Professional Documents
Culture Documents
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 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
A family of viruses that causes illness such as severe acute respiratory 323,000 deaths, while nearly 1.68 million people have recovered [3].
syndrome (SARS), the common cold, and Middle East respiratory syn- According to the European Centre for Disease Prevention and Control,
drome (MERS) is named a coronavirus [1]. In December 2019, a new the novel coronavirus originated from animals. It is transmitted from
coronavirus originated in Wuhan, China, known as acute respiratory human to human via tiny respiratory droplets through coughing,
syndrome coronavirus 2 (SARS-CoV-2). The disease this virus causes is sneezing, or when an affected person is nearby (at less than 1-m dis-
called coronavirus disease 2019, or COVID-19, commonly known as the tance) to other persons [4]. These droplets can be inhaled, or people can
coronavirus. Being infectious, the spread of COVID-19 was quite rapid; be infected when they touch a contaminated surface (which contains
as of March 11, 2020, there were 118,000 cases in 114 countries around such droplets) and then touch their nose, eyes, or mouth [5]. At its origin
the globe, and 4291 people were dead. Considering this life-threatening in December 2019, no vaccine was available to immunize people and
situation, on March 11, 2020, the World Health Organization (WHO) protect them from the coronavirus. Moreover, there is no standard
declared the outbreak of COVID-19 a ‘pandemic’ [2]. As of May 20, treatment for patients with the coronavirus. However, the treatment of
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 Renewable and Sustainable Energy Reviews 151 (2021) 111574
the novel coronavirus is based on dealing with its symptoms (such as integrating external and internal risks in the supply chain network (3)
fever, cough, and difficulty of breathing) and the clinical condition of a investigate the post coronavirus sustainability of the biomass supply
patient [4]. Considering the facts of rapid human-to-human disease chain and (4) identify the timeline when the biomass supply chain could
transmission and with no available vaccine for COVID-19, the world attain pre-coronavirus conditions. Pre-analysis for the modelling in-
responded by closing public and private institutions, schools, colleges, cludes identifying risk elements associated with the biomass supply
universities, local and international business ventures. Moreover, travel chain, key data sources from expert knowledge, and an in-depth litera-
restrictions have been put in place, and the enforced lockdown has been ture review. A dynamic risk assessment model will develop a causal
in effect worldwide. To prevent the spread of the coronavirus and ensure network using identified risk factors and probabilistic data. The risk
the safety of citizens, governments around the world have issued interdependencies in this study will be constructed using causal effects,
stay-at-home orders in various countries [6]. This lockdown causes i.e., the influence of one risk factor on another and throughout the
significant global environmental, social and economic impacts. The network. The model will be tested on the biomass supply chain of a U.
global environmental impact of COVID-19 includes the decrease in S.-based biofuel company, and model error will be calculated by
emissions of pollutants and greenhouse gases. For example, the Centre model validation.
for Research on Energy and Clean Air reported at the end of March 2020 This paper is organized into five major sections. The introduction
that travel bans in China had resulted in a 25% reduction in the emission describes the impact of COVID-19 on the biomass supply chain and in-
of carbon in the country over a four-week period, which amounts to 200 cludes a brief discussion on the integrated biomass supply chain. The
million metric tons of less carbon dioxide than in 2019 for the same second section is based on a literature review of the pertinent literature.
period [7]. Unlike the environmental impacts of the pandemic, social This section is further classified into two subsections: the first part
and economic impacts are detrimental. The United Nations Department presents the current methodologies used to assess the risks to the
of Economic and Social Affairs (UN DESA) identified emerging social biomass supply chain. The second part highlights the literature of the
issues due to COVID-19 and have reported that the people living in dynamic Bayesian network (DBN). The literature review is summarised
poverty are the most vulnerable due to the outbreak of COVID-19 and by identifying research gaps and the need for the current study. The
the lockdown. For example, due to a lack of safe shelter, homeless third section outlines the proposed methodology and validation and is
people have a high possibility of coronavirus exposure [8]. Moreover, further divided into two subsections, a case study of a U.S. biofuel
the absence of comprehensive universal social protection systems will company and the model’s applications to it. The fourth section is results
increase poverty and present fewer employment opportunities. The and discussion, presenting the results and reviewing the findings of this
extent of economic damage depends upon the speed of the spread of the research work. Then the research is concluded, along with suggested
coronavirus. For example, at its initiation, the spread of this coronavirus future work directions.
