Renewable and Sustainable Energy Reviews 151 (2021) 111574

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

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

A dynamic risk assessment model to assess the impact of the coronavirus

(COVID-19) on the sustainability of the biomass supply chain: A case study
of a U.S. biofuel industry
Zaman Sajid
Department of Business Administration, University of the People, 225 S. Lake Ave., Pasadena, CA, 91101, USA
Department of Process Engineering, Faculty of Engineering & Applied Science, Memorial University of Newfoundland, St. John’s, N.L., A1B 3X5, Canada

A R T I C L E I N F O A B S T R A C T

Keywords: The novel coronavirus (COVID-19) is highly detrimental, and its death distribution peculiarity has severely
Risk modeling affected people’s health and the operations of businesses. COVID-19 has wholly undermined the global economy,
Risk assessment in the biomass supply chain including inflicting significant damage to the ever-emerging biomass supply chain; its sustainability is dis-
Bayesian network
integrating due to the coronavirus. The biomass supply chain must be sustainable and robust enough to adapt to
Biomass
Supply chain
the evolving and fluctuating risks of the market due to the coronavirus or any potential future pandemics.
COVID-19 However, no such study has been performed so far. To address this issue, investigating how COVID-19 influences
Pandemic a biomass supply chain is vital. This paper presents a dynamic risk assessment methodological framework to
Coronavirus model biomass supply chain risks due to COVID-19. Using a dynamic Bayesian network (DBN) formalism, the
Virus impacts of COVID-19 on the performance of biomass supply chain risks have been studied. The proposed model
Dynamic Bayesian Network has been applied to the biomass supply chain of a U.S.-based Mahoney Environmental® company in Washington,
Biofuel USA. The case study results show that it would take one year to recover from the maximum damage to the
Model Validation
biomass supply chain due to COVID-19, while full recovery would require five years. Results indicate that
Supply chain risk
biomass feedstock gate availability (FGA) is 2%, due to pandemic and lockdown conditions. Due to the avail-
Sustainability
Feedstock gate availability ability of vaccination and gradual business reopenings, this availability increases to 92% in the second year.
Results also indicate that the price of fossil-based fuel will gradually increase after one year of the pandemic;
however, the market prices of fossil-based fuel will not revert to pre-coronavirus conditions even after nine years.
K-fold cross-validation is used to validate the DBN. Results of validation indicate a model accuracy of 95%. It is
concluded that the pandemic has caused risks to the sustainability of biomass feedstock, and the current study
can help develop risk mitigation strategies.

1. Introduction
A family of viruses that causes illness such as severe acute respiratory
syndrome (SARS), the common cold, and Middle East respiratory syn-
drome (MERS) is named a coronavirus [1]. In December 2019, a new
coronavirus originated in Wuhan, China, known as acute respiratory
syndrome coronavirus 2 (SARS-CoV-2). The disease this virus causes is
called coronavirus disease 2019, or COVID-19, commonly known as the
coronavirus. Being infectious, the spread of COVID-19 was quite rapid;
as of March 11, 2020, there were 118,000 cases in 114 countries around
the globe, and 4291 people were dead. Considering this life-threatening
situation, on March 11, 2020, the World Health Organization (WHO)
declared the outbreak of COVID-19 a ‘pandemic’ [2]. As of May 20,
2020, the COVID-19 has spread to 188 countries and has affected more
than 4.89 million people around the globe. This situation has resulted in
323,000 deaths, while nearly 1.68 million people have recovered [3].
According to the European Centre for Disease Prevention and Control,
the novel coronavirus originated from animals. It is transmitted from
human to human via tiny respiratory droplets through coughing,
sneezing, or when an affected person is nearby (at less than 1-m dis-
tance) to other persons [4]. These droplets can be inhaled, or people can
be infected when they touch a contaminated surface (which contains
such droplets) and then touch their nose, eyes, or mouth [5]. At its origin
in December 2019, no vaccine was available to immunize people and
protect them from the coronavirus. Moreover, there is no standard
treatment for patients with the coronavirus. However, the treatment of

E-mail addresses: zaman.sajid@mun.ca, zaman.sajid@uopeople.edu.

