va­ models. Reliab. Eng. Syst. Saf. 201, 106956. https://doi.org/10.1016/j.ress.2020.
106956
lues of tinj . This implied that earlier implementation of intervention Henry, D., Emmanuel Ramirez-Marquez, J., 2012. Generic metrics and quantitative
action ensured safer and more resilient operation. This case study approaches for system resilience as a function of time. Reliab. Eng. Syst. Saf. 99,
proved the applicability of the framework. 114–122. https://doi.org/10.1016/j.ress.2011.09.002
Holling, C.S., 1973. Resilience and stability of ecological systems. Annu. Rev. Ecol. Syst.
It should be noted that only the constraints on the performance 4, 1–23.
variables from a safety perspective were considered. Additional Hollnagel, E., Woods, D.D., Leveson, N., 2006. Resilience Engineering: Concepts and
constraints such as economic constraints can also be included for Precepts. Ashgate Publishing, Ltd.
Jackson, S., 2009. Architecting Resilient Systems: Accident Avoidance and Survival and
future work; framing a multi objective optimization problem in Recovery from Disruptions 66 John Wiley & Sons.
which the economic profit as well as the system’s resilience needs to Jain, P., Chakraborty, A., Pistikopoulos, E.N., Mannan, M.S., 2018a. Resilience-based
be optimized. The modeling of failure rates and recovery rates of process upset event prediction analysis for uncertainty management using
bayesian deep learning: application to a polyvinyl chloride process system. Ind.
process systems have not been explored thoroughly. Therefore, the
Eng. Chem. Res. 57, 14822–14836. https://doi.org/10.1021/acs.iecr.8b01069
use of Monte Carlo simulations for modeling failure time distribu­ Jain, P., Mentzer, R., Mannan, M.S., 2018b. Resilience metrics for improved process-risk
tions and failure rates of process systems is another important di­ decision making: Survey. Anal. Appl. Saf. Sci. 108, 13–28. https://doi.org/10.1016/j.
rection for future research. ssci.2018.04.012
Jain, P., Pistikopoulos, E.N., Mannan, M.S., 2019. Process resilience analysis based data-
driven maintenance optimization: application to cooling tower operations. Comput.
Declaration of Competing Interest Chem. Eng. 121, 27–45. https://doi.org/10.1016/j.compchemeng.2018.10.019
Kummer, A., Varga, T., Nagy, L., 2020. Semi-batch reactor control with NMPC avoiding
thermal runaway. Comput. Chem. Eng. 134. https://doi.org/10.1016/j.
The authors declare that they have no known competing fi­ compchemeng.2019.106694
nancial interests or personal relationships that could have appeared Liu, G., Wilhite, B.A., 2019. Model-based design for inhibition of thermal runaway in
to influence the work reported in this paper. free-radical polymerization. Ind. Eng. Chem. Res. 58, 17244–17254. https://doi.
org/10.1021/acs.iecr.9b02007
Moreno-Sader, K., Jain, P., Tenorio, L.C.B., Mannan, M.S., El-Halwagi, M.M., 2019.
References Integrated approach of safety, sustainability, reliability, and resilience analysis via
a return on investment metric. ACS Sustain. Chem. Eng. 7, 19522–19536. https://
Abimbola, M., Khan, F., 2019. Resilience modeling of engineering systems using dy­ doi.org/10.1021/acssuschemeng.9b04608
namic object-oriented Bayesian network approach. Comput. Ind. Eng. 130, Nan, C., Sansavini, G., 2017. A quantitative method for assessing resilience of inter­
108–118. https://doi.org/10.1016/j.cie.2019.02.022 dependent infrastructures. Reliab. Eng. Syst. Saf. 157, 35–53. https://doi.org/10.
Adger, W.N., 2006. Vulnerability. Glob. Environ. Chang. 16, 268–281. https://doi.org/ 1016/j.ress.2016.08.013
10.1016/j.gloenvcha.2006.02.006 Ni, L., Mebarki, A., Jiang, J., Zhang, M., Pensee, V., Dou, Z., 2016. Thermal risk in batch
Ahmed, F., Kilic, K., 2019. Fuzzy analytic hierarchy process: a performance analysis of reactors: theoretical framework for runaway and accident. J. Loss Prev. Process
various algorithms. Fuzzy Sets Syst. 362, 110–128. https://doi.org/10.1016/j.fss. Ind. 43, 75–82. https://doi.org/10.1016/j.jlp.2016.04.004
2018.08.009 Nocera, F., Gardoni, P., Cimellaro, G.P., 2019. Time-dependent probability of exceeding
Aidoo, I., Fugar, F., Adinyira, E., Ansah, N.B., 2022. Understanding the concept of re­ a target level of recovery. ASCE-ASME J. Risk Uncertain. Eng. Syst. Part A Civ. Eng.
silience in construction safety management systems. In: Gorse, C., Scott, L., Booth, 5, 04019013. https://doi.org/10.1061/ajrua6.0001019
C., Dastbaz, M. (Eds.), Climate Emergency – Managing, Building, and Delivering Patriarca, R., Falegnami, A., Costantino, F., Bilotta, F., 2018. Resilience engineering for
the Sustainable Development Goals. Springer International Publishing Cham, pp. socio-technical risk analysis: application in neuro-surgery. Reliab. Eng. Syst. Saf.
217–233. 180, 321–335. https://doi.org/10.1016/j.ress.2018.08.001