was rapid in China. This situation reduced automobile sales by 80%, and
exports fell 17.2% in January and February 2020. The Organization for 1.1. Integrated biomass supply chain
Economic Co-operation and Development (OECD) reported that due to
this epidemic, the global GDP growth would dwindle to 2.4% from 2.9% Typically, a biomass supply chain consists of (i) a supply chain
in 2020 and can fall to zero in a worst-case scenario [9]. The supply network including biomass handling, preprocessing and the movement
chain of a country is an integral part of its economy, and the re- of material from biomass origin to biorefineries, (ii) production of bio-
percussions of the COVID-19 outbreak have served to create and fuels in biorefineries, and (iii) product distribution to the consumers
accentuate external risks to supply chain sustainability. The Institute for [13]. Due to these business functions, the current biomass supply chain
Supply Chain Management has reported that 75% of companies have management is a conundrum that plays a crucial role in managing
been crumbling due to transportation restrictions and are facing dis- bioenergy and bioproduction processes [14]. Various risks are associ-
ruptions in their supply chain [10]. Due to the global coronavirus ated with the biomass supply chain that distinguish it from other
pandemic, business closure and lockdown implications also have inev- products’ traditional supply chains. Such risks are uncertainty in
itable and uncertain impacts on the biomass supply chain [11]. Due to biomass availability, limited biomass supply due to seasons and
the future need for sustainable energy, biomass conversion into useful weather, fluctuations in physical and chemical compositions of biomass,
bioenergy can be used as an alternate energy source to fossil fuels. biomass transport density, geographical distributions of feedstock, local
However, the current pandemic is making this goal unviable. In a recent transportation systems, and distribution infrastructures [15]. The
survey by Hawkins Wright studying the impact of COVID-19 on global complexity is compounded when costs of raw feedstock and bio-
biomass feedstock availability, 58% of wood pellet producers reported a refineries operations are compared with the market price of fossil fuels.
negative impact of the coronavirus on their business. In comparison, As highlighted by Ref. [16], the biomass supply chain is an integration of
33% reported a reduced production due to the pandemic [12]. The four system components: (i) harvesting/collection of biomass from sin-
participants who belonged to all regions of the world also indicated that gle or multiple locations and the pre-treatment, (ii) intermediate storage
82% of them are negatively affected by the coronavirus, while the of biomass, (iii) transportation, and (iv) bioconversion. In addition to
businesses of 14% of them are severely affected. Considering such these discrete processes, the final product’s (biofuel) distribution also
adverse impacts of COVID-19 on the biomass supply chain, there are needs to be embedded in the biomass supply chain, as shown in Fig. 1. In
many risks associated with the sustainability of biomass feedstock. Fig. 1, all components are interconnected and interdependent, and
These include uncertainties in the biomass supply chain, biomass pro- therefore no distinct boundary can be drawn between two pieces. In this
duction, logistics, biomass harvesting, transportation, labour availabil- supply chain, biomass supply is the first and foremost point for biofuel
ity, and high preprocessing costs. In order to ensure the sustainability of production.
biorefineries and biofuels in this uncertain time, there is a need to Biomass is defined as any renewable organic material which is
explore and understand the impact of COVID-19 on the biomass supply available regularly, such as crops and their wastes, woods, and waste of
chain. The risk assessment must primarily focus on integrating sustain- wood processing, algae, aquatic plants, animal manure, and wood and
able biomass resources, should consider the biomass symbiosis rela- its residues. Other biomass materials are waste from food processing,
tionship, interdependencies of risks in the biomass supply chain waste cooking oil, food waste, and waste materials which produce fuels,
network, the sustainability of biomass resources, given fossil fuel prices, energy, and chemicals [17]. Plant-based organic materials that are not
and the lockdown of all workers in the industry due to COVID-19. To meant to be used as food, such as palm kernel shells, rice husks, fruits’
address these challenges, this paper aims to (1) study the impact of wastes, and paddy straw, are also categorized as biomass feedstock [15].
COVID-19 on the sustainability of the biomass supply chain, (2) develop Moreover, non-edible resources such as jatropha, Linum usitatissimum
a risk assessment methodology for the biomass supply chain by (linseed), Azadirachta indica (neem), etc., and waste cooking oil are also
2
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574
used as biomass feedstock [18]. Studies have identified that various alternatives; anaerobic digestion and fermentation [27]. The physico-
factors, such as inappropriate crop harvesting, unfavourable weather, chemical transformation has various process options, such as ammonia
operational constraints, and the geographical region where the biomass fibre explosion, steam explosion, and wet explosion [28]. This conver-
is cultivated can affect biomass availability [19]. Biomass is considered sion is out of the scope of the current study. The scope of the present
to remain biologically living matter and chemically active throughout work is sketched in Fig. 2 and includes biomass supply collection to the
the biomass supply chain. For example, for algae biomass, the active gate of a biorefinery. The final product, shown in Fig. 1, could be of two
biological ingredients are proteins, polyphenols, polysaccharides (bio- categories: biofuels or biochemicals. The primary products of a bio-
polymers of glycosidic bonds and monosaccharides), polyunsaturated refinery include biofuels, such as biodiesel (produced from animal fats
fatty acids (PUFAs), minerals, and dyes [20]. The biological activity in and vegetable oils), biogas (produced from decomposing food scraps,
biomass can alter the composition of the biomass, which could affect the animal manure, yard waste in landfills) or ethanol (also called bio-
biomass’s conversion to biofuel in the biorefinery. Using waste cooking ethanol, produced from crops of sugar cane and corn). The biochemicals
oil as a source of biomass feedstock, the water contents and free fatty are derived from value-added products such as styrene and phenolics
acids are dominant factors. Hence the quality of biomass is a crucial risk used for various purposes [29]. Transportation connects all components
factor in maintaining the efficiency of the biomass supply chain. In this in a biomass supply chain, and hence transportation cost plays an
regard, appropriate harvesting and collection schedules can help ensure essential role in an economical and sustainable supply chain network. As
the quality of biomass is within the desired limitations, up to the point highlighted by Ref. [15], transportation scheduling and carriage routes
when it is delivered to the doorstep of the biorefinery. As shown in are key risk factors in biomass logistics. A viable schedule helps to
Fig. 1, pre-treatment activities are performed on the biomass collection ensure that both feedstock and finished products are delivered on time
site before sending the biomass to a biofuel processing facility. to the respective biorefinery and consumers. Selecting an appropriate
Pre-treatment of biomass includes chemical, physical, biological, and transport route helps to minimize travel time and cost; it also reduces the
physicochemical processes, as highlighted in the literature [21,22]. environmental impacts of activities in the biomass supply chain.
Depending upon the characteristics of raw material and end products,
the pre-treatment or preprocessing increases biomass porosity, enhances 2. Literature review
the efficiency of the enzyme, improves the accessibility of the enzyme,
and minimizes the losses of carbohydrates [23]. In preprocessing, the The literature review of this paper consists of two subsections. The
dense biomass feedstock is converted to be less dense, which reduces first section highlights the current state-of-the-art methodologies for risk
associated biomass costs such as transportation, handling, and storage, assessment of the biomass supply chain. The second part of the literature
and increases the product quality. The preprocessed biomass is trans- review discusses the paradigm of the dynamic Bayesian network (DBN).
ported to biorefineries, where it is converted into bioenergy (biofuel and In the past, various researchers have proposed techniques to model
electricity) through the bioconversion process. There are three leading biomass supply chains. In a study, a mixed-integer linear programming
technologies for the bioconversion process: thermochemical, biochem- (MILP) model was presented to assess the size and geographic location of
ical, and physicochemical [24,25]. Thermochemical conversion of a bioethanol plant in Austria [30]. The model studied various stages of
biomass is mainly performed using three pathways, i.e. pyrolysis, gasi- biofuel production, including biomass supply chain, heat utilization in
fication, and combustion [26]. biofuel production, and demand for fossil and biofuels. A mathematical
In contrast, biochemical conversion involves two primary process model was presented in another study to model biomass supply chain
3
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574
and logistics in a biofuel production facility [31]. The proposed model distribution) or more than two (multivariate distribution) random var-
could identify the location, number and size of biorefineries based on iables. Despite such characteristics, BN models cannot consider tempo-
available biomass. The study applied the proposed model to the biomass ral information [39], which means a BN cannot model numerous
supply chain in the State of Mississippi. Another study presented a phenomena over time. This limitation is overcome by the Dynamic
mixed-integer linear programming (MILP) model to design the supply Bayesian Network (DBN). A DBN helps to model and quantify temporal
chain risk of biomass-based ethanol [32]. The model used a case study of relationships among different data points [40]. It represents the ‘evo-
an Italian biomass-based ethanol system to analyze the ethanol supply lution’ of given random numbers as a function of the time steps sequence
chain’s strategic design and biomass planning decisions. Their model (a discrete sequence). Intuitively, a DBN is the ‘evolution’ of risk over
was able to capture the multiple biomass technology choices and time. The terminology of ‘dynamic’ in DBN does not mean the change in
considered the risk mitigation preferences of decision-makers. Another network architecture but the change in model parameters [40]. A ‘static’
work proposed a multistage stochastic programming model to deal with BN is denoted by nodes (N) and edges (E) that connect the nodes in a
uncertainties in quality (operational factor), the availability of biomass directed position. For each node (denoted by y), a conditional proba-
feedstock, and their impact on net profit [33]. The study showed a bility distribution (CPD) is associated. The network structure, consisting
trade-off between biomass quality and profit and helped control varia- of directed relationships between parents and child nodes, interpreted
tions in storage levels. In a study, a Bayesian network (BN) approach was with their CPDs, defines a BN. The factorized joint probability (FJP) on
used to predict the complex behaviour of risk propagation in the supply N (denoted by P(N)) is represented as:
chain [34]. They studied the resilience index and risk exposure indices ∏ y
for various nodes within the supply chain network. Their results showed P(N) = P( )
y∈N
πy
the vulnerable nodes due to disruption propagation in the supply chain
network and were helpful in supply chain risk management. A simula- where πy indicates the parents of child node y. This definition of
tion modelling approach was used to evaluate the risks of operational ‘static’ BN is transformed to a ‘dynamic’ mode by considering two as-
disruptions on the cost of the biomass supply chain, production level, pects: first, the initial or prior distribution of random variables in the
and inventory [35]. The study considered three scenarios, and the static BN at time 0. Secondly, the dynamic two-step BN is investigated;
simulation model was analyzed for seven years. The study showed that this describes the transformation from the previous time t-1 to the cur-
risk vindication alone at a facility is not enough. The concurrent inclu- rent time t, having the probability P( yt | yt−1 ) for any node y, which
sion of failure dependencies at the system level is required for the sus- belongs to N in a DAG. Hence, the JP for two sets of nodes, namely Nt
tainable design of the biomass supply chain. and Nt−1 , is:
In a recent study, researchers studied operational and disruption ∏
Nt yt
risks to the bioethanol supply chain [36]. They proposed a MILP based P( )= P( )
model to identify, analyze the design, and risk planning of a supply chain
Nt−1 y∈N, π y∈N
πyt
network of multi-feedstock lignocellulosic bioethanol. Using a case
The computation of the JP law is performed by ascertaining the
study in Iran, their model provided optimal supply chain decision var-
sequence in the network over a series of times. For the total length of the
iables. In another study, investigators evaluated reduction opportunities
path Z and JP of the initial network P(N0 ), the probability of trans-
of supply, market, and operational risks for the supply chain of a
forming from N0 to Z is computed as:
cellulosic biorefinery, which adopted a distributed depot concept [37].
Their study was based on geographically distributed depots and showed ∏ y0 ∏Z ∏
yt
that operational and market risks were reduced for a distributed depot P(N0:Z ) = P(N0 ).P(N1:Z ) P( )⋅ P( )
y∈N
π (x 0 ) t=1 y∈N
π (y t)
supply system.
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,
2.1. Dynamic Bayesian Network (DBN) denoted as t-1 [41]. A DBN generalizes the Hidden Markov Model
(HMM). A more complex DBN could have multiple input variables
A Bayesian Network (BN) is a directed acyclic graph (DAG) in which connecting last time slices, as shown in Fig. 3. A DBN defines the de-
nodes represent random variables, and the directed arcs show the con- pendencies between parent and child nodes over time. As shown in
ditional dependency of one node (called parent) on another (called Fig. 3, nodes (y1 to y5) in a DBN are still linked through a DAG;
child) [38]. The influence of parent on child corresponds to a directed nevertheless, cycles between nodes are allowed in the DBN. A DBN is a
edge. The strength of parent and child nodes is defined by assigning series of BNs developed for varying time units; each BN is called a time
marginal and conditional probability tables. A BN characterizes a slice. As shown in Fig. 3, temporal links connect time slices, and the
factorization of joint probability (JP) distribution over two (bivariate
4
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574
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 the precoronavirus situation? What will be the impact of fuel prices on
can help to study the dependence of current and previous years’ varia- biomass availability? The results from this research work can help
bles—for example, the dependence of this year’s biomass supply on last governments, investors, biomass supply chain managers and biomass
year’s rain. In DBN, all times are zero-based, which means t = 0, rather business owners to make informed decisions as they assess the
than t = 1, which indicates the first time step. post-coronavirus situation. This study can help them to evaluate,
Previously, the DBN has been used for modelling various systems manage, minimize, and mitigate the risks to their businesses due to
such as predictions of crop growth [42], process safety [43], system COVID-19. The dynamic risk modelling framework for the biomass
performance measurements [44] and healthcare [45]. Although the supply chain risk is applied to a biofuel company located in Washington
DBN has been used in various fields of study, it has never been used to state of the United States of America (USA). The case study helps to
address the dynamic behaviour of the biomass supply chain system. To illustrate the proposed methodology, results and findings.
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- 3. Methodology
dicates two categories of biomass supply chain risks: internal opera-
tional risks and internal supply chain risks [46]. Internal operational The research methodology in this study consists of five steps. The
risks refer to risks to the regular operations, such as risks to the distri- prime objective of this work is to investigate the temporal and quanti-
bution, manufacturing, process improvements, product development, tative strengths of the relationships among various risk factors of the
and infrastructure. The internal supply chain risks indicate the potential biomass supply chain. The definition of strength is reported in the
risks to the upstream and downstream supply chain of the biomass, such literature [47] and is applied here. The input data in this study were the
as unpredictable demand by consumers or interruptions in the avail- interdependency relationships among risk factors, which were defined
ability of biomass feedstock or bioproducts. As evident through the using experts’ elicitation. In this study, experts were a research group
literature review, the traditional risk assessment techniques have only consisting of three senior university professors in a U.S. university who
focused on internal operational and internal supply chain risks to the have an in-depth knowledge of the biomass supply chain and extensive
biomass network. However, the biomass supply chain risks are evolving, experience in chemical engineering research and development. The
especially with the current pandemic conditions due to COVID-19. The steps followed include:
impacts of external risks on the biomass supply chain and the in-
teractions of external risks with supply chain components have never Step 1: Identification of risk categories
been addressed. This work aims to fill this research gap. This paper has
three primary contributions and innovations; (1) prior research on In this step, experts identified six risk categories associated with the
biomass supply chains has been focused on studying the static behaviour biomass supply chain. Those were qualitative factors, quantitative fac-
of the biomass supply chain and does not consider time variability in the tors, labour, biomass transportation, oil, and the pandemic. The quali-
supply chain system, and hence cannot provide biomass supply chain tative factors were defined as those risk factors related to the quality or
behaviour over time. This research investigates such dynamic behav- characteristics of biomass, such as the chemical and biological compo-
iour. This study provides a dynamic framework to analyze the evolution sition of the biomass. Risk factors in quantitative conditions were
of biomass supply chain risks over time. (2) This paper is the first to defined as those risk factors that could affect the quantity or amount of
assess the probabilistic impact of COVID-19 on a biomass supply chain biomass available, such as the closure of biomass production sites. The
and identifies the recovery time of loss due to the coronavirus. (3) This labour category consisted of those risk factors related to the workforce
study provides a risk network model which captures the uncertainties of employed in the biomass supply chain, such as labour availability. The
biomass supply chains. It aims to answer the questions of the future biomass transportation category was defined as risk factors associated
impact of COVID-19 on the biomass supply chain. How much will the with moving biomass from one point to another—for example, the
cost of the biomass supply chain be affected due to this coronavirus? transportation schedule. The oil category was described as a category
How much is the sustainability of the biomass supply chain affected due that consists of risk factors related to fossil-based oil, either diesel or
to the ongoing pandemic? When will the biomass supply chain return to gasoline, used in the transportation of the biomass in the biomass supply
5
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574
chain network: for example, oil demand. The pandemic risk category of 100 cases (n = 100) was compiled, and each case had values of the
was defined as those risk factors based on the epidemic situation, such as response variable (biomass feedstock gate availability) and covariates.
a countrywide lockdown. The database file consisted of 100 samples with known values of input
nodes (as presented in the model) and the node of ‘biomass feedstock
Step 2: Identification of risk factors gate availability’ (final output node). The randomized data file was
divided into ten equal sets or segments (k-value in k-fold validation).
In this step, risk categories defined in step 1 were further classified The value of 10 sets is the optimal value of k in k-fold cross-validation, as
into risk factors using previously available studies [37,47–52]. The supported in the literature [60]. The first ten parts of data, representing
identification of risk factors is based on the risk category under study ten sets (n = 100/k = 10 = 10), were set aside, and the model was
and the type of raw material used to produce biofuel—for example, parameterized with the leftover cases, i.e., (n-n/k). Then the model was
waste cooking oil or crops. tested against other sets of the database for accuracy and comparison of
model results to known outcomes in each case. This procedure was
Step 3: Dynamic Bayesian Network (DBN) modelling repeated for all ten segments. GeNIe was used as a tool to develop
randomized subsets and to perform k-fold cross-validation. Results were
In this step, a DBN model was developed. This task was accomplished presented and discussed in terms of the accuracy of nodes and selected
in two phases, (a) structural learning and (b) parameter learning. receiver operating characteristic (ROC) curves.
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 3.1. Case study of a U.S. Biofuel company
marginal probability distributions of root nodes in the network and
conditional probability distributions of non-rooted nodes in the BN. A To demonstrate the applications of the proposed methodology, a case
detailed description of the two steps is provided below. 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
Step 3a. Structural learning 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
This step consisted of building the interdependency relationships confirmed cases in different states is shown in the choropleth map
among risk factors through experts’ knowledge. The expert elicitation sketched in Fig. 4. On March 13, 2020, the U.S. government declared a
technique was used to define such relationships, as discussed in the national emergency to deal with the COVID-19 crisis [62]. As of April 7,
literature [53]. This structured expert elicitation technique has been 2020, several states ordered the closure of businesses, workplaces, and
extensively used in the literature, such as in Refs. [54–56]. Experts schools and enforced social distancing [63], followed by a partial or
defined the interdependencies among stochastic risk factors based on statewide lockdown all over the U.S., as shown in the choropleth map of
the causal relationships between stochastic nodes. Once the relation- Fig. 5. This situation means that the U.S. economy was shut down, and
ships were identified, the directed arcs were drawn between nodes as per employees and employers could not work or receive customers. Road,
causality defined by the experts. The result of this step was a graph air, and sea transportation were also limited. This economic shutdown
structure of B.N. also closed biomass business units across the country. In the U.S.,
various biomass feedstocks are being used, such as crops (sugar beets
Step 3b. Parameter learning and sugarcane), crop residues (bagasse, barley straw, corn stover, rice
straw, grain sorghum stubble, and wheat straw), wood (forest residues,
In this step, prior probability distributions were defined for root primary mill residues, urban wood, secondary mill residues), and waste
nodes at time t = 0 (first-time step), and conditional probability tables cooking oil [18]. For this research work, Mahoney Environmental®
(CPTs) were defined for non-root nodes at time t = 0. Since COVID-19 is company, located in Washington state, was chosen to elaborate on the
new and there are no historical data available, experts’ judgments were methodological framework. The company collects waste cooking oils,
used in this step. The outcome of this step was a B.N. with temporal which are used to produce biodiesel, and has been a leader in recycling
relationships, i.e., a DBN. The analysis was performed for ten time steps waste cooking oil for over 65 years in the U.S. Regularly, the company
(t = 9). GeNIe Modeler by Bays Fussion, LLC (https://www.bayesfusion. collects used cooking oils from more than thousands of shopping malls,
com/genie/) was used as a tool to build DBN and perform Bayesian food restaurants, bars, commercial food producers, stadiums and mili-
inference, as discussed in the next step. tary stations across Washington state and produces low-cost commer-
cial-grade biodiesel along with co-products such as solvents, lubricants
Step 4. Risk-based inference 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
In this step, the risk-based inference was performed using the DBN locations for waste cooking oil, shown in Fig. 6, and details of all 13
developed in step 3. The temporal feature of GeNIe allowed updating sites are presented in Table 1. Due to COVID-19 and the lockdown, the
beliefs of all time steps at once rather than observing results one at a collection of used cooking oil, its transportation to the chemical plant
time. This step provided the posterior probabilities of each node in DBN, site, and biodiesel distribution have been negatively affected. This study
which were conditionally based on the evidence. The computational aims to analyze the probabilistic magnitude of this impact and assess the
mechanism of solving DBN has been presented earlier. The probabilistic risk-based performance of the biomass supply chain using the proposed
results were reported and discussed. 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
Step 5: Model validation the gate of Mahoney Environmental® for the pre-treatment of waste
cooking oil and subsequently the shortfall of biomass for biodiesel
As a final step of this methodology, the model presented was vali- production. The objective of the current case study is to identify and
dated using the k-fold cross-validation method. This method helped to analyze the risks to waste cooking oil feedstocks which would help
provide information on the robustness and accuracy of the model in ensure a continuous supply to the Mahoney Environmental recycles
predicting the outcomes. The use and reliability of this method to pre- under investigation. Steps presented in the methodology section are
dict model accuracy has been extensively studied in the literature, for applied to this case study, and results are presented and discussed in the
example [57–60]. A database file (presented in Data-In-Brief) consisting next section.
6
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).
8
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574
9
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 Renewable and Sustainable Energy Reviews 151 (2021) 111574
aids each node to be conditionally dependent on its parents. At the same probability to reduce to 0.084. As shown in Fig. 10, this value should
time, DBN also allows a node to be conditionally dependent on its par- approach zero, as in the pre-coronavirus situation; however, such results
ents, at previous time slices. are not obtained until the fifth year. In other words, it will take five years
The latter behaviour is shown in Fig. 8 as a number within the small for the biomass supply chain to complete recovery from the COVID-19
square. For example, in Fig. 8, number 2 in the small green square on hit. It is noteworthy here that the U.S. House of Representatives, on
low fuel demand indicates that there are two time slices in temporal May 15, 2020, voted to pass the COVID-19 stimulus package. The bill,
dependencies of low fuel demand. While the arc between transport and known as the ‘Health and Economic Recovery Omnibus Emergency
low fuel demand shows number 4 in a small square, this arc shows the Solutions Act or the HEROES Act’ aims to provide relief to biofuel
temporal influence, while its number indicates its order, also called lag. producers in the U.S. The financial benefits include providing a 45 cents
Lag in DBN shows the order of the temporal (time-based) link. The per gallon payment to qualified biofuel producers from January 1, 2020,
number 4 suggests that the influence of transport spans over four time to May 1, 2020 [65]. For interested readers, the entire bill is available at
steps. In other words, a temporal link with order four from the transport the House of Representatives, accessible at [66]. With these financial
node to another node, low fuel demand, as shown in Fig. 8, is interpreted incentives, the time of five years of recovery, as revealed in this study,
as transport links to low fuel demand in the later future, i.e., four-time may be shortened.
intervals or steps. The low fuel demand node has a link from the The DBN model provided lays the imperative groundwork for
transport node in the past, i.e., four-time intervals or steps in the past. probabilistic approaches to assess the disruption to the biomass supply
Results in Fig. 8 show that the transportation cost is influenced by low chain due to fuel prices. Supply chain networks of biomass feedstock
fuel prices, sustainable transport planning, high preprocessing cost, and may be complex systems. They may comprise many interrelated and
less quantity of biomass available. interconnected factors that may introduce nonlinearity into the system.
In terms of the quality of biomass, changes in biomass composition, Depending upon the complexity of the network, discretization can be
and the absence of the usual collection schedule, cause low quality introduced into the model, which can help capture and compute non-
biomass. Results in Fig. 8 also show that a decrease in labour availability linear relationships among risk variables [67]. Moreover, discretiza-
results in high operational costs. The graphical structure of the DBN tion would help to reduce computational power to solve the model [68].
model, shown in Fig. 8, is similar to its static BN model, shown in Fig. 7, The results of the impact of COVID-19 on fuel prices are drawn in
except that there are additional arcs in the DBN model that quantify the Fig. 11. Fig. 11 shows that the probability of the fuel price being lower is
temporal relationships among risk factors. The dynamic or temporal arcs higher at the start of the coronavirus spread, shown in the blue colour
represent the changes in the probabilistic values of risk variables over line. In other words, the likelihood of fuel prices being higher is low at
time. Fig. 9 captures two unrolled time steps (Step 0 and 1) of the DBN the beginning of the coronavirus spread (shown in the orange colour
model presented in Fig. 8. Unrolling helps to covert DBN into its line). Values of both curves in Fig. 11 are complementary to each other.
equivalent static BN, which quickly helps to understand the DBN The DBN predicts that the probability of the fuel price being higher will
structure. As shown in Fig. 9, all variables are repeated in each time step gradually start to increase and will take one year to reach a value of
except the pandemic, effectiveness of vaccine and lockdown. Such 0.84; after this time, there is a gradual increase in fuel prices over time.
behaviour of the model is because these three variables do not repeat in However, the fuel price does not reach a pre-coronavirus situation, even
each time step but influence the rest of the variables in each time step. after nine years, as shown in Fig. 11. The increasing trend in fuel prices
Considering the observed dynamic evidence of QT5, BT4, and L2, the in Fig. 11 is compared with fuel price trends from the U.S. Energy In-
DBN model encodes the probability distribution over gate availability of formation Administration (EIA). Their data of weekly U.S. regular
biomass feedstock. This simulation results in calculating the risk of convention retail prices of gasoline are drawn in Fig. 12. The data of the
having no biomass feedstock available at the factory gate, given the U.S. EIA show that after May 4, 2020, there is an increasing trend in fuel
conditions of low transportation cost, high operational cost, and less price. Hence, this confirms the results of this study. The methodological
quantity of biomass available at a given time, i.e., P(FGA t/E1), where E framework presented in this study is consistent with [69], in which re-
= QT5t-1 = less, BT4t-1 = low, L2t-1 = High. It shows that conditional searchers used DBN to model risk assessment of flood control operations.
probability distribution for gate availability of biomass feedstock de- As the DBN model shows the changes in probabilities of the various
pends on the results in the previous time step for risk factors of the risk factors of the biomass supply chain over time, it can help develop
available quantity of biomass, transportation cost, and operational cost. contingency plans to manage risks to the biomass supply chain. To
The results are drawn in Fig. 10, which shows the probability of having demonstrate the operationalization of the model, this study chose a
no biomass feedstock at the factory gate, given this dynamic evidence, i. biofuel production facility that utilizes waste cooking oil as a raw ma-
e., P (FGA (no)/E). The plot in Fig. 10 shows the risk of having no terial. However, the model presented can be applied to supply chain
biomass feedstock availability at the company’s gate over ten time steps. networks of other biofuel raw materials such as sugar beet, corn, sug-
This result can help develop production planning and the optimal time arcane, jatropha, grass, switchgrass, woody crops (lignocellulosic
for biodiesel production in a pandemic situation. biomass), woodchips, agricultural residues, soybean oil, rapeseed oil, or
Results in Fig. 10 show that in the pre-coronavirus situation, gate other plant oils. In any case, the risk categories of quality should include
availability of biomass feedstock was regular, and there was no external risk factors such as biological activities and a proper harvesting
risk of COVID-19. Fig. 10 shows that when the pandemic hits Wash- schedule. The risk category of quantity should consist of unseasonable
ington state, the probability of unavailability of biomass supply at the weather, rainfall, and droughts. However, the risk factor of trans-
company’s gate reaches its maximum value of 0.98 in the lockdown portation should be based on vehicles to collect the biomass from fields
situation. The high unavailability of biomass indicates the need to secure and its transportation to bioconversion sites. The methodological
waste cooking oil from more remote locations than shown in Table 1. framework presented is consistent with [69], in which, using experts’
This action may not be environmentally friendly, as more vehicle opinions, researchers developed a DBN model to perform a risk assess-
transportation would release more carbon by fuel burning, and nega- ment of flood control operations. Future work is recommended to
tively affect the environment. Moreover, collecting waste cooking oil expand the analysis by including the risk factors associated with
from farther places would not be economical, as it would involve the bioconversion and biofuel delivery to customers. Since this will intro-
additional costs of the operations of the biomass supply chain. duce structural complexity in the network, a dynamic object-oriented
Results also indicate that it would take at least one year for Bayesian network approach [71,72] is recommended.
1
Evidence.
11
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574
4
Each point on a ROC curve is called a classifier. COVID-19 has hit the whole world and has affected various
13
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574
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 Nevertheless, coordination at a global level can help to curb COVID-19
has significant repercussions for the sustainability of the biofuel in- impacts and bolster the world’s bioeconomies.
dustry. As COVID-19 is new, adequate risk management strategies are
essential to ensure the sustainability of the biomass supply chain. Credit author statement
Although the biofuel industry is nascent in this crucial time, the per-
formance of the biomass supply chain has become more critical. In this Zaman Sajid: Conceptualization; Data curation; Formal analysis;
study, a risk-based dynamic Bayesian network (DBN) model is presented Investigation; Methodology; Project administration; Resources; Soft-
to assess the biomass supply chain’s risk over ten years. Biomass supply ware; Validation; Visualization; Roles/Writing - original draft; Writing –
chain risk is defined as the unavailability of biomass feedstock to pro- review & editing.
duce biofuel. Dealing with such risk is vital since an uninterpreted
continuous supply of the biomass supply chain will guarantee the Declaration of competing interest
optimal production of biofuel. The presented model is applied to a U.S.
biofuel company. The results indicate that the pandemic has affected the The authors declare that they have no known competing financial
supply chain of biomass, and there are 85.01% chances that the biomass interests or personal relationships that could have appeared to influence
collected in a pandemic will be of low quality. With the development of the work reported in this paper.
vaccinations and partial reopening of businesses, this chance falls to
54.23%. Results also indicate that low-quality biomass increases the Acknowledgments
preprocessing cost of biomass feedstock to 84.60% during the pandemic.
Study shows that due to lockdown and lower demand of fossil-based oil, The author would like to thank three anonymous professors at the
fuel prices would drop to 89% of the regular prices. The dynamic nature University of Pittsburgh, USA, who provided valuable comments to
of the model indicates a gradual increase in prices and demand for fossil improve the contents of this work. The author would also like to thank
fuels. Results show that the fuel prices will not reach a pre-coronavirus anonymous reviewers and the editorial team for their comments and
situation, even after nine years of businesses’ reopenings. Biomass FGA suggestions to improve this work.
is low at the start of the pandemic, with the unavailability of 98%.
However, vaccination availability helps to drop this unavailability to References
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 [1] Mayo C. Coronavirus disease 2019 (COVID-19) - symptoms and causes - mayo
clinic. https://www.mayoclinic.org/diseases-conditions/coronavirus/symptoms-c
ROC curves drawn indicate that the data used in this study are not auses/syc-20479963. [Accessed 20 May 2020].
random guesses and have an accuracy of 95% for all time steps. [2] WHO. WHO Director-General’s opening remarks at the media briefing on COVID-
Measuring risk to the biomass supply chain in this study helps assess the 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-
biorefinery’s “riskiness” concerning the risk of fuel market prices. Since 19—11-march-2020. [Accessed 20 May 2020].
COVID-19 is new and the biofuel industries have never seen its impact in [3] CSSE. Coronavirus COVID-19 (2019-nCoV). Johns Hopkins Univ; 2020.
the past, the risk data aresparse, and therefore expert elicitation was https://gisanddata.maps.arcgis.com/apps/opsdashboard/index.html#/bda759474
0fd40299423467b48e9ecf6. [Accessed 20 May 2020].
used in this study. The results of this study can help biofuel investors, [4] ECDC Q, COVID-19 A on. Eur Cent Dis Prev Control 2020. https://www.ecdc.
managers, and biofuel policy-makers to make well-informed decisions europa.eu/en/covid-19/questions-answers. [Accessed 20 May 2020].
on the sustainability of the biomass supply chain in the post-coronavirus [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/
situation. In conclusion, COVID-19 is causing the performance of the
question-and-answers-hub/q-a-detail/q-a-coronaviruses. [Accessed 21 May 2020].
biomass supply chain to plummet and is presenting challenges to the
sustainability of the biomass supply chain and biorefineries.
14
Z. Sajid Renewable and Sustainable Energy Reviews 151 (2021) 111574
[6] Haddad M. Coronavirus: how much more time are people spending at home? | [35] Sharma B, Clark R, Hilliard MR, Webb EG. Simulation modeling for reliable
News | Al Jazeera. 2020. https://www.aljazeera.com/news/2020/04/corona biomass supply chain design under operational disruptions. Front Energy Res 2018;
virus-world-staying-home-200406122943899.html. [Accessed 22 May 2020]. 6:100.
[7] Myllyvirta L. Analysis: coronavirus temporarily reduced China’s CO2 emissions by [36] Mousavi Ahranjani P, Ghaderi SF, Azadeh A, Babazadeh R. Robust design of a
a quarter. Carbon Br 2020. https://www.carbonbrief.org/analysis-coronavirus-h sustainable and resilient bioethanol supply chain under operational and disruption
as-temporarily-reduced-chinas-co2-emissions-by-a-quarter. [Accessed 22 May risks. Clean Technol Environ Policy 2020;22:119–51. https://doi.org/10.1007/
2020]. s10098-019-01773-2.
[8] UN DESA. The social impact of COVID-19 | DISD. United Nation Dep Econ Soc Aff; [37] Mamun S, Hansen JK, Roni MS. Supply, operational, and market risk reduction
2020. https://www.un.org/development/desa/dspd/2020/04/social-impact-of-co opportunities: managing risk at a cellulosic biorefinery. Renew Sustain Energy Rev
vid-19/. [Accessed 22 May 2020]. 2020;121:109677. https://doi.org/10.1016/j.rser.2019.109677.
[9] Segal S, Gerstel D. The global economic impacts of COVID-19 | center for strategic [38] Nielsen TD, Jensen FV. Bayesian networks and decision graphs. Springer Science &
and international studies. Cent Strateg Int Stud 2020. https://www.csis.org/analys Business Media; 2009.
is/global-economic-impacts-covid-19. [Accessed 22 May 2020]. [39] Liang H, Ganeshbabu U, Thorne T. A dynamic Bayesian network approach for
[10] Boyd J. COVID-19 Survey: impacts on global supply chain. USA: Institute for analysing topic-sentiment evolution. IEEE Access 2020;8:54164–74.
Supply Management ® study reveals virus’ supply chain effects. Tempe; 2020. [40] Murphy KP. Dynamic bayesian networks. Probabilistic Graph Model M Jordan
[11] Forestry B. Covid-19 Biomass supply chain response - Usewoodfuel. Usewoodfuel 2002;7:431.
Scotl 2020. https://usewoodfuel.co.uk/covid-19-biomass-supply-chain-response/. [41] Benhamou E, Atif J, Laraki R. A new approach to learning in Dynamic