https://doi.org/10.1016/j.rser.2021.111574
Received 11 July 2020; Received in revised form 4 July 2021; Accepted 6 August 2021
Available online 14 August 2021
1364-0321/© 2021 Elsevier Ltd. All rights reserved.
Z. Sajid
the novel coronavirus is based on dealing with its symptoms (such as
fever, cough, and difficulty of breathing) and the clinical condition of a
patient [4]. Considering the facts of rapid human-to-human disease
transmission and with no available vaccine for COVID-19, the world
responded by closing public and private institutions, schools, colleges,
universities, local and international business ventures. Moreover, travel
restrictions have been put in place, and the enforced lockdown has been
in effect worldwide. To prevent the spread of the coronavirus and ensure
the safety of citizens, governments around the world have issued
stay-at-home orders in various countries [6]. This lockdown causes
significant global environmental, social and economic impacts. The
global environmental impact of COVID-19 includes the decrease in
emissions of pollutants and greenhouse gases. For example, the Centre
for Research on Energy and Clean Air reported at the end of March 2020
that travel bans in China had resulted in a 25% reduction in the emission
of carbon in the country over a four-week period, which amounts to 200
million metric tons of less carbon dioxide than in 2019 for the same
period [7]. Unlike the environmental impacts of the pandemic, social
and economic impacts are detrimental. The United Nations Department
of Economic and Social Affairs (UN DESA) identified emerging social
issues due to COVID-19 and have reported that the people living in
poverty are the most vulnerable due to the outbreak of COVID-19 and
the lockdown. For example, due to a lack of safe shelter, homeless
people have a high possibility of coronavirus exposure [8]. Moreover,
the absence of comprehensive universal social protection systems will
increase poverty and present fewer employment opportunities. The
extent of economic damage depends upon the speed of the spread of the
coronavirus. For example, at its initiation, the spread of this coronavirus
was rapid in China. This situation reduced automobile sales by 80%, and
exports fell 17.2% in January and February 2020. The Organization for
Economic Co-operation and Development (OECD) reported that due to
this epidemic, the global GDP growth would dwindle to 2.4% from 2.9%
in 2020 and can fall to zero in a worst-case scenario [9]. The supply
chain of a country is an integral part of its economy, and the re-
percussions of the COVID-19 outbreak have served to create and
accentuate external risks to supply chain sustainability. The Institute for
Supply Chain Management has reported that 75% of companies have
been crumbling due to transportation restrictions and are facing dis-
ruptions in their supply chain [10]. Due to the global coronavirus
pandemic, business closure and lockdown implications also have inev-
itable and uncertain impacts on the biomass supply chain [11]. Due to
the future need for sustainable energy, biomass conversion into useful
bioenergy can be used as an alternate energy source to fossil fuels.
However, the current pandemic is making this goal unviable. In a recent
survey by Hawkins Wright studying the impact of COVID-19 on global
biomass feedstock availability, 58% of wood pellet producers reported a
negative impact of the coronavirus on their business. In comparison,
33% reported a reduced production due to the pandemic [12]. The
participants who belonged to all regions of the world also indicated that
82% of them are negatively affected by the coronavirus, while the
businesses of 14% of them are severely affected. Considering such
adverse impacts of COVID-19 on the biomass supply chain, there are
many risks associated with the sustainability of biomass feedstock.
These include uncertainties in the biomass supply chain, biomass pro-
duction, logistics, biomass harvesting, transportation, labour availabil-
ity, and high preprocessing costs. In order to ensure the sustainability of
biorefineries and biofuels in this uncertain time, there is a need to
explore and understand the impact of COVID-19 on the biomass supply
chain. The risk assessment must primarily focus on integrating sustain-
able biomass resources, should consider the biomass symbiosis rela-
tionship, interdependencies of risks in the biomass supply chain
network, the sustainability of biomass resources, given fossil fuel prices,
and the lockdown of all workers in the industry due to COVID-19. To
address these challenges, this paper aims to (1) study the impact of
COVID-19 on the sustainability of the biomass supply chain, (2) develop
a risk assessment methodology for the biomass supply chain by
2
Renewable and Sustainable Energy Reviews 151 (2021) 111574
integrating external and internal risks in the supply chain network (3)
investigate the post coronavirus sustainability of the biomass supply
chain and (4) identify the timeline when the biomass supply chain could
attain pre-coronavirus conditions. Pre-analysis for the modelling in-
cludes identifying risk elements associated with the biomass supply
chain, key data sources from expert knowledge, and an in-depth litera-
ture review. A dynamic risk assessment model will develop a causal
network using identified risk factors and probabilistic data. The risk
interdependencies in this study will be constructed using causal effects,
i.e., the influence of one risk factor on another and throughout the
network. The model will be tested on the biomass supply chain of a U.
S.-based biofuel company, and model error will be calculated by
model validation.
This paper is organized into five major sections. The introduction
describes the impact of COVID-19 on the biomass supply chain and in-
cludes a brief discussion on the integrated biomass supply chain. The
second section is based on a literature review of the pertinent literature.
This section is further classified into two subsections: the first part
presents the current methodologies used to assess the risks to the
biomass supply chain. The second part highlights the literature of the
dynamic Bayesian network (DBN). The literature review is summarised
by identifying research gaps and the need for the current study. The
third section outlines the proposed methodology and validation and is
further divided into two subsections, a case study of a U.S. biofuel
company and the model’s applications to it. The fourth section is results
and discussion, presenting the results and reviewing the findings of this
research work. Then the research is concluded, along with suggested
future work directions.
1.1. Integrated biomass supply chain
Typically, a biomass supply chain consists of (i) a supply chain
network including biomass handling, preprocessing and the movement
of material from biomass origin to biorefineries, (ii) production of bio-
fuels in biorefineries, and (iii) product distribution to the consumers
[13]. Due to these business functions, the current biomass supply chain
management is a conundrum that plays a crucial role in managing
bioenergy and bioproduction processes [14]. Various risks are associ-
ated with the biomass supply chain that distinguish it from other
products’ traditional supply chains. Such risks are uncertainty in
biomass availability, limited biomass supply due to seasons and
weather, fluctuations in physical and chemical compositions of biomass,
biomass transport density, geographical distributions of feedstock, local
transportation systems, and distribution infrastructures [15]. The
complexity is compounded when costs of raw feedstock and bio-
refineries operations are compared with the market price of fossil fuels.
As highlighted by Ref. [16], the biomass supply chain is an integration of
four system components: (i) harvesting/collection of biomass from sin-
gle or multiple locations and the pre-treatment, (ii) intermediate storage
of biomass, (iii) transportation, and (iv) bioconversion. In addition to
these discrete processes, the final product’s (biofuel) distribution also
needs to be embedded in the biomass supply chain, as shown in Fig. 1. In
Fig. 1, all components are interconnected and interdependent, and
therefore no distinct boundary can be drawn between two pieces. In this
supply chain, biomass supply is the first and foremost point for biofuel
production.
Biomass is defined as any renewable organic material which is
available regularly, such as crops and their wastes, woods, and waste of
wood processing, algae, aquatic plants, animal manure, and wood and
its residues. Other biomass materials are waste from food processing,
waste cooking oil, food waste, and waste materials which produce fuels,
energy, and chemicals [17]. Plant-based organic materials that are not
meant to be used as food, such as palm kernel shells, rice husks, fruits’
wastes, and paddy straw, are also categorized as biomass feedstock [15].
Moreover, non-edible resources such as jatropha, Linum usitatissimum
(linseed), Azadirachta indica (neem), etc., and waste cooking oil are also
Z. Sajid
Fig. 1. An integrated biomass supply chain.
used as biomass feedstock [18]. Studies have identified that various
factors, such as inappropriate crop harvesting, unfavourable weather,
operational constraints, and the geographical region where the biomass
is cultivated can affect biomass availability [19]. Biomass is considered
to remain biologically living matter and chemically active throughout
the biomass supply chain. For example, for algae biomass, the active
biological ingredients are proteins, polyphenols, polysaccharides (bio-
polymers of glycosidic bonds and monosaccharides), polyunsaturated
fatty acids (PUFAs), minerals, and dyes [20]. The biological activity in
biomass can alter the composition of the biomass, which could affect the
biomass’s conversion to biofuel in the biorefinery. Using waste cooking
oil as a source of biomass feedstock, the water contents and free fatty
acids are dominant factors. Hence the quality of biomass is a crucial risk
factor in maintaining the efficiency of the biomass supply chain. In this
regard, appropriate harvesting and collection schedules can help ensure
the quality of biomass is within the desired limitations, up to the point
when it is delivered to the doorstep of the biorefinery. As shown in
Fig. 1, pre-treatment activities are performed on the biomass collection
site before sending the biomass to a biofuel processing facility.
Pre-treatment of biomass includes chemical, physical, biological, and
physicochemical processes, as highlighted in the literature [21,22].
Depending upon the characteristics of raw material and end products,
the pre-treatment or preprocessing increases biomass porosity, enhances
the efficiency of the enzyme, improves the accessibility of the enzyme,
and minimizes the losses of carbohydrates [23]. In preprocessing, the
dense biomass feedstock is converted to be less dense, which reduces
associated biomass costs such as transportation, handling, and storage,
and increases the product quality. The preprocessed biomass is trans-
ported to biorefineries, where it is converted into bioenergy (biofuel and
electricity) through the bioconversion process. There are three leading
technologies for the bioconversion process: thermochemical, biochem-
ical, and physicochemical [24,25]. Thermochemical conversion of
biomass is mainly performed using three pathways, i.e. pyrolysis, gasi-
fication, and combustion [26].
In contrast, biochemical conversion involves two primary process
3
Renewable and Sustainable Energy Reviews 151 (2021) 111574
alternatives; anaerobic digestion and fermentation [27]. The physico-
chemical transformation has various process options, such as ammonia
fibre explosion, steam explosion, and wet explosion [28]. This conver-
sion is out of the scope of the current study. The scope of the present
work is sketched in Fig. 2 and includes biomass supply collection to the
gate of a biorefinery. The final product, shown in Fig. 1, could be of two
categories: biofuels or biochemicals. The primary products of a bio-
refinery include biofuels, such as biodiesel (produced from animal fats
and vegetable oils), biogas (produced from decomposing food scraps,
animal manure, yard waste in landfills) or ethanol (also called bio-
ethanol, produced from crops of sugar cane and corn). The biochemicals
are derived from value-added products such as styrene and phenolics
used for various purposes [29]. Transportation connects all components
in a biomass supply chain, and hence transportation cost plays an
essential role in an economical and sustainable supply chain network. As
highlighted by Ref. [15], transportation scheduling and carriage routes
are key risk factors in biomass logistics. A viable schedule helps to
ensure that both feedstock and finished products are delivered on time
to the respective biorefinery and consumers. Selecting an appropriate
transport route helps to minimize travel time and cost; it also reduces the
environmental impacts of activities in the biomass supply chain.
2. Literature review
The literature review of this paper consists of two subsections. The
first section highlights the current state-of-the-art methodologies for risk
assessment of the biomass supply chain. The second part of the literature
review discusses the paradigm of the dynamic Bayesian network (DBN).
In the past, various researchers have proposed techniques to model
biomass supply chains. In a study, a mixed-integer linear programming
(MILP) model was presented to assess the size and geographic location of
a bioethanol plant in Austria [30]. The model studied various stages of
biofuel production, including biomass supply chain, heat utilization in
biofuel production, and demand for fossil and biofuels. A mathematical
model was presented in another study to model biomass supply chain
Z. Sajid
Fig. 2. Scope of the study.
and logistics in a biofuel production facility [31]. The proposed model
could identify the location, number and size of biorefineries based on
available biomass. The study applied the proposed model to the biomass
supply chain in the State of Mississippi. Another study presented a
mixed-integer linear programming (MILP) model to design the supply
chain risk of biomass-based ethanol [32]. The model used a case study of
an Italian biomass-based ethanol system to analyze the ethanol supply
chain’s strategic design and biomass planning decisions. Their model
was able to capture the multiple biomass technology choices and
considered the risk mitigation preferences of decision-makers. Another
work proposed a multistage stochastic programming model to deal with
uncertainties in quality (operational factor), the availability of biomass
feedstock, and their impact on net profit [33]. The study showed a
trade-off between biomass quality and profit and helped control varia-
tions in storage levels. In a study, a Bayesian network (BN) approach was
used to predict the complex behaviour of risk propagation in the supply
chain [34]. They studied the resilience index and risk exposure indices
for various nodes within the supply chain network. Their results showed
the vulnerable nodes due to disruption propagation in the supply chain
network and were helpful in supply chain risk management. A simula-
tion modelling approach was used to evaluate the risks of operational
disruptions on the cost of the biomass supply chain, production level,
and inventory [35]. The study considered three scenarios, and the
simulation model was analyzed for seven years. The study showed that
risk vindication alone at a facility is not enough. The concurrent inclu-
sion of failure dependencies at the system level is required for the sus-
tainable design of the biomass supply chain.
In a recent study, researchers studied operational and disruption
risks to the bioethanol supply chain [36]. They proposed a MILP based
model to identify, analyze the design, and risk planning of a supply chain
network of multi-feedstock lignocellulosic bioethanol. Using a case
study in Iran, their model provided optimal supply chain decision var-
iables. In another study, investigators evaluated reduction opportunities
of supply, market, and operational risks for the supply chain of a
cellulosic biorefinery, which adopted a distributed depot concept [37].
Their study was based on geographically distributed depots and showed
that operational and market risks were reduced for a distributed depot
supply system.
2.1. Dynamic Bayesian Network (DBN)
A Bayesian Network (BN) is a directed acyclic graph (DAG) in which
nodes represent random variables, and the directed arcs show the con-
ditional dependency of one node (called parent) on another (called
child) [38]. The influence of parent on child corresponds to a directed
edge. The strength of parent and child nodes is defined by assigning
marginal and conditional probability tables. A BN characterizes a
factorization of joint probability (JP) distribution over two (bivariate
4
Renewable and Sustainable Energy Reviews 151 (2021) 111574
distribution) or more than two (multivariate distribution) random var-
iables. Despite such characteristics, BN models cannot consider tempo-
ral information [39], which means a BN cannot model numerous
phenomena over time. This limitation is overcome by the Dynamic
Bayesian Network (DBN). A DBN helps to model and quantify temporal
relationships among different data points [40]. It represents the ‘evo-
lution’ of given random numbers as a function of the time steps sequence
(a discrete sequence). Intuitively, a DBN is the ‘evolution’ of risk over
time. The terminology of ‘dynamic’ in DBN does not mean the change in
network architecture but the change in model parameters [40]. A ‘static’
BN is denoted by nodes (N) and edges (E) that connect the nodes in a
directed position. For each node (denoted by y), a conditional proba-
bility distribution (CPD) is associated. The network structure, consisting
of directed relationships between parents and child nodes, interpreted
with their CPDs, defines a BN. The factorized joint probability (FJP) on
N (denoted by P(N)) is represented as:
∏ y
P(N) = P( )
y∈N
πy
where πy indicates the parents of child node y. This definition of
‘static’ BN is transformed to a ‘dynamic’ mode by considering two as-
pects: first, the initial or prior distribution of random variables in the
static BN at time 0. Secondly, the dynamic two-step BN is investigated;
this describes the transformation from the previous time t-1 to the cur-
rent time t, having the probability P( yt | yt−1 ) for any node y, which
belongs to N in a DAG. Hence, the JP for two sets of nodes, namely Nt
and Nt−1 , is:
∏
Nt yt
P( )= P( )
Nt−1 y∈N, π y∈N
πyt
The computation of the JP law is performed by ascertaining the
sequence in the network over a series of times. For the total length of the
path Z and JP of the initial network P(N0 ), the probability of trans-
forming from N0 to Z is computed as:
∏ y0 ∏Z ∏
yt
P(N0:Z ) = P(N0 ).P(N1:Z ) P( )⋅ P( )
y∈N
π (x 0 ) t=1 y∈N
π (y t)
Thus, DBN follows the Markov property, which states that the CPD at
a given time t is solely dependent upon the state at the previous time,
denoted as t-1 [41]. A DBN generalizes the Hidden Markov Model
(HMM). A more complex DBN could have multiple input variables
connecting last time slices, as shown in Fig. 3. A DBN defines the de-
pendencies between parent and child nodes over time. As shown in
Fig. 3, nodes (y1 to y5) in a DBN are still linked through a DAG;
nevertheless, cycles between nodes are allowed in the DBN. A DBN is a
series of BNs developed for varying time units; each BN is called a time
slice. As shown in Fig. 3, temporal links connect time slices, and the
Z. Sajid
Fig. 3. (a) An ordinary Bayesian network (B.N.) having five nodes, (b) Dynamic Bayesian network showing three slices.
outcome is a dynamic model. In the biomass supply chain, this property
can help to study the dependence of current and previous years’ varia-
bles—for example, the dependence of this year’s biomass supply on last
year’s rain. In DBN, all times are zero-based, which means t = 0, rather
than t = 1, which indicates the first time step.
Previously, the DBN has been used for modelling various systems
such as predictions of crop growth [42], process safety [43], system
performance measurements [44] and healthcare [45]. Although the
DBN has been used in various fields of study, it has never been used to
address the dynamic behaviour of the biomass supply chain system. To
fill this research gap, in this paper, a DBN for the biomass supply chain is
proposed, and the application is shown using a case study. Fig. 2 in-
dicates two categories of biomass supply chain risks: internal opera-
tional risks and internal supply chain risks [46]. Internal operational
risks refer to risks to the regular operations, such as risks to the distri-
bution, manufacturing, process improvements, product development,
and infrastructure. The internal supply chain risks indicate the potential
risks to the upstream and downstream supply chain of the biomass, such
as unpredictable demand by consumers or interruptions in the avail-
ability of biomass feedstock or bioproducts. As evident through the
literature review, the traditional risk assessment techniques have only
focused on internal operational and internal supply chain risks to the
biomass network. However, the biomass supply chain risks are evolving,
especially with the current pandemic conditions due to COVID-19. The
impacts of external risks on the biomass supply chain and the in-
teractions of external risks with supply chain components have never
been addressed. This work aims to fill this research gap. This paper has
three primary contributions and innovations; (1) prior research on
biomass supply chains has been focused on studying the static behaviour
of the biomass supply chain and does not consider time variability in the
supply chain system, and hence cannot provide biomass supply chain
behaviour over time. This research investigates such dynamic behav-
iour. This study provides a dynamic framework to analyze the evolution
of biomass supply chain risks over time. (2) This paper is the first to
assess the probabilistic impact of COVID-19 on a biomass supply chain
and identifies the recovery time of loss due to the coronavirus. (3) This
study provides a risk network model which captures the uncertainties of
biomass supply chains. It aims to answer the questions of the future
impact of COVID-19 on the biomass supply chain. How much will the
cost of the biomass supply chain be affected due to this coronavirus?
How much is the sustainability of the biomass supply chain affected due
to the ongoing pandemic? When will the biomass supply chain return to
5
Renewable and Sustainable Energy Reviews 151 (2021) 111574
the precoronavirus situation? What will be the impact of fuel prices on
biomass availability? The results from this research work can help
governments, investors, biomass supply chain managers and biomass
business owners to make informed decisions as they assess the
post-coronavirus situation. This study can help them to evaluate,
manage, minimize, and mitigate the risks to their businesses due to
COVID-19. The dynamic risk modelling framework for the biomass
supply chain risk is applied to a biofuel company located in Washington
state of the United States of America (USA). The case study helps to
illustrate the proposed methodology, results and findings.
3. Methodology
The research methodology in this study consists of five steps. The
prime objective of this work is to investigate the temporal and quanti-
tative strengths of the relationships among various risk factors of the
biomass supply chain. The definition of strength is reported in the
literature [47] and is applied here. The input data in this study were the
interdependency relationships among risk factors, which were defined
using experts’ elicitation. In this study, experts were a research group
consisting of three senior university professors in a U.S. university who
have an in-depth knowledge of the biomass supply chain and extensive
experience in chemical engineering research and development. The
steps followed include:
Step 1: Identification of risk categories
In this step, experts identified six risk categories associated with the
biomass supply chain. Those were qualitative factors, quantitative fac-
tors, labour, biomass transportation, oil, and the pandemic. The quali-
tative factors were defined as those risk factors related to the quality or
characteristics of biomass, such as the chemical and biological compo-
sition of the biomass. Risk factors in quantitative conditions were
defined as those risk factors that could affect the quantity or amount of
biomass available, such as the closure of biomass production sites. The
labour category consisted of those risk factors related to the workforce
employed in the biomass supply chain, such as labour availability. The
biomass transportation category was defined as risk factors associated
with moving biomass from one point to another—for example, the
transportation schedule. The oil category was described as a category
that consists of risk factors related to fossil-based oil, either diesel or
gasoline, used in the transportation of the biomass in the biomass supply
Z. Sajid
chain network: for example, oil demand. The pandemic risk category
was defined as those risk factors based on the epidemic situation, such as
a countrywide lockdown.
Step 2: Identification of risk factors
In this step, risk categories defined in step 1 were further classified
into risk factors using previously available studies [37,47–52]. The
identification of risk factors is based on the risk category under study
and the type of raw material used to produce biofuel—for example,
waste cooking oil or crops.
Step 3: Dynamic Bayesian Network (DBN) modelling
In this step, a DBN model was developed. This task was accomplished
in two phases, (a) structural learning and (b) parameter learning.
Structural learning considered the interdependencies among the sto-
chastic risk variables defined in step 2 and resulted in the graphical
structure of a dynamic BN. Parameter learning was used to define prior
marginal probability distributions of root nodes in the network and
conditional probability distributions of non-rooted nodes in the BN. A
detailed description of the two steps is provided below.
Step 3a. Structural learning
This step consisted of building the interdependency relationships
among risk factors through experts’ knowledge. The expert elicitation
technique was used to define such relationships, as discussed in the
literature [53]. This structured expert elicitation technique has been
extensively used in the literature, such as in Refs. [54–56]. Experts
defined the interdependencies among stochastic risk factors based on
the causal relationships between stochastic nodes. Once the relation-
ships were identified, the directed arcs were drawn between nodes as per
causality defined by the experts. The result of this step was a graph
structure of B.N.
Step 3b. Parameter learning
In this step, prior probability distributions were defined for root
nodes at time t = 0 (first-time step), and conditional probability tables
(CPTs) were defined for non-root nodes at time t = 0. Since COVID-19 is
new and there are no historical data available, experts’ judgments were
used in this step. The outcome of this step was a B.N. with temporal
relationships, i.e., a DBN. The analysis was performed for ten time steps
(t = 9). GeNIe Modeler by Bays Fussion, LLC (https://www.bayesfusion.
com/genie/) was used as a tool to build DBN and perform Bayesian
inference, as discussed in the next step.
Step 4. Risk-based inference
In this step, the risk-based inference was performed using the DBN
developed in step 3. The temporal feature of GeNIe allowed updating
beliefs of all time steps at once rather than observing results one at a
time. This step provided the posterior probabilities of each node in DBN,
which were conditionally based on the evidence. The computational
mechanism of solving DBN has been presented earlier. The probabilistic
results were reported and discussed.
Step 5: Model validation
As a final step of this methodology, the model presented was vali-
dated using the k-fold cross-validation method. This method helped to
provide information on the robustness and accuracy of the model in
predicting the outcomes. The use and reliability of this method to pre-
dict model accuracy has been extensively studied in the literature, for
example [57–60]. A database file (presented in Data-In-Brief) consisting
6
Renewable and Sustainable Energy Reviews 151 (2021) 111574
of 100 cases (n = 100) was compiled, and each case had values of the
response variable (biomass feedstock gate availability) and covariates.
The database file consisted of 100 samples with known values of input
nodes (as presented in the model) and the node of ‘biomass feedstock
gate availability’ (final output node). The randomized data file was
divided into ten equal sets or segments (k-value in k-fold validation).
The value of 10 sets is the optimal value of k in k-fold cross-validation, as
supported in the literature [60]. The first ten parts of data, representing
ten sets (n = 100/k = 10 = 10), were set aside, and the model was
parameterized with the leftover cases, i.e., (n-n/k). Then the model was
tested against other sets of the database for accuracy and comparison of
model results to known outcomes in each case. This procedure was
repeated for all ten segments. GeNIe was used as a tool to develop
randomized subsets and to perform k-fold cross-validation. Results were
presented and discussed in terms of the accuracy of nodes and selected
receiver operating characteristic (ROC) curves.
3.1. Case study of a U.S. Biofuel company
To demonstrate the applications of the proposed methodology, a case
study of a U.S. biofuel company in Washington state is presented. At the
time of doing this research on May 10, 2020, there have been 1.307
million confirmed cases of COVID-19 in the U.S., while 78,794 people
there have lost their lives due to COVID-19 [61]. The number of
confirmed cases in different states is shown in the choropleth map
sketched in Fig. 4. On March 13, 2020, the U.S. government declared a
national emergency to deal with the COVID-19 crisis [62]. As of April 7,
2020, several states ordered the closure of businesses, workplaces, and
schools and enforced social distancing [63], followed by a partial or
statewide lockdown all over the U.S., as shown in the choropleth map of
Fig. 5. This situation means that the U.S. economy was shut down, and
employees and employers could not work or receive customers. Road,
air, and sea transportation were also limited. This economic shutdown
also closed biomass business units across the country. In the U.S.,
various biomass feedstocks are being used, such as crops (sugar beets
and sugarcane), crop residues (bagasse, barley straw, corn stover, rice
straw, grain sorghum stubble, and wheat straw), wood (forest residues,
primary mill residues, urban wood, secondary mill residues), and waste
cooking oil [18]. For this research work, Mahoney Environmental®
company, located in Washington state, was chosen to elaborate on the
methodological framework. The company collects waste cooking oils,
which are used to produce biodiesel, and has been a leader in recycling
waste cooking oil for over 65 years in the U.S. Regularly, the company
collects used cooking oils from more than thousands of shopping malls,
food restaurants, bars, commercial food producers, stadiums and mili-
tary stations across Washington state and produces low-cost commer-
cial-grade biodiesel along with co-products such as solvents, lubricants
and glycerine at their plant site, shown in Fig. 5. Residents of Seattle and
nearby have also dropped off their used cooking oil at various drop-off
locations for waste cooking oil, shown in Fig. 6, and details of all 13
sites are presented in Table 1. Due to COVID-19 and the lockdown, the
collection of used cooking oil, its transportation to the chemical plant
site, and biodiesel distribution have been negatively affected. This study
aims to analyze the probabilistic magnitude of this impact and assess the
risk-based performance of the biomass supply chain using the proposed
model. Risk in this study is defined as the probability of having an
insufficient supply of waste cooking oil from all 13 drop-off locations to
the gate of Mahoney Environmental® for the pre-treatment of waste
cooking oil and subsequently the shortfall of biomass for biodiesel
production. The objective of the current case study is to identify and
analyze the risks to waste cooking oil feedstocks which would help
ensure a continuous supply to the Mahoney Environmental recycles
under investigation. Steps presented in the methodology section are
applied to this case study, and results are presented and discussed in the
next section.
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574

Fig. 4. Confirmed cases of COVID-19 in the USA* as of May 10, 2020. Data adapted from Johns Hopkins CSSE [61] *Notes: Names of some states are omitted to avoid
label clustering. Map developed in Datawrapper.

Fig. 5. Choropleth map of the United States* with statewide business closures due to COVID-19 as of April 3, 2020 [64]. The red circle shows the biofuel plant
location on the map. *Notes: Names of some states are omitted to avoid label clustering. Map developed in Datawrapper.

7
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574

Fig. 6. Public cooking oil drop-off locations for waste cooking oil in Seattle and nearby vicinities. Details are presented in Table 1. For an interactive map, readers
can click here. (Map courtesy of Google Maps).

4. Results and discussions

Table 1
Details of Waste Cooking oil drop-off sites.
The results of six risk categories and their associated subcategories as
Tag in Name Locations defined in steps 1 and 2 of the methodology are presented in Table 2. As
Fig. 6
shown in Table 2, each risk category is further classified into sub-
A Bainbridge - Transfer Station 7215 Vincent Road categories of risk factors. In total, 23 subcategories are defined. As
Bainbridge, WA 98110 shown in Table 2, the risk classification is based on risks associated with
a
B Federal Way - French Lake Dog Park 31531 1st Ave. S. Federal
Way, WA 98003
the biomass supply chain of the Mahoney Environmental® company.
Mahoney Environmental Restaurant 6333 1st Ave. S. Seattle, WA Table 2 also shows the definitions of risk terms identified in this study.
Cooking Oil Recycling 98108 The results of static BN are shown in Fig. 7, which shows the in-
D SODO - Republic Services 54 S. Dawson St. Seattle, WA terdependencies among risk factors. It indicates both external and in-
98134
ternal risks. External risks are the pandemic situation and lockdown,
E Madison Park - Bert’s Red Apple 1801 41st Ave. E. Seattle, WA
98112 while the internal risks are shown in the green area in Fig. 7. Fig. 7 is a
F Mercer Island - Presbyterian Church 3605 84th Ave. SW Mercer static BN model of the problem and includes the associated risk factors
Island, WA 98040 and their interdependencies. Arcs in Fig. 7 indicate the probabilistic
a
G Olympia - Thurston County 2420 Hogum Bay Rd. N.E. relationships among variables and are quantified by the conditional
HazoHouse Olympia, WA 98516
Ha Redmond - City of Redmond Drop 8703 160th Ave, NE,
probability distributions, as presented and discussed in the attached
Offsite Redmond, WA 98052 Data-in-Brief. Each directed arc is represented in red colour while the
I Renton - Republic Services 501 Monster Rd. SW Renton, direction of the arrow indicates the direction of causality. Fig. 7 shows
WA 98134 that the pandemic causes a lockdown situation that influences both
Ja City of Auburn 1020H Street S.E., Auburn,
businesses and regular transportation and transit services within
WA 98002
K Kent - Republic Services 22010 76th Ave S, Kent, WA Washington.
98032 Due to the business closures across Washington state, there is an
La City of Sedro-Woolley Recycling 315 Sterling St., Sedro- impact on the availability of the labour to perform routine day-to-day
Facility Woolley, WA 98284 operations in the biomass supply chain. Fig. 7 also reveals that the de-
a
M Whatcom County Disposal 3505 Airport Dr, Bellingham,
WA 98226
mand for fossil-based fuel and its production affects its market prices.
The results of the BN indicate that the closures of restaurants and
a
Not shown in Fig. 6 due to space limitations. collection sites to collect waste cooking oil have an impact on the
quantity of biomass available. This outcome is true since restaurants and

8
Z. Sajid
Table 2
Risk categories and risk factors.
Risk Category Risk factors Definitions
Qualitative factors Quality of biomass This risk factor indicates the
(QL) (QL1) quality of waste cooking oil
collected.
Cooking temperature The temperature at which fresh
(QL2) cooking oil is cooked and
transformed into waste cooking
oil.
Changes in waste This factor refers to the chemical
cooking oil composition of waste cooking oil,
composition (QL3) e.g., changes in free fatty acids,
and moisture contents.
Usual collection This risk factor shows the usual
schedule (QL4) schedule to pick up the waste
cooking oil.
Preprocessing cost This risk element indicates the
(QL5) cost to process waste cooking oil
to make it useable before using it
to produce biodiesel e.g. filtration,
sedimentation, and moisture
removal from waste cooking oil.
Quantitative Unavailability of fresh This factor shows the
factors (QT) vegetable oil (QT1) unavailability of fresh vegetable
oil in the market due to market
closures and transportation halts
at a large scale.
Less recycling by the It represents the less use of fresh
consumers (QT2) cooking oils by consumers due to
market closure and subsequently
less recycling of used cooking oil.
Closure of collection It indicates the closure of
sites (QT3) collection sites depicted in
Table 1.
Restaurants not open It shows the closures of restaurant
(QT4) businesses due to COVID-19.
Quantity of biomass Indicates the amount of biomass
available (QT5) available as feedstock to produce
biodiesel.
Labour (L) Labour availability (L1) It shows the availability of the
labour working in biomass supply
chain networks such as managers,
engineers, production staff,
collection teams, drivers, and
other staff.
Operational cost (L2) It is the expenses incurred to run
business operations such as
administrative costs, maintenance
costs and salaries of the
employees.
Biomass Robust transport It refers to the transport schedule
transportation schedule (BT1) to collect waste cooking oil.
(BT) Viable transport route It shows the routes adapted to
(BT2) collect waste cooking oil and
transfer it to the biodiesel
production site.
Sustainable transport Transport planning refers to
planning (BT3) policies to move waste cooking oil
from its source to the factory gate.
Transportation cost It is cost associated with
(BT4) transporting waste cooking oil to
bio-conversion site (Mahoney
Environmental® in the current
case study).
Oil (OL) Fuel demand (OL1) It indicates the demand for fossil-
based fuel for transportation such
as diesel and gasoline.
Fuel prices (OL2) This risk factor refers to the
market price of fossil-based fuel.
Production of fuel It denotes the production of fossil-
(OL3) based fuel.
Pandemic (P) Lockdown (P1) It indicates the condition in which
people are required to stay at
home, and no free movement is
allowed.
9
Renewable and Sustainable Energy Reviews 151 (2021) 111574
Table 2 (continued )
Risk Category Risk factors Definitions
Business operational It refers to the working of a
(P2) business.
Transport (Regular It shows the regular public and
transportation) (P3) private transport by roads, air,
and water.
Effectiveness of vaccine This factor refers to the efficacy of
(V) the COVID-19 vaccine.
Feedstock gate – It indicates the availability of
availability waste cooking oil at the factory
(FGA) gate.
collection sites are the major places to collect waste cooking oil for the
Mahoney Environmental® company. Their closure due to COVID-19
causes an impact on the quantity of biomass collected. Results in
Fig. 7 also show that due to less labour available to collect waste cooking
oil from the suppliers, there is an impact on the collection schedule.
Results also reveal that a high cooking temperature causes a change in
the chemical composition of waste cooking oil, affecting the quality of
biomass, which influences its preprocessing cost. One of the prominent
features of the model is the impact of the vaccine effectiveness on the
network. As shown in Fig. 7, vaccination affected the pandemic. This
result is based on the fact that immunization induces the reduction of the
population’s vulnerability to the risk of COVID-19.
Results in Fig. 7 show that three risk factors are affecting the gate
availability of biomass feedstock, the transportation cost, the quantity of
biomass available, and operational costs. The probabilistic results of the
BN analysis provide more depth to understand the impact of COVID-19
on the biomass supply chain. Due to space limitations, the results of
selected risk factors are discussed here. The results of P(Low QL1/QL3,
QL4) show a probability value of 0.851, which indicates that, based on
changes in the composition of waste cooking oil and due to the un-
availability of the usual collection schedule, there is a high probability
that the biomass collected will be of low quality, due to the pandemic.
The results of preprocessing cost validate such findings, since P(High
QL5/Low QL1) = 0.846, indicating a higher probability of preprocessing
cost based on the low quality of this biomass. The high preprocessing
cost influences the operational cost, and its probability is given as P
(High L2/High QL5) = 0.821. This result shows that there is a strong
possibility to have an increased operational cost of the biomass supply
chain, based on the high preprocessing cost of the biomass.
In terms of the available quantity of biomass, the results indicate that
P(Less QT5/QT2, QT3, QT4) = 0.93. This result shows that the COVID-
19 has a significant impact on the quantity of biomass available, given
that restaurant and waste oil collection sites are closed. There is less
recycling of waste cooking oil. The results of BN in Fig. 7 also help to
understand the impact of COVID-19 on fuel prices. The probabilistic
analysis shows that P(Low OL2/Low OL1, High OL3) = 0.891. This
result indicates that due to COVID-19, there is less demand for fossil-
based fuel, and due to its high production, it suffers from lower prices.
This situation indicates an abundance of fuel in the market. The DBN
analysis shows the evolution of these results over time, and a screenshot
of DBN simulation in GeNIe software is shown in Fig. 8.
Fig. 7 shows how the risk factors influence one another in non-
temporal behaviour, while Fig. 8 enables us to examine the temporal
behaviour of these risk factors over time. Fig. 8 also allows us to study
the influence of COVID-19 on the performance of the supply chain
network, as it indicates the DBN of the biomass supply chain. The model
presented in Fig. 8 helps to understand the impact of COVID-19 vacci-
nation on the biomass supply chain network. The use of dynamic BN
permits us to perform quantitative modelling of various risk factors that
influence the gate availability of biomass feedstock. The DBN presented
in Fig. 8 is an extension of static BN, shown in Fig. 7. DBN supports the
temporal evaluation of risk factors over a discretized timeframe. As
shown in Fig. 8, the timeline is divided into ten time slices, and the DBN
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574

Fig. 7. A Bayesian Network (BN) model for the biomass supply chain. The area in green represents the internal risks.

Fig. 8. A Dynamic Bayesian Network (DBN) model for biomass supply chain.

10
Z. Sajid
aids each node to be conditionally dependent on its parents. At the same
time, DBN also allows a node to be conditionally dependent on its par-
ents, at previous time slices.
The latter behaviour is shown in Fig. 8 as a number within the small
square. For example, in Fig. 8, number 2 in the small green square on
low fuel demand indicates that there are two time slices in temporal
dependencies of low fuel demand. While the arc between transport and
low fuel demand shows number 4 in a small square, this arc shows the
temporal influence, while its number indicates its order, also called lag.
Lag in DBN shows the order of the temporal (time-based) link. The
number 4 suggests that the influence of transport spans over four time
steps. In other words, a temporal link with order four from the transport
node to another node, low fuel demand, as shown in Fig. 8, is interpreted
as transport links to low fuel demand in the later future, i.e., four-time
intervals or steps. The low fuel demand node has a link from the
transport node in the past, i.e., four-time intervals or steps in the past.
Results in Fig. 8 show that the transportation cost is influenced by low
fuel prices, sustainable transport planning, high preprocessing cost, and
less quantity of biomass available.
In terms of the quality of biomass, changes in biomass composition,
and the absence of the usual collection schedule, cause low quality
biomass. Results in Fig. 8 also show that a decrease in labour availability
results in high operational costs. The graphical structure of the DBN
model, shown in Fig. 8, is similar to its static BN model, shown in Fig. 7,
except that there are additional arcs in the DBN model that quantify the
temporal relationships among risk factors. The dynamic or temporal arcs
represent the changes in the probabilistic values of risk variables over
time. Fig. 9 captures two unrolled time steps (Step 0 and 1) of the DBN
model presented in Fig. 8. Unrolling helps to covert DBN into its
equivalent static BN, which quickly helps to understand the DBN
structure. As shown in Fig. 9, all variables are repeated in each time step
except the pandemic, effectiveness of vaccine and lockdown. Such
behaviour of the model is because these three variables do not repeat in
each time step but influence the rest of the variables in each time step.
Considering the observed dynamic evidence of QT5, BT4, and L2, the
DBN model encodes the probability distribution over gate availability of
biomass feedstock. This simulation results in calculating the risk of
having no biomass feedstock available at the factory gate, given the
conditions of low transportation cost, high operational cost, and less
quantity of biomass available at a given time, i.e., P(FGA t/E1), where E
= QT5t-1 = less, BT4t-1 = low, L2t-1 = High. It shows that conditional
probability distribution for gate availability of biomass feedstock de-
pends on the results in the previous time step for risk factors of the
available quantity of biomass, transportation cost, and operational cost.
The results are drawn in Fig. 10, which shows the probability of having
no biomass feedstock at the factory gate, given this dynamic evidence, i.
e., P (FGA (no)/E). The plot in Fig. 10 shows the risk of having no
biomass feedstock availability at the company’s gate over ten time steps.
This result can help develop production planning and the optimal time
for biodiesel production in a pandemic situation.
Results in Fig. 10 show that in the pre-coronavirus situation, gate
availability of biomass feedstock was regular, and there was no external
risk of COVID-19. Fig. 10 shows that when the pandemic hits Wash-
ington state, the probability of unavailability of biomass supply at the
company’s gate reaches its maximum value of 0.98 in the lockdown
situation. The high unavailability of biomass indicates the need to secure
waste cooking oil from more remote locations than shown in Table 1.
This action may not be environmentally friendly, as more vehicle
transportation would release more carbon by fuel burning, and nega-
tively affect the environment. Moreover, collecting waste cooking oil
from farther places would not be economical, as it would involve the
additional costs of the operations of the biomass supply chain.
Results also indicate that it would take at least one year for
1
Evidence.
11
Renewable and Sustainable Energy Reviews 151 (2021) 111574
probability to reduce to 0.084. As shown in Fig. 10, this value should
approach zero, as in the pre-coronavirus situation; however, such results
are not obtained until the fifth year. In other words, it will take five years
for the biomass supply chain to complete recovery from the COVID-19
hit. It is noteworthy here that the U.S. House of Representatives, on
May 15, 2020, voted to pass the COVID-19 stimulus package. The bill,
known as the ‘Health and Economic Recovery Omnibus Emergency
Solutions Act or the HEROES Act’ aims to provide relief to biofuel
producers in the U.S. The financial benefits include providing a 45 cents
per gallon payment to qualified biofuel producers from January 1, 2020,
to May 1, 2020 [65]. For interested readers, the entire bill is available at
the House of Representatives, accessible at [66]. With these financial
incentives, the time of five years of recovery, as revealed in this study,
may be shortened.
The DBN model provided lays the imperative groundwork for
probabilistic approaches to assess the disruption to the biomass supply
chain due to fuel prices. Supply chain networks of biomass feedstock
may be complex systems. They may comprise many interrelated and
interconnected factors that may introduce nonlinearity into the system.
Depending upon the complexity of the network, discretization can be
introduced into the model, which can help capture and compute non-
linear relationships among risk variables [67]. Moreover, discretiza-
tion would help to reduce computational power to solve the model [68].
The results of the impact of COVID-19 on fuel prices are drawn in
Fig. 11. Fig. 11 shows that the probability of the fuel price being lower is
higher at the start of the coronavirus spread, shown in the blue colour
line. In other words, the likelihood of fuel prices being higher is low at
the beginning of the coronavirus spread (shown in the orange colour
line). Values of both curves in Fig. 11 are complementary to each other.
The DBN predicts that the probability of the fuel price being higher will
gradually start to increase and will take one year to reach a value of
0.84; after this time, there is a gradual increase in fuel prices over time.
However, the fuel price does not reach a pre-coronavirus situation, even
after nine years, as shown in Fig. 11. The increasing trend in fuel prices
in Fig. 11 is compared with fuel price trends from the U.S. Energy In-
formation Administration (EIA). Their data of weekly U.S. regular
convention retail prices of gasoline are drawn in Fig. 12. The data of the
U.S. EIA show that after May 4, 2020, there is an increasing trend in fuel
price. Hence, this confirms the results of this study. The methodological
framework presented in this study is consistent with [69], in which re-
searchers used DBN to model risk assessment of flood control operations.
As the DBN model shows the changes in probabilities of the various
risk factors of the biomass supply chain over time, it can help develop
contingency plans to manage risks to the biomass supply chain. To
demonstrate the operationalization of the model, this study chose a
biofuel production facility that utilizes waste cooking oil as a raw ma-
terial. However, the model presented can be applied to supply chain
networks of other biofuel raw materials such as sugar beet, corn, sug-
arcane, jatropha, grass, switchgrass, woody crops (lignocellulosic
biomass), woodchips, agricultural residues, soybean oil, rapeseed oil, or
other plant oils. In any case, the risk categories of quality should include
risk factors such as biological activities and a proper harvesting
schedule. The risk category of quantity should consist of unseasonable
weather, rainfall, and droughts. However, the risk factor of trans-
portation should be based on vehicles to collect the biomass from fields
and its transportation to bioconversion sites. The methodological
framework presented is consistent with [69], in which, using experts’
opinions, researchers developed a DBN model to perform a risk assess-
ment of flood control operations. Future work is recommended to
expand the analysis by including the risk factors associated with
bioconversion and biofuel delivery to customers. Since this will intro-
duce structural complexity in the network, a dynamic object-oriented
Bayesian network approach [71,72] is recommended.
Z. Sajid
Fig. 11. Impact of COVID-19 on fuel price.
Fig. 12. Prices of regular gasoline in the U.S. (November 4, 2019, to June 15,
2020), data adapted from [70].
Fig. 13. Accuracy of model for each time step.
random guess [73]. This fact indicates that a value above 0.5 and closer
to 1 indicates a better judgment, as indicated in Fig. 14. An increase in
AUROC shows high values of true positives, which suggests that model
results are accurate.
The dotted diagonal line in Fig. 14 shows a baseline ROC curve of a
hypothetical classifier4 which is worthless and has random
4
Each point on a ROC curve is called a classifier.
13
Renewable and Sustainable Energy Reviews 151 (2021) 111574
classification. Any classifier above the diagonal line is classified as
better, while below the diagonal line is worse. It is worth mentioning
here that a classifier below a random classification line may have useful
information, but the information is being applied incorrectly. Hence, for
a model to be validated, the classifier should fall above the diagonal line,
and the AUROC should be closer to 1. Being expressed in one number,
AUROC shows the quality of the model. Results shown in Fig. 14 (for
time steps 2 and 6) indicate that the model offers better performance for
both time steps and that results are reliable. The model data has a
slightly better quality for time step 2 than time step 6 since the AUROC
value for time step 2 (0.7062) is marginally better than the AUROC
value for time step 6 (0.6839).
4.2. Practical implications of this study
This study’s results can help develop a sustainable and robust
biomass supply chain to mitigate the effects of COVID-19. The study can
help decision-makers to better plan and cost optimize biofuel production
in post-coronavirus conditions. As suggested by this study, one of the
significant risk factors in the biomass supply chain is the high trans-
portation cost to move biomass from its origin to the bioprocessing fa-
cility. This study finds challenges in selecting an appropriate biorefinery
where the presented model could be applied. For this purpose, Mahoney
Environmental® is chosen in this study. This company has been a leader
in recycling used cooking oils in the U.S. for over 65 years. Furthermore,
selecting appropriate bioconversion technology to convert biomass
feedstock into biofuel is crucial for developing a sustainable and resilient
biomass supply chain. The selection can be made based on the techno-
economic studies of the bioconversion process.
Future perspectives of this study are that it provides a risk assessment
framework for decision-makers to understand the risk to the biomass
supply chain in the post-COVID era when the biofuel industries will
emerge as more crucial and critical than ever before. This work finds its
applications in developing robust risk management strategies that will
help minimize the risk of COVID-19 to the biomass supply chain and
ensure the sustainability of the bioeconomy. Moreover, such a design
will redefine the role of emerging biofuel industries in the global
economy. The results of this research are essential to developing con-
tingency plans to combat the vulnerabilities to the biomass supply chain
due to the coronavirus or similar future pandemics. Due to mobility
restrictions, the pandemic has caused various socio-economic crises.
Such crises have reshaped the investment in renewable and sustainable
energy resources, and significant disruption and delays have been wit-
nessed, which have created uncertainty in the production and sustain-
ability of biofuel for years to come. The economic crisis may include
fewer employment opportunities in the biofuel sector, closures or
limited biomass processing sites, cheaper availability of conventional
fossil fuel than biofuel, and lower or limited household incomes due to
the pandemic. Socio-crises of the pandemic associated with the biomass
supply chain may include discrimination in assigning shift work, espe-
cially for hourly waged workers, job layoffs without considering
seniority, and an increase in inequality. This research can help deal with
socio-economic challenges by providing better resource allocation and
sustainable resource management during the pandemic and post-
coronavirus situations. The model-building process in this research
also faced challenges of identifying risk categories for this study. This
challenge is overcome by using expert elicitation and relevant literature.
For future work, it is recommended to develop U.S. biofuel policies in
light of this study, considering the pandemic challenges. It is also rec-
ommended to look at how COVID-19 variants can affect the presented
model– a recent variant has been detected in March 2021 named
B.1.617.2 (Delta), initially originating in India in December 2020 [74].
5. Conclusions and future work
COVID-19 has hit the whole world and has affected various
Z. Sajid
Fig. 14. ROC curves for Biomass FGA = no for time steps 2 and 6, representing AUROCs values of 0.7062 and 0.6839, respectively.
industries, including the biomass supply chain. The COVID-19 outbreak
has significant repercussions for the sustainability of the biofuel in-
dustry. As COVID-19 is new, adequate risk management strategies are
essential to ensure the sustainability of the biomass supply chain.
Although the biofuel industry is nascent in this crucial time, the per-
formance of the biomass supply chain has become more critical. In this
study, a risk-based dynamic Bayesian network (DBN) model is presented
to assess the biomass supply chain’s risk over ten years. Biomass supply
chain risk is defined as the unavailability of biomass feedstock to pro-
duce biofuel. Dealing with such risk is vital since an uninterpreted
continuous supply of the biomass supply chain will guarantee the
optimal production of biofuel. The presented model is applied to a U.S.
biofuel company. The results indicate that the pandemic has affected the
supply chain of biomass, and there are 85.01% chances that the biomass
collected in a pandemic will be of low quality. With the development of
vaccinations and partial reopening of businesses, this chance falls to
54.23%. Results also indicate that low-quality biomass increases the
preprocessing cost of biomass feedstock to 84.60% during the pandemic.
Study shows that due to lockdown and lower demand of fossil-based oil,
fuel prices would drop to 89% of the regular prices. The dynamic nature
of the model indicates a gradual increase in prices and demand for fossil
fuels. Results show that the fuel prices will not reach a pre-coronavirus
situation, even after nine years of businesses’ reopenings. Biomass FGA
is low at the start of the pandemic, with the unavailability of 98%.
However, vaccination availability helps to drop this unavailability to
7.9% in year two and then there is a gradual decrease in its value.
The presented model is validated using k-fold cross-validation. The
ROC curves drawn indicate that the data used in this study are not
random guesses and have an accuracy of 95% for all time steps.
Measuring risk to the biomass supply chain in this study helps assess the
biorefinery’s “riskiness” concerning the risk of fuel market prices. Since
COVID-19 is new and the biofuel industries have never seen its impact in
the past, the risk data aresparse, and therefore expert elicitation was
used in this study. The results of this study can help biofuel investors,
managers, and biofuel policy-makers to make well-informed decisions
on the sustainability of the biomass supply chain in the post-coronavirus
situation. In conclusion, COVID-19 is causing the performance of the
biomass supply chain to plummet and is presenting challenges to the
sustainability of the biomass supply chain and biorefineries.
14
Renewable and Sustainable Energy Reviews 151 (2021) 111574
Nevertheless, coordination at a global level can help to curb COVID-19
impacts and bolster the world’s bioeconomies.
Credit author statement
Zaman Sajid: Conceptualization; Data curation; Formal analysis;
Investigation; Methodology; Project administration; Resources; Soft-
ware; Validation; Visualization; Roles/Writing - original draft; Writing –
review & editing.
Declaration of competing interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
Acknowledgments
The author would like to thank three anonymous professors at the
University of Pittsburgh, USA, who provided valuable comments to
improve the contents of this work. The author would also like to thank
anonymous reviewers and the editorial team for their comments and
suggestions to improve this work.
References
[1] Mayo C. Coronavirus disease 2019 (COVID-19) - symptoms and causes - mayo
clinic. https://www.mayoclinic.org/diseases-conditions/coronavirus/symptoms-c
auses/syc-20479963. [Accessed 20 May 2020].
[2] WHO. WHO Director-General’s opening remarks at the media briefing on COVID-
19 - 11 March 2020. World Heal Organ; 2020. https://www.who.int/dg/speech
es/detail/who-director-general-s-opening-remarks-at-the-media-briefing-on-covid-
19—11-march-2020. [Accessed 20 May 2020].
[3] CSSE. Coronavirus COVID-19 (2019-nCoV). Johns Hopkins Univ; 2020.
https://gisanddata.maps.arcgis.com/apps/opsdashboard/index.html#/bda759474
0fd40299423467b48e9ecf6. [Accessed 20 May 2020].
[4] ECDC Q, COVID-19 A on. Eur Cent Dis Prev Control 2020. https://www.ecdc.
europa.eu/en/covid-19/questions-answers. [Accessed 20 May 2020].
[5] WHOHT. Q&A on coronaviruses (COVID-19) - how does COVID-19 spread? Heal
Top. 2020. https://www.who.int/emergencies/diseases/novel-coronavirus-2019/
question-and-answers-hub/q-a-detail/q-a-coronaviruses. [Accessed 21 May 2020].
Z. Sajid
[6] Haddad M. Coronavirus: how much more time are people spending at home? |
News | Al Jazeera. 2020. https://www.aljazeera.com/news/2020/04/corona
virus-world-staying-home-200406122943899.html. [Accessed 22 May 2020].
[7] Myllyvirta L. Analysis: coronavirus temporarily reduced China’s CO2 emissions by
a quarter. Carbon Br 2020. https://www.carbonbrief.org/analysis-coronavirus-h
as-temporarily-reduced-chinas-co2-emissions-by-a-quarter. [Accessed 22 May
2020].
[8] UN DESA. The social impact of COVID-19 | DISD. United Nation Dep Econ Soc Aff;
2020. https://www.un.org/development/desa/dspd/2020/04/social-impact-of-co
vid-19/. [Accessed 22 May 2020].
[9] Segal S, Gerstel D. The global economic impacts of COVID-19 | center for strategic
and international studies. Cent Strateg Int Stud 2020. https://www.csis.org/analys
is/global-economic-impacts-covid-19. [Accessed 22 May 2020].
[10] Boyd J. COVID-19 Survey: impacts on global supply chain. USA: Institute for
Supply Management ® study reveals virus’ supply chain effects. Tempe; 2020.
[11] Forestry B. Covid-19 Biomass supply chain response - Usewoodfuel. Usewoodfuel
Scotl 2020. https://usewoodfuel.co.uk/covid-19-biomass-supply-chain-response/.
[Accessed 22 May 2020].
[12] Wright H. Survey: COVID-19 impacts raw material availability for pellets |
Biomassmagazine.com. Biomass Mag; 2020. http://biomassmagazine.com/articles
/17056/survey-covid-19-impacts-raw-material-availability-for-pellets. [Accessed
23 May 2020].
[13] Bairamzadeh S, Pishvaee MS, Saidi-Mehrabad M. Multiobjective robust
possibilistic programming approach to sustainable bioethanol supply chain design
under multiple uncertainties. Ind Eng Chem Res 2016;55:237–56. https://doi.org/
10.1021/acs.iecr.5b02875.
[14] Gold S, Seuring S. Supply chain and logistics issues of bio-energy production.
J Clean Prod 2011;19:32–42.
[15] Hong BH, How BS, Lam HL. Overview of sustainable biomass supply chain: from
concept to modelling. Clean Technol Environ Policy 2016;18:2173–94.
[16] Iakovou E, Karagiannidis A, Vlachos D, Toka A, Malamakis A. Waste biomass-to-
energy supply chain management: a critical synthesis. Waste Manag 2010;30:
1860–70. https://doi.org/10.1016/j.wasman.2010.02.030.
[17] Sundaram S. Biorefineries and chemical processes: design, integration and
sustainability analysis. Green Process Synth 2015;4:65–6.
[18] Sajid Z, Zhang Y, Khan F. Process design and probabilistic economic risk analysis of
bio-diesel production. Sustain Prod Consum 2016;5:1–15. https://doi.org/
10.1016/j.spc.2015.10.003.
[19] Lozano-Moreno JA, Maréchal F. Biomass logistics and environmental impact
modelling for sugar-ethanol production. J Clean Prod 2019;210:317–24. https://
doi.org/10.1016/j.jclepro.2018.10.310.
[20] Konkol D, Górniak W, Świniarska M, Korczyński M. Algae biomass in animal
production. Algae biomass charact. Appl. Springer; 2018. p. 123–30.
[21] Chiaramonti D, Prussi M, Ferrero S, Oriani L, Ottonello P, Torre P, et al. Review of
pretreatment processes for lignocellulosic ethanol production, and development of
an innovative method. Biomass Bioenergy 2012;46:25–35.
[22] Mood SH, Golfeshan AH, Tabatabaei M, Jouzani GS, Najafi GH, Gholami M, et al.
Lignocellulosic biomass to bioethanol, a comprehensive review with a focus on
pretreatment. Renew Sustain Energy Rev 2013;27:77–93.
[23] Zhang K, Pei Z, Wang D. Organic solvent pretreatment of lignocellulosic biomass
for biofuels and biochemicals: a review. Bioresour Technol 2016;199:21–33.
https://doi.org/10.1016/j.biortech.2015.08.102.
[24] Adams P, Bridgwater T, Lea-Langton A, Ross A, Watson I. Chapter 8 - biomass
conversion technologies. In: Thornley P, adams PBT-GGB of BS. Academic Press;
2018. p. 107–39. https://doi.org/10.1016/B978-0-08-101036-5.00008-2.
[25] Sajid Z, Khan F, Zhang Y. Process simulation and life cycle analysis of biodiesel
production. Renew Energy 2016;85. https://doi.org/10.1016/j.
renene.2015.07.046.
[26] Sadhwani N, Liu Z, Eden MR, Adhikari S. Simulation, analysis, and assessment of
CO2 enhanced biomass gasification. In: Kraslawski A, Turunen I, 23rd, editors. Eur.
Symp. Comput. Aided process eng., vol. 32. Elsevier; 2013. p. 421–6. https://doi.
org/10.1016/B978-0-444-63234-0.50071-3.
[27] Gautam P, Kumar S, Lokhandwala S. Chapter 11 - energy-aware intelligence in
megacities. In: Kumar S, Kumar R, Pandey A, editors. Curr. Dev. Biotechnol.
Bioeng. Elsevier; 2019. p. 211–38. https://doi.org/10.1016/B978-0-444-64083-
3.00011-7.
[28] Papadokonstantakis S, Johnsson F. Biomass conversion technologies - D3.1 Report
on definition of parameters for defining biomass conversion technologies. 2017.
Gothenburg.
[29] Pandey A, Mohan SV, Chang J-S, Hallenbeck PC, Larroche C. Biomass, biofuels,
biochemicals. Biohydrogen. Elsevier; 2019.
[30] Leduc S, Schwab D, Dotzauer E, Schmid E, Obersteiner M. Optimal location of
wood gasification plants for methanol production with heat recovery. Int J Energy
Res 2008;32:1080–91.
[31] Ekşioğlu SD, Acharya A, Leightley LE, Arora S. Analyzing the design and
management of biomass-to-biorefinery supply chain. Comput Ind Eng 2009;57:
1342–52.
[32] Giarola S, Bezzo F, Shah N. A risk management approach to the economic and
environmental strategic design of ethanol supply chains. Biomass Bioenergy 2013;
58:31–51. https://doi.org/10.1016/j.biombioe.2013.08.005.
[33] Shabani N, Sowlati T. A hybrid multi-stage stochastic programming-robust
optimization model for maximizing the supply chain of a forest-based biomass
power plant considering uncertainties. J Clean Prod 2016;112:3285–93. https://
doi.org/10.1016/j.jclepro.2015.09.034.
[34] Ojha R, Ghadge A, Tiwari MK, Bititci US. Bayesian network modelling for supply
chain risk propagation. Int J Prod Res 2018;56:5795–819.
15
Renewable and Sustainable Energy Reviews 151 (2021) 111574
[35] Sharma B, Clark R, Hilliard MR, Webb EG. Simulation modeling for reliable
biomass supply chain design under operational disruptions. Front Energy Res 2018;
6:100.
[36] Mousavi Ahranjani P, Ghaderi SF, Azadeh A, Babazadeh R. Robust design of a
sustainable and resilient bioethanol supply chain under operational and disruption
risks. Clean Technol Environ Policy 2020;22:119–51. https://doi.org/10.1007/
s10098-019-01773-2.
[37] Mamun S, Hansen JK, Roni MS. Supply, operational, and market risk reduction
opportunities: managing risk at a cellulosic biorefinery. Renew Sustain Energy Rev
2020;121:109677. https://doi.org/10.1016/j.rser.2019.109677.
[38] Nielsen TD, Jensen FV. Bayesian networks and decision graphs. Springer Science &
Business Media; 2009.
[39] Liang H, Ganeshbabu U, Thorne T. A dynamic Bayesian network approach for
analysing topic-sentiment evolution. IEEE Access 2020;8:54164–74.
[40] Murphy KP. Dynamic bayesian networks. Probabilistic Graph Model M Jordan
2002;7:431.
[41] Benhamou E, Atif J, Laraki R. A new approach to learning in Dynamic Bayesian
Networks (DBNs). 2018. ArXiv Prepr ArXiv181209027.
[42] Kocian A, Massa D, Cannazzaro S, Incrocci L, Di Lonardo S, Milazzo P, et al.
Dynamic Bayesian network for crop growth prediction in greenhouses. Comput
Electron Agric 2020;169:105167. https://doi.org/10.1016/j.
compag.2019.105167.
[43] Cai B, Liu Y, Liu Z, Chang Y, Jiang L. A multiphase dynamic bayesian network
methodology for the determination of safety integrity levels. Bayesian networks
reliab. Eng. Springer; 2020. p. 217–37.
[44] Ramírez PAP, Utne IB. Use of dynamic Bayesian networks for life extension
assessment of ageing systems. Reliab Eng Syst Saf 2015;133:119–36. https://doi.
org/10.1016/j.ress.2014.09.002.
[45] Li M, Liu Z, Li X, Liu Y. Dynamic risk assessment in healthcare based on Bayesian
approach. Reliab Eng Syst Saf 2019;189:327–34. https://doi.org/10.1016/j.
ress.2019.04.040.
[46] Christopher M, Peck H. Building the resilient supply chain. 2004.
[47] Sajid Z, Khan F, Zhang Y. Integration of interpretive structural modelling with
Bayesian network for biodiesel performance analysis. Renew Energy 2017;107.
https://doi.org/10.1016/j.renene.2017.01.058.
[48] Amundson J, Faulkner W, Sukumara S, Seay J, Badurdeen F. A bayesian network-
based approach for risk modeling to aid in development of sustainable biomass
supply chains. In: Bogle IDL, fairweather MBT-CACE. 22 Eur. Symp. Comput. Aided
process Eng., vol. 30. Elsevier; 2012. p. 152–6. https://doi.org/10.1016/B978-0-
444-59519-5.50031-9.
[49] Rauch P. Developing and evaluating strategies to overcome biomass supply risks.
Renew Energy 2017;103:561–9. https://doi.org/10.1016/j.renene.2016.11.060.
[50] Sajid Z, Khan F, Zhang Y. A novel process economics risk model applied to a
biodiesel production system. Renew Energy 2018;118:615–26. https://doi.org/
10.1016/j.renene.2017.11.022.
[51] Williams PRD, Inman D, Aden A, Heath GA. Environmental and sustainability
factors associated with next-generation biofuels in the US: what do we know?
Environ Sci Technol 2009;43:4763–75.
[52] Zhang Y, Hu Q, Xiao D, Liu X, Shen Y. Spatial-temporal change of agricultural
biomass and carbon capture capability in the mid-south of hebei province. In: Li R,
Monti A, editors. L. Alloc. Biomass crop. Challenges oppor. With chang. L. Use.
Cham: Springer International Publishing; 2018. p. 159–87. https://doi.org/
10.1007/978-3-319-74536-7_9.
[53] Ayyub BM. Elicitation of expert opinions for uncertainty and risks. CRC Press;
2001.
[54] Larkin P, Gracie R, Shafiei A, Dusseault M, Sarkarfarshi M, Aspinall W, et al.
Uncertainty in risk issues for carbon capture and geological storage: findings from
structured expert elicitation. Int J Risk Assess Manag 2019;22:429–63.
[55] Hemming V, Burgman MA, Hanea AM, McBride MF, Wintle BC. A practical guide
to structured expert elicitation using the IDEA protocol. Methods Ecol Evol 2018;9:
169–80.
[56] Schmidt O, Gambhir A, Staffell I, Hawkes A, Nelson J, Few S. Future cost and
performance of water electrolysis: an expert elicitation study. Int J Hydrogen
Energy 2017;42:30470–92.
[57] Arlot S, Celisse A. A survey of cross-validation procedures for model selection. Stat
Surv 2010;4:40–79.
[58] Govindarajan M. Text mining technique for data mining application. Proc world
Acad Sci Eng Technol 2007;26:544–9. Citeseer.
[59] Geroldinger A, Lusa L, Nold M, Heinze G. On resampling methods for model
assessment in penalized and unpenalized logistic regression. ArXiv Prepr
ArXiv210107640. 2021.
[60] Marcot BG, Hanea AM. What is an optimal value of k in k-fold cross-validation in
discrete Bayesian network analysis? Comput Stat 2020:1–23.
[61] COVID-19 United JH. States cases by county - Johns Hopkins coronavirus resource
center. Johns Hopkins Univ; 2020. https://coronavirus.jhu.edu/us-map. [Accessed
10 May 2020].
[62] Cathey L. Government coronavirus response: trump declares a national emergency,
says he “likely” will get tested - ABC News 2020 [)], https://abcnews.go.com/Poli
tics/government-coronavirus-response-live-updates-trump-declares-national/story
?id=69580277. [Accessed 10 May 2020].
[63] Train R. Coronavirus lockdown in the USA: which states and what measures have
been applied? AsCom. 2020. https://en.as.com/en/2020/04/07/other_sports/158
6293073_218312.html. [Accessed 11 May 2020].
[64] Schumaker E. Here are the states that have shut down nonessential businesses -
ABC News. ABC News; 2020. https://abcnews.go.com/Health/states-shut-essent
ial-businesses-map/story?id=69770806. [Accessed 11 May 2020].
 processes
Article
Computational Fluid Dynamics (CFD) Modeling and Analysis
of Hydrocarbon Vapor Cloud Explosions (VCEs) in Amuay
Refinery and Jaipur Plant Using FLACS
Zaman Sajid 1,2, * , Muhammad Kashif Khan 3,4 , Alireza Rahnama 2 , Farzan Sahari Moghaddam 2 , Kirti Vardhan 2
and Reema Kalani 2
!"#!$%&'(!
!"#$%&'
Citation: Sajid, Z.; Khan, M.K.;
Rahnama, A.; Moghaddam, F.S.;
Vardhan, K.; Kalani, R.
Computational Fluid Dynamics
(CFD) Modeling and Analysis of
Hydrocarbon Vapor Cloud
Explosions (VCEs) in Amuay
Refinery and Jaipur Plant Using
FLACS. Processes 2021, 9, 960.
https://doi.org/10.3390/pr9060960
Academic Editor: Chi-Min Shu
Received: 14 April 2021
Accepted: 26 May 2021
Published: 28 May 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Processes 2021, 9, 960. https://doi.org/10.3390/pr9060960
1 Department of Business Administration, University of the People, 225 S Lake Ave, Pasadena, CA 91101, USA
2 Department of Process Engineering, Faculty of Engineering & Applied Science, Memorial University of
Newfoundland, St. John’s, NL A1C 5S7, Canada; arahnamamah@mun.ca (A.R.);
fsaharimogh@mun.ca (F.S.M.); kvardha@mun.ca (K.V.); rakalan@mun.ca (R.K.)
3 School of Mechanical Engineering, Sungkyunkwan University, 2066 Seobu-Ro, Jangan-Gu Suwon,
Gyeong Gi-Do, Seoul 16419, Korea; kashif.khan167@gmail.com
4 School of Chemical Engineering, Sungkyunkwan University, 2066 Seobu-Ro, Jangan-Gu Suwon,
Gyeong Gi-Do, Seoul 16419, Korea
* Correspondence: zaman.sajid@mun.ca; Tel.: +1-(0)709-765-8844
Abstract: Process safety helps prevent the unexpected and unplanned release of flammable and toxic
chemicals, leading to poisonous gas clouds, fires, and explosions. Vapor cloud explosions (VCEs)
are among the most severe hazards to humans and the environment in process facilities. Therefore,
process safety demands to use best and reliable techniques to model VCEs in process industries
and storage tanks of flammable chemicals. In this regard, the Computational Fluid Dynamics (CFD)
models are more appropriate, as these models provide three-dimensional (3D) modeling of all
sequences of events in an accident. In this study, CFD is used to model VCE in two industrial
accidents: the Amuay refinery disaster (happened in 2012) and the Indian Oil Corporation’s (IOC)
Jaipur terminal (2009). This work studies 3D CFD modeling of flammable cloud explosion in the
real-time configuration for both accidents. FLACS (FLame ACceleration Simulator), a CFD software,
is used to simulate the loss of hydrocarbon containment, cloud formation, and explosion in both
industrial case studies. The ignition locations and grid sizes were varied to analyze their influence on
explosion overpressure, temperature, vapor velocity, and fuel mass. This work also investigated the
effect of geometry complexity on the explosion. Results showed that, as opposed to the coarse grid,
the fine grid provides more precision in the analysis. The study also reveals an explosion overpressure
of the order 4–15 bar (g) for the given case studies. This study’s results can help perform a qualitative
and quantitative risk assessment of the Amuay refinery accident and Jaipur fire. The simulation of
different scenarios can help develop and improve safety guidelines to mitigate similar accidents.
Keywords: process safety; Computational Fluid Dynamics (CFD); gas explosion; vapor cloud explo-
sion (VCE) simulation; Amuay refinery; Jaipur fire
1. Introduction
Fuels are a significant energy source for power generation and transportation. In
general, fuels such as petrol, natural gas, diesel, and petroleum gas have high demand.
The need for energy is pushing every country to develop refinement facilities for crude
oil processing and refining. The proliferation of processing industries has increased the
potential hazards to the safety of people and process facilities [1]. Today, process safety
intermingles in the design, production, and maintenance of every processing facility. How-
ever, despite all safety guidelines and measures, accidents are happening. The magnitude
of these accidents ranges from minor damage to equipment or personnel to catastrophic
https://www.mdpi.com/journal/processes
Processes 2021, 9, 960 3 of 18

The studies presented in the literature review section helped propose better process
design facilities with more robust production and maintenance plans. However, such
studies have not considered the variations in fuel type (light versus heavy fuel), within
the site geometry in explosion modeling. Fuel type plays an essential role in defining the
intensity of the explosion. Moreover, previous studies consider limited accident scenarios.
The site geometry of the accident and the location of an ignition source are essential
aspects that need to be considered in computational-based studies. For example, the heat
flux distributions of fire and the flame behavior are highly dependent upon the ignition
source’s location, as highlighted in the literature [14]. Hence, an ignition source site plays
a vital role in assessing the overpressure in a VCE. Therefore, it should be included in
the CFD modeling. The present work demonstrates the applications of CFD simulation
to simulate Amuay refinery and Jaipur fire accidents. This study aims to predict various
characteristics of VCE in both industrial accidents. To the best of the authors’ investigations,
there has been no study conducted on the comparative behavior of VCEs in both accidents,
and hence is the objective of current work. This work also discusses various technical
aspects of both accidents to understand gas explosions better and investigates factors
that affect the CFD model. In these two case studies, the variations in ignition locations,
temperature, pressure, vapor velocity, and the mass of combustible fuels are also studied.
This work also identifies proper grid size for higher precision and captures the effect of gas
composition and geometry complexity using different ignition locations in two accidents.
A methodological framework of this study is shown in Figure 1. As shown in Figure 1, as a
preliminary step of the study, different scenarios are developed in step 1 for both accidents
under investigation. In step 2, input information is retrieved. This information includes
the meteorological conditions and details on the release of hydrocarbon for both accidents.
Step 3 is aimed to define the domain and geometry for the simulation of both accidents.
Computational grid and initial conditions are described in steps 4 and 5, respectively, while
in the last step, simulations are performed, and results are analyzed. Further details of the
methodology are presented in Section 3 of this paper.

Figure 1. A glimpse of the methodological framework for this study.

Processes 2021, 9, 960 4 of 18

This paper consists of two parts. The first part provides a conceptual background
of accidents in the Amuay refinery and Jaipur fire, and the second part describes the
methodological framework of the study.

2. Background
2.1. Vapour Cloud Explosion (VCE)
In general, the combustion of a large amount of flammable vapor cloud involving
more than one tonne of combustible fuel causes an explosion. As a result of quick energy
release, the explosion causes to have a magnificent overpressure. Consequently, more
severe fatalities occur due to the minimal time to revert to protective action [15]. The
release of combustible materials in the atmosphere results in the formation of a vapor cloud.
If the propagating flammable elements find a source of ignition, it results in a VCE. The
VCE demands analytical, computational (with CFD tool), and experimental investigations
to mitigate future accidents. The most likely outcome of a flammable fluid discharge over
some particular time is to form a flammable vapor cloud. The failure of an under-pressure
vessel, pipe, or other facilities can lead to instant dispersion of flammable vapors. The
general event tree for cloud vapor explosion is presented in Figure 2.

Release of Instant Delayed Consequence

Explosion
hydrocarbon ignition ignition

Yes Fire

No Release to
atmosphere
No

No Fire
Yes

Yes VCE
Figure 2. An event tree of VCE.

2.2. Detonation and Deflagration

The overpressure produced due to the fuel mixture’s fast combustion spreads through
the fuel mixture as a shock wave that causes to ignite the fuel mixture. The shock wave
and flame front move together at supersonic velocity. Due to this characteristic, detonation
is considered more destructive as compared to deflagration. In deflagration, there is the
subsonic velocity (<100 m/s) of flame propagation and a modest overpressure (<7.3 psi).
However, in detonation, the flame propagation velocity is supersonic (up to 2000 m/s), and
there is high overpressure (290 psi). Under certain conditions, deflagration combustion is
quickly transformed into a detonation explosion. The phenomenon is called deflagration
to detonation transition (DDT). It has been revealed that the DDT phenomenon occurred
in the Amuay refinery accident and Jaipur fire [16]. There are strong reasons behind the
existence of DDT in these industrial VCEs. For example, there was a high flame speed
(closer to the speed of sound) in both accidents. Under such circumstances, there is a high
possibility that deflagration would have been transitioned to detonation [17]. To reduce the
likelihood of DDT, researchers have pointed to having an obstacle-free path and avoiding
congestion such as residential buildings in a fire radiation radius [10]. These research
findings are being modeled in this study using the geometry of the accidents.
Processes 2021, 9, 960 5 of 18

2.3. Affecting Parameters

Different parameters analyzed in this study are discussed below.

2.3.1. Geometry Complexity

One of the primary conditions in which the subsonic transforms into supersonic flame
speed is the geometry complexity (obstacles and partial confinement in the flame path).
The geometry of an accident scene can significantly affect the behavior of the explosion,
and the gas explosion concentration can vigorously differ according to the flame’s velocity.
Changes in boundary conditions can result in a remarkable change within the propagation
of gas. The obstacle in the path of the explosion propagation has a considerable impact
on increasing the gas velocity. The location of the obstruction profoundly influences the
propagation of the flame. Research shows that a faster propagation of the flame is noticed
when the obstacles are closer to the ignition source [18].

2.3.2. Ignition Source

According to a study, some of the significant sources of accidents in process industries
are listed below [19]:
• flames being exposed to the environment, such as flares;
• burning substances;
• solid surfaces have a high temperature;
• electrical sparks beside similar other means of discharge;
• confinement having an adiabatic situation.
The minimum ignition temperature of the fuel-air mixture is highly critical. Further-
more, the ratio for mixtures of hydrocarbon mixtures being investigated at their various
temperatures of the mixture has a crucial role in the explosion modeling [20].

2.3.3. Wind Effect

A study investigating the accidents due to the VCEs shows that the wind speed was
low in more than 70% of accidents [21]. A low rate of wind speed causes the leakage of fuel
by gravity. The results advise that nil or low wind speed conditions can affect the entire
explosion risk in many oil and gas plants. The weather conditions are occasionally rare
than low wind velocity. However, they are associated with gigantic clouds with a much
higher risk of ignition.
Moreover, wind conditions with low speed allow cloud formation through the low
release rates of vapor. This condition has the potential to cause accidents as there is a high
likelihood of continuous release. Hence, the modeling of the risk of VCE in petrochemical
industries needs to include wind conditions. In a review study on VCEs, researchers
categorized various industrial accidents based on wind conditions [21]. In their study,
both accidents (Jaipur and Amuay) were classified as those which occurred in nil/low
wind conditions. The study also identified vapor release rates of 67 kg/s and 3 kg/s for
Amuay and Jaipur accidents, respectively. The wind data in their study was based on
meteorological records and the shape of the cloud formed. A low wind speed was defined
as the wind speed that spreads the vapor cloud to an equal extent and in all fuel source
directions, and this definition is adopted in the current study.

2.4. The Explosion at Punto Fijo Refinery, Venezuela

An explosion occurred at Amuay (Punto Fijo) refinery, the second-largest refinery in
the world, on 25 August 2012, around 01:11 a.m. (GMT) in the northwest of Venezuela.
A plant layout of the Amuay refinery is shown in Figure 3. The shock wave released by
the explosion affected the neighboring residential area. There was significant damage to
people and property. More than 50 people lost their lives, over 100 people were seriously
injured, several light injuries were reported, and more than 1600 houses were damaged.
Residents of over 200 homes were evacuated to safer places. The damage in monetary
terms was more than one billion U.S. dollars.
Processes 2021, 9, 960 6 of 18

Figure 3. An Overview and layout of the Amuay Refinery located in Venezuela, Falcón State, Carirubana city. An
approximate location of the disaster zone is highlighted in a red box. Reproduced with permission from Kirti Bhushan
Mishra, Klaus-Dieter Wehrstedt, Holger Krebs, Fuel Processing Technology; published by Elsevier, 2014, with permission
from Elsevier.

Amuay incident, a layout shown in Figure 4, began with a leak of high vapor pressure
substances such as alkenes, liquefied natural gas (LNG), or liquid butane (or propane) and
formed a denser flammable vapor cloud in the atmosphere, as shown in Figure 4. The
dispersed cloud stuck near the ground and created fog under low wind conditions. A study
reported the spread of dispersed fuel to a radius of 0.6 km, due to low wind speed [10].
The accumulation of the dispersed vapor cloud proved to be catastrophic when the cloud
caught an ignition source. Researchers believed that the ignition source was most probably
not inside the Amuay refinery but passing by vehicle [22]. The overpressure generated,
as the result of the explosion, was of high magnitude. It crushed steel fences and cars,
blown the refinery structure, and bent the infrastructure and vehicles. Based on the media
sources, various initiating events led to make this accident a catastrophic one. Those events
were the negligence towards the regular maintenance and ignoring the leak and associated
smells in the plant before the incident [10].
An installed component’s functional failure is the most frequent hazard in the petro-
chemical industries, including valves, pumps, and piping. These operational failures occur
due to an improper maintenance schedule, corrosion, and inadequate safety knowledge
practices. Some witnesses in the refinery reported seeing a liquefied gaseous fuel leak-
age [22]. Other studies reported initiating events as malfunctioning of the valve, pumping
system, and piping connected to the olefin tanks [23]. In summary, it was believed that
the Amuay incident occurred due to the failure of a valve in piping. Venezuela’s socialist
government reported that the sabotage and manipulation of several bolts from a gas pump
caused the rupture, which initiated a gas leakage in the flange of a pump. Subsequently, a
quick alkene leak occurred due to a 7.4 cm opening in a pipeline [24].
Dispersion of high vapor pressure olefin fuel started to accumulate a combustible
cloud in the atmosphere after a valve’s failure, as shown in Figure 5. This cloud of olefin
stuck to the ground due to its higher density than air. It formed white irregular fog with no
particular smell unless it contained some additives like hydrogen sulfide. Due to slow wind
speed, the dispersed fuel started to accumulate and covered a radius of nearly 0.6 km [10].
A passing-by vehicle was recognized as the most likely source of ignition in the Amuay
Processes 2021, 9, 960 7 of 18

accident. The blast generated an overpressure in the range of 150 kPa to 2000 kPa, which
proved to be destructive for property and human lives [10].

Figure 4. A satellite map of vapor cloud formation in Amuay refinery in Venezuela, Falcón State, Carirubana city, was
reproduced with permission from Kirti Bhushan Mishra, Klaus-Dieter Wehrstedt, Holger Krebs, Fuel Processing Technology;
published by Elsevier, 2014, with permission from Elsevier.

Figure 5. (a) Tank fire, (b) wind pushing the fire to other tanks, (c) fire affecting olefin tanks, (d) site showing three tanks
subjected to fire after the accident, reproduced with permission from Kirti Bhushan Mishra, Klaus-Dieter Wehrstedt, Holger
Krebs. Fuel Processing Technology; published by Elsevier, 2014, with permission from Elsevier.
Processes 2021, 9, 960 8 of 18

2.5. Explosion at Indian Oil Corporation’s (IOC) Jaipur Terminal, India

The accident shown in Figure 6 occurred on 29 October 2009, in a giant tank at the
Indian Oil Corporation oil depot, in Jaipur India, that held 8000 kiloliters of petrol. The
accident’s primary reason was gasoline leakage from a valve connected to the pipeline [25].
The fire erupted in multiple tanks and continued for over a week [26]. The explosion created
a massive fireball that covered the entire installations and started a fire in the nearby fuel
tanks. The investigating team estimated that there was an uncontrolled leakage for 90 min.
Eighty-one tonnes of gasoline might have vaporized. The vapor cloud eventually exploded
with an intensity equivalent to 38 tonnes of TNT explosives. The management decided
to let the petroleum products burnt to avoid another potential accident due to fire [16].
The leaking gasoline created a vapor cloud that exploded and started a fire in the storage
tanks. This accident claimed twelve lives, and several people were injured and lost more
than Rs. 2.8 billion (Indian rupees). The investigating team found that the ignition source
was flameproof electrical equipment or firing up of an internal combustion engine in the
nearby areas. The fire in the storage tanks spread to various other tanks, setting all the
60,000 kiloliters of fuel on the fire. The fire caused catastrophic damage in a radius of nearly
2 km. It was observed that the vapor cloud engulfed the entire facility, including the plant,
truck offloading area, and parts of the pipeline terminal. Gasoline concentration in the area
was 7.8 percent to 1.2 percent, within the gasoline flammability range [13].

Figure 6. Satellite map of Jaipur plant indicating the location of leak origin and the top view of the tanks. The area on
the map is India, Rajasthan State, Jaipur city. Reproduced with permission from Kirti Bhushan Mishra, Journal of Loss
Prevention in the Process Industries; published by Elsevier, 2018, with permission from Elsevier.
Processes 2021, 9, 960 9 of 18

Below is the summary of the sequence of events [27];

1. loss of primary gasoline containment;
2. escape of gasoline from secondary containment;
3. the operating staff was ordained useless;
4. lack of tools to tackle the situation;
5. flaws in the engineering design;
6. no team available for operating from the site and in the case of essential areas.
The flammable cloud was formed because a massive quantity of gasoline escaped
from containment and then vaporized. Various parameters contributed to the explosion:
the explosive properties; weather conditions; and ignition source location. The energy
emitted in the explosion created a blast wave that changed the nature of the blast. In a blast
wave, the pressure increases instantaneously and drops very fast to return to atmospheric
pressure. In this case, the fundamental reason was the complete absence of the specific site
operating procedures and leakage mitigation equipment [28].
Computational Fluid Dynamics (CFD) techniques are being progressively applied
to better understand fire and explosion behaviors [29]. In CFD, mathematics is applied
through algorithms and numerical techniques to understand fluid flow problems with or
without chemical reactions [30]. Although analytical methods provide great usefulness
in studying various cloud explosions parameters, a CFD concept is broadly utilized for
multiple applications such as VCE [29].
CFD allows the modeling of a detailed site geometry of the VCE facility. Moreover,
CFD enables modification of wind conditions, as per case studies, in large-scale modeling
experiments [31]. However, there is no experimental data that could help study the
continuous release of flammable material in calm wind conditions in which vapor transport
is caused by gravity. A few studies have focused on using CFD to analyze VCE at the
Buncefield oil depot accident in low wind conditions [32,33]. The only quantitative data
used in these studies was vapor front progress; however, the movement of the vapor front
does not affect the cloud structure itself. Another challenge is to model CFD when there is
a long-range transport of cloud.
Moreover, the selection of grid size in CFD affects the precision of the results. To fill
this research gap, this paper models Amuay refinery and Jaipur fire accidents using CFD.
This work uses the actual fuel compositions at the time of accidents. Both accidents are
reconstructed in this simulation study, and analyses are performed.

3. Modeling Methodology
The CFD models are the most significant models to perform 3D (3 Dimensional)
modeling and allow using the equations that govern fluid motion for a vast range of
complex scenarios [12]. FLACS (FLame ACceleration Simulator), developed by GexCon
AS (https://www.gexcon.com/, accessed on 9 May 2021), is a frequently utilized CFD tool
in chemical and petroleum industries for explosion modeling in process safety and risk
analysis. The FLACS implements and solves the compressible conservation equations (such
as momentum, mass, species, and enthalpy) on a 3D Cartesian grid using finite volume [17].
Among CFD software for explosion modeling, FLACS is a proper research tool for inves-
tigating explosion details through various influencing factors in a gas explosion, such as
gas cloud size and ignition position [34]. Moreover, FLACS can also help analyze other
affecting parameters, including flow dispersion, shock waves, and the cloud’s dimensions.
Due to such features, FLACS v10.4 was used to model and simulate the Amuay refinery
accident and Jaipur fire in this study.

3.1. Input Data for Simulation

In this simulation study, 3D (3 Dimensional) models of the Amuay refinery and Jaipur
fire accidents were built. The momentum and mass equations for both accidents and
numerical models were adopted from the literature [35]. The general specification of the
geometry within different x, y, and z directions was 25.6 m, 8 m, and 8 m, respectively. Two
Processes 2021, 9, 960 10 of 18

fuel compositions were studied in the simulation. The first fuel was flammable octane,
i.e., the fuel composition closer to that of in Jaipur oil depot fire, compared to flammable
natural gas composition (i.e., CH4 91.3%; C2 H6 7%, C3 H8 1.3%) as the second fuel, i.e.,
the fuel composition closer to that of in Amuay refinery explosion. Initial and boundary
conditions were defined as 2 m/s for initial wind speed. The stability in the atmospheric
conditions was achieved by setting relative turbulence to zero. This condition helped to
achieve calm/zero wind conditions, as discussed earlier.
In both studies, simulations were performed with coarse and fine mesh in which the
grid numbers and the graphical presentation of the grids developed are shown in Table 1
and Figure 7, respectively.

Table 1. The number of grids in different dimensions for both coarse and fine meshing.

Mesh X Y Z
Coarse grid numbers 28 12 8
Fine grid numbers 112 48 32

Figure 7. Geometry with coarse grid (left) and fin grid (right).

Two different grid sizing, i.e., coarse and fine grids, were used for simulation to assess
the results’ precision based on the grid sizing parameter.
The explosions for fuel compositions in various aspects, such as the released pressure,
temperature, fluid velocity, and fuel level, were evaluated. Besides, the ignition location
was also assessed in the simulations since it was another profoundly influencing parameter
for both accidents, as discussed previously. The effect of geometry complexity on the
explosion was analyzed by considering two ignition source locations, i.e., case 1 (18 m,
5 m, 5 m) and case 2 (13 m, 5 m, 5 m). FLACS has already been validated as a valuable
tool for developing CFD code for various gas dispersion cases [36,37] and, therefore, such
validation is not repeated in this study.

4. Results and Discussion

The highest achieved pressures, using both fuels, during the explosion with coarse
and fine mesh calculations are summarized in Table 2.

Table 2. Results of simulation with coarse and fine meshing.

Highest Pressure for (bar (g))

Case Study
A Fine Mesh A Coarse Mesh
Case1 Octane 15.00 8.00–8.50
Case2 Octane 5.25–5.50 3.00–3.25
Case 1 Natural gas 7.50–8.00 5.50–6.00
Case 2 Natural gas 4.00 1.70–2.00
Processes 2021, 9, 960 11 of 18

Table 2 indicates that grid sizing is a critical parameter that significantly influences the
precision of the analysis. In Table 2, Case 1 and Case 2 represent the different locations of
the ignition source in modeling, as discussed in Section 3.1 of this paper. In Table 2, “Case 1
Octane” and “Case 2 Octane” both indicate the use of octane as fuel, while “Case 1 Natural
gas” and “Case 2 Natural gas” both show the use of natural gas as fuel in respective case
studies. Results suggest that higher-pressure values are observed for all four cases in
Table 2 when a grid with fine mesh size is used. This analysis of grid sizing shows that, as
opposed to coarse mesh, the fine mesh size results in higher accuracy. The variations of
pressure, fluid velocity, and fuel consumed in the explosion over time for both cases are
shown in Figure 8. In Figure 8, the peak of pressure after the explosion is presented with
other parameters such as fuel level and fluid velocity for octane explosion in both cases of
ignition locations by implementing fine grid calculations. The results show that in case 1,
the pressure starts to increase from 0.37 s and reaches a maximum pressure of 15 bar (g)
in 0.42 s. The peak pressure drops to 0 bar (g) at 0.49 s. However, case 2 indicates that
pressure starts to increase from 0.34 s and gains a peek at 5.24 bar (g) at 0.42 s. The pressure
drops to 0 bar (g) at 0.49 s. These results show that the ignition location case1 resulted in
a much higher pressure peak. This trend is due to the high-pressure explosion (i.e., over
14 bar (g)) in comparison with the lower pressure peak for case 2 ignition location (i.e.,
over 5 bar (g)). The natural gas explosion results are presented in Figure 9 for both ignition
sites with fine grid calculations.

Figure 8. (a) Pressure, fluid velocity, and fuel graph due to the explosion with octane, Case 1; (b)
pressure, fluid velocity, and fuel graph due to the explosion with octane, Case2.
Processes 2021, 9, 960 13 of 18
Processes 2021, 9, x FOR PEER REVIEW 14 of 20

FigureFigure 10. Pressure and temperature distribution profiles Case 1 octane explosion.
10. Pressure and temperature distribution profiles Case 1 octane explosion.
After observing the highest pressure in the confined geometry, the pressure distrib-
After observing the highest pressure in the confined geometry, the pressure distributes
utes significantly out of the confinement. The results of the pressure and temperature dis-
significantly out of for
tribution profiles thecase
confinement.
2 are shownThe results
in Figure 11. of the pressure
It shows and temperature
that the temperature inside distribu-
tionand
profiles for
outside thecase 2 are shown
confinement startedintoFigure
enhance11. It shows that the temperature inside and
remarkably.
outside the confinement started to enhance remarkably.
The temperature and pressure distribution profiles of cases 1 and 2 for the natural gas
explosion are shown in Figures 12 and 13, respectively.
Results in Figure 12 show that the distribution profiles for pressure and temperature
of natural gas explosion follow the same trend as octane explosion in case 1, but with lower
intensities of both variables. It shows that the VCE due to natural gas will have lower
destruction compared to using octane fuel.
Results in Figure 13 reveal that both temperature and pressure profiles in case 2 of
using natural gas are slightly different than shown in Figure 11. As shown in Figure 13,
there is a significant variation in pressure once the cloud explosion reaches outside the
confinement. The temperature profile shows the same behavior in Figure 13.
CFD simulation helped to understand better fire and explosion modeling for the
Amuay refinery accident and Jaipur fire; however, there are still critical criteria that chal-
lenge applying these types of software. As an illustration, simulation time profoundly
affects the performance of the simulation studies. A fast simulation speeds up the analysis
process and enables us to study different scenarios, as presented in this study. Results show
that the analysis’s precision significantly reduces due to higher grid sizing, which results
in faster simulations.
Processes 2021, 9, 960 14 of 18
Processes 2021, 9, x FOR PEER REVIEW 15 of 20

Figure
Processes 2021, 9, x FOR 11.
PEER Pressure
REVIEW
Figure and temperature
11. Pressure distribution
and temperature profiles
distribution profiles CaseCase 2 octane
2 octane explosion.
explosion. 16 of 20

The temperature and pressure distribution profiles of cases 1 and 2 for the natural
gas explosion are shown in Figures 12 and 13, respectively.

Figure 12. Pressure and temperature

Figure 12. Pressure distribution
and temperature profiles
distribution profiles for for
CaseCase
1 for a1natural
for a gas
natural gas explosion.
explosion.

Results in Figure 12 show that the distribution profiles for pressure and temperature
of natural gas explosion follow the same trend as octane explosion in case 1, but with
lower intensities of both variables. It shows that the VCE due to natural gas will have
lower destruction compared to using octane fuel.
Processes 2021, 9, 960 15 of 18

Processes 2021, 9, x FOR PEER REVIEW 17 of 20

Figure 13. Pressure and temperature distribution profiles for Case 2 for a natural gas explosion.
Figure 13. Pressure and temperature distribution profiles for Case 2 for a natural gas explosion.
Results in Figure 13 reveal that both temperature and pressure profiles in case 2 of
As using
shown in this
natural study,
gas are complex
slightly geometries
different than and processes
shown in Figure 11. As showncan make
in Figure 13,the simulation
time, forthere
precise calculations,
is a significant variationeven longer.
in pressure onceSuch delay
the cloud can only
explosion reachesbeoutside
avoidedthe with the use
confinement. The temperature profile shows the same behavior in Figure 13.
of advanced computers and a higher software capability. A high level of precision is
CFD simulation helped to understand better fire and explosion modeling for the
an essential consideration
Amuay refinery accident and thatJaipur
cannot be avoided
fire; however, instill
there are safety-related applications.
critical criteria that chal- In this
study, simulation
lenge applyingin computer
these utilized
types of software. As anthe full capacity
illustration, simulationoftime
oneprofoundly
central af- processing unit
(CPU) core. Therefore,
fects the performance the simulation’s
of the simultaneous
simulation studies. operation
A fast simulation speeds up should be decided based
the analysis
process and enables us to study different scenarios, as presented in this study. Results
on the availability of higher computer capacity. If this is not followed
show that the analysis’s precision significantly reduces due to higher grid sizing, which consistently, the
simulation time
results length
in faster will increase significantly, which will affect the study. In this
simulations.
study’s simulations,
As shown inone this critical parameter
study, complex is the
geometries andignition
processes location
can make the in simulation
the simulated complex
time, for precise calculations, even longer. Such delay can only be avoided with the use of
geometry. This study reveals differences in the values of pressure
advanced computers and a higher software capability. A high level of precision is an es-
and temperature by
changing theconsideration
sential ignition location
that cannot inbesimulations. It indicates
avoided in safety-related the effect
applications. ofstudy,
In this geometry and the
presencesimulation
of different obstacles
in computer inthe
utilized thefullgeometries
capacity of oneunder
central study.
processingAsunit
shown in Figures 8 and 9,
(CPU) core.
the overpressure changes approximately two times more than when the ignition location
was moved to a more constricted location inside the plant. These results are consistent with
the literature. A study identified that the increased confinement and congestion caused
the increase in burning rate and pressure buildup in the Jaipur fire accident [16]. The
composition of the exploding gas also has a noticeable effect on the impact of the explosion.
This study demonstrates that octane fuel has much higher released impacts than natural
gas. In the constricted ignition location, the octane fuel creates an overpressure of more
than 15 bar, whereas methane creates a 7.5 bar pressure. In both cases, the rate of decline
in fluid velocity is lesser than the drop-in pressure. The overpressure results in this study
are consistent with the values reported in the previous studies. Research work on the
Amuay refinery reported the range of overpressure between 1.5 bar and 20 bar [10]. The
Processes 2021, 9, 960 16 of 18

consistencies of results in this study with previously conducted research indicate that the
simulation results of this study are reliable. As shown in this study, the overpressure builds
up rapidly after the fuel ignition, reaching its peak value instantaneously. Results show
that the highest temperature is achieved after the pressure drops following the buildup
to the maximum. The analysis indicates that geometry complexity is essential in CFD
modeling due to changes in the ignition locations and released temperature and pressure.
The size of the extricated cloud, compared to the confined geometry at the point of the
explosion, can significantly differentiate the resulted impact scenarios.
On the other hand, the wind speed and the direction of the wind also affect the
results. The released temperature is also a critical evaluating factor in this study. A
significant temperature rise for both explosion compositions is noticed after achieving the
highest released pressure due to the explosions’ shock wave. In a process industry, hazard
reduction is achieved through robust process design. Future work is recommended to
include secondary containment, such as dikes surrounding the leaking tank, in the CFD
model. This model can help assess temperature and overpressure variations in reducing
and preventing hazards from the tanks. A risk-based economic analysis [38–40] is also
recommended to estimate the financial loss due to accidents in process industries.

5. Conclusions
Research and development are critical to ensure a high safety level at industrial plants.
There will always be hazardous conditions associated with process industries. However,
hazard identification and risk assessment can help to maintain the sustainability and
reliability of production processes. The VCEs are among the most dangerous accidents
in process industries due to a combination of fire elements such as fuel, oxygen, and an
ignition source. Therefore, there is a need to adopt preventive measures to avoid disasters
of fires and explosions. A CFD simulation tool, FLACS, used in this study, provides
valuable insight into understanding cloud explosion phenomena in the Amuay refinery
and Jaipur oil depot fire. This study shows that precision in geometrical shapes is vital in
understanding VCE, though such simulation requires higher simulation time. This work
conducts detailed analyses of the parameters influencing an explosion. This information can
improve facilities’ design in mitigating leaks and understand worst-case scenario events,
leading to catastrophic accidents. The study concludes that the geometrical complexity and
composition of the flammable fuels are amongst the critical influencing factors in modeling
VCE. As a future direction, the CFD model developed in this study could be integrated
with the layout of surrounding buildings of the accident site. This modified model would
help to simulate scenarios in terms of the layer of protection analysis (LOPA).

Author Contributions: Conceptualization, Z.S., M.K.K., A.R., F.S.M., K.V. and R.K.; Data curation,
M.K.K.; Formal analysis, A.R., F.S.M., K.V. and R.K.; Investigation, Z.S.; Methodology, Z.S., A.R.,
F.S.M., K.V. and R.K.; Project administration, Z.S. and M.K.K.; Software, Z.S., M.K.K., A.R., F.S.M.,
K.V. and R.K.; Writing—original draft, A.R., F.S.M., K.V., and R.K.; Writing—review & editing, Z.S.
and M.K.K. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.
Processes 2021, 9, 960 17 of 18

References
1. Woodward, J.L.; Pitbaldo, R. LNG Risk-Based Safety: Modeling and Consequence Analysis; John Wiley & Sons: Hoboken, NJ, USA,
2010.
2. King, R.; Hirst, R. King’s Safety in the Process Industries; Butterworth-Heinemann: Oxford, UK, 1998.
3. Tian, D.; Zhao, C.; Wang, B.; Zhou, M. A MEMCIF-IN method for safety risk assessment in oil and gas industry based on interval
numbers and risk attitudes. Eng. Appl. Artif. Intell. 2019, 85, 269–283. [CrossRef]
4. Casal, J. Evaluation of the Effects and Consequences of Major Accidents in Industrial Plants; Elsevier: Amsterdam, The Netherlands,
2017.
5. Zhou, J.; Reniers, G. Modeling, and analysis of vapour cloud explosions knock-on events by using a Petri-net approach. Saf. Sci.
2018, 108, 188–195. [CrossRef]
6. Oran, E.S.; Chamberlain, G.; Pekalski, A. Mechanisms and occurrence of detonations in vapor cloud explosions. Prog. Energy
Combust. Sci. 2020, 77, 100804. [CrossRef]
7. Johnson, D.M. The potential for vapour cloud explosions—Lessons from the Buncefield accident. J. Loss Prev. Process. Ind. 2010,
23, 921–927. [CrossRef]
8. Venart, J. Flixborough: The Explosion and its Aftermath. Process. Saf. Environ. Prot. 2004, 82, 105–127. [CrossRef]
9. Sharma, R.; Gurjar, B.; Wate, S.; Ghuge, S.; Agrawal, R. Assessment of an accidental vapour cloud explosion: Lessons from the
Indian Oil Corporation Ltd. accident at Jaipur, India. J. Loss Prev. Process. Ind. 2013, 26, 82–90. [CrossRef]
10. Mishra, K.B.; Wehrstedt, K.-D.; Krebs, H. Amuay refinery disaster: The aftermaths and challenges ahead. Fuel Process. Technol.
2014, 119, 198–203. [CrossRef]
11. Schmidt, S.; Mishra, K.B.; Wehrstedt, K.-D. CFD based reproduction of Amuay refinery accident 2012. Chem. Eng. 2016, 48.
[CrossRef]
12. Dasgotra, A.; Teja, G.V.; Sharma, A.; Mishra, K.B. CFD modeling of large-scale flammable cloud dispersion using FLACS. J. Loss
Prev. Process. Ind. 2018, 56, 531–536. [CrossRef]
13. Mishra, K.B. The influence of volume blockage ratio on IOCL Jaipur explosion. J. Loss Prev. Process. Ind. 2018, 54, 196–205.
[CrossRef]
14. Xiao, H.; Duan, Q.; Jiang, L.; Sun, J. Effects of ignition location on premixed hydrogen/air flame propagation in a closed
combustion tube. Int. J. Hydrog. Energy 2014, 39, 8557–8563. [CrossRef]
15. Schaschke, A. Dictionary of Chemical Engineering; OUP Oxford: Oxford, UK, 2014.
16. Sharma, R.K. A violent, episodic vapour cloud explosion assessment: Deflagration-to-detonation transition. J. Loss Prev. Process.
Ind. 2020, 65, 104086. [CrossRef]
17. Hansen, R.; Johnson, D.M. Improved far-field blast predictions from fast deflagrations, DDTs and detonations of vapour clouds
using FLACS CFD. J. Loss Prev. Process Ind. 2015, 35, 293–306. [CrossRef]
18. Yi, Y.; Xueqiu, H.; Geng, L.; Huia, W. Effect of meshy obstacle on methane gas explosion. Procedia Eng. 2011, 26, 70–74. [CrossRef]
19. Eckhoff, R.K. Explosion Hazards in the Process Industries; Gulf Professional Publishing: Cambridge, MA, USA; London, UK, 2016.
20. Sinha, A.; Wen, J.X. A simple model for calculating peak pressure in vented explosions of hydrogen and hydrocarbons. Int. J.
Hydrogen Energy 2019, 44, 22719–22732. [CrossRef]
21. Atkinson, G.; Cowpe, E.; Halliday, J.; Painter, D. A review of very large vapour cloud explosions: Cloud formation and explosion
severity. J. Loss Prev. Process. Ind. 2017, 48, 367–375. [CrossRef]
22. Ohep, E. More than 100 Dead in PDVSA Amuay Explosion—CEO Fedecamaras Falcon. 5 September 2012. Available online:
http://www.petroleumworld.com/storyt12090501.htm (accessed on 28 June 2020).
23. The Global Barrel. Amuay Refinery Disaster: Syrian Naphtha & Chavez ’Petroleum Revolution’ in Flames. 30 August 2012.
Available online: https://globalbarrel.com/2012/08/30/amuay-refinery-disaster-syrian-naphtha-chavez-petroleum-revolution-
in-flames/ (accessed on 28 June 2020).
24. Pearson, T. Venezuelan Report: Refinery Disaster Caused by Intentional Manipulation of Gas Pump Bolts. 10 September 2013.
Available online: https://venezuelanalysis.com/news/10013 (accessed on 29 June 2020).
25. Zhang, S.; Wang, X.; Cheng, Y.F.; Shuai, J. Modeling and analysis of a catastrophic oil spill and vapor cloud explosion in a confined
space upon oil pipeline leaking. Pet. Sci. 2020, 17, 556–566. [CrossRef]
26. Girdhar, M. Jaipur Fire and its Environmental effects. Fire Eng. 2012, 37, 21–22.
27. Lal, M.B. Committee Report; Oil Industry Safety Directorate: Noida, India, 2015.
28. Sharma, R.K.; Gurjar, B.R.; Ho, H.K.P.; Ghuge, S.P.; Agrawal, R. Lessons Learned from Industrial Chemical Accidents: A Case
Study of Indian Oil Corporation (IOC) Terminal, Jaipur, India. 2015. Available online: https://pscmembers.tamu.edu/wp-
content/uploads/2015/2015Proceedings/045_RaviSharma.pdf (accessed on 29 June 2020).
29. Hurley, M.J.; Gottuk, D.T.; Hall, J.R., Jr.; Harada, K.; Kuligowski, E.D.; Puchovsky, M.; Torero, J.L.; Watts, J.M., Jr.; Wieczorek, C.J.
SFPE Handbook of Fire Protection Engineering; Springer: Berlin/Heidelberg, Germany, 2015.
30. Dan, S.; Kim, H.; Shin, D.; Yoon, E.S. Quantitative Risk Analysis of New Energy Stations by CFD-Based Explosion Simulation. In
Computer Aided Chemical Engineering; Elsevier: Amsterdam, The Netherlands, 2012; Volume 31, pp. 305–309.
31. Gant, S.; Narasimhamurthy, V.; Skjold, T.; Jamois, D.; Proust, C. Evaluation of multi-phase atmospheric dispersion models for
application to Carbon Capture and Storage. J. Loss Prev. Process. Ind. 2014, 32, 286–298. [CrossRef]