92
B. Pawar, M. Huffman, F. Khan et al. Process Safety and Environmental Protection 163 (2022) 82–93

Pawar, B., Park, S., Hu, P., Wang, Q., 2021. Applications of resilience engineering processes for national infrastructure resilience. J. Homel. Secur. Emerg. Manag. 16,
principles in different fields with a focus on industrial systems: a literature re­ 1–17. https://doi.org/10.1515/jhsem-2017-0019
view. J. Loss Prev. Process Ind. 69, 104366. https://doi.org/10.1016/j.jlp.2020. Wang, Q., 2010. Theoretical and Experimental Evaluation of Chemical Reactivity. PhD
104366 Dissertation, Texas A&M University, College Station, Texas, USA. 〈https://www.
Perrings, C., 2006. Resilience and sustainable development. Environ. Dev. Econ. 11, proquest.com/openview/eefcf7719551c2553f174fe56ebfafdb/1?pq-origsite=
417–427. gscholar&cbl=18750〉.
Righi, A.W., Saurin, T.A., Wachs, P., 2015. A systematic literature review of resilience Westerterp, K.R., Molga, E.J., 2006. Safety and runaway prevention in batch and
engineering: Research areas and a research agenda proposal. Reliab. Eng. Syst. Saf. semibatch reactors - A review. Chem. Eng. Res. Des. 84, 543–552. https://doi.org/
141, 142–152. https://doi.org/10.1016/j.ress.2015.03.007 10.1205/cherd.05221
Rostamabadi, A., Jahangiri, M., Zarei, E., Kamalinia, M., Alimohammadlou, M., 2020. A Wiig, S., Aase, K., Billett, S., Canfield, C., Røise, O., Njå, O., Guise, V., Haraldseid-
novel Fuzzy Bayesian Network approach for safety analysis of process systems; an Driftland, C., Ree, E., Anderson, J.E., Macrae, C., Bourrier, M., Berg, S.H., Bergerød,
application of HFACS and SHIPP methodology. J. Clean. Prod. 244, 118761. https:// I.J., Schibevaag, L., Øyri, S.F., Sjøseth, S., O’Hara, J., Kattouw, C.E., Kalakou, F.T.,
doi.org/10.1016/j.jclepro.2019.118761 Bentsen, S.B., Manser, T., Jeppesen, E., RiH-team, 2020. Defining the boundaries
Salehi, V., Veitch, B., Musharraf, M., 2020. Measuring and improving adaptive capacity and operational concepts of resilience in the resilience in healthcare research
in resilient systems by means of an integrated DEA-Machine learning approach. program. BMC Health Serv. Res. 20, 330.
Appl. Ergon. 82, 102975. https://doi.org/10.1016/j.apergo.2019.102975 Woods, D.D., 2015. Four concepts for resilience and the implications for the future of
Sarker, P., Lester, H.D., 2019. Post-disaster recovery associations of power systems resilience engineering. Reliab. Eng. Syst. Saf. 141, 5–9. https://doi.org/10.1016/j.
dependent critical infrastructures. Infrastructures 4, 1–16. https://doi.org/10. ress.2015.03.018
3390/infrastructures4020030 Yarveisy, R., Gao, C., Khan, F., 2020. A simple yet robust resilience assessment metrics.
Sharma, N., Tabandeh, A., Gardoni, P., 2018. Resilience analysis: a mathematical for­ Reliab. Eng. Syst. Saf. 197, 106810. https://doi.org/10.1016/j.ress.2020.106810
mulation to model resilience of engineering systems. Sustain. Resilient Yazdi, M., Kabir, S., 2017. A fuzzy Bayesian network approach for risk analysis in
Infrastruct. 3, 49–67. https://doi.org/10.1080/23789689.2017.1345257 process industries. Process Saf. Environ. Prot. 111, 507–519. https://doi.org/10.
Smit, B., Wandel, J., 2006. Adaptation, adaptive capacity and vulnerability. Glob. 1016/j.psep.2017.08.015
Environ. Chang. 16, 282–292. https://doi.org/10.1016/j.gloenvcha.2006.03.008 Yodo, N., Wang, P., 2018. A control-guided failure restoration framework for the design
Stoessel, F., 2009. Planning protection measures against runaway reactions using of resilient engineering systems. Reliab. Eng. Syst. Saf. 178, 179–190. https://doi.
criticality classes. Process Saf. Environ. Prot. 87, 105–112. https://doi.org/10.1016/j. org/10.1016/j.ress.2018.05.018
psep.2008.08.003 Zarei, E., Ramavandi, B., Darabi, A.H., Omidvar, M., 2021. A framework for resilience
Thomas, J.E., Eisenberg, D.A., Seager, T.P., Fisher, E., 2019. A resilience engineering assessment in process systems using a fuzzy hybrid MCDM model. J. Loss Prev.
approach to integrating human and socio-technical system capacities and Process Ind. 69, 104375. https://doi.org/10.1016/j.jlp.2020.104375

93

You might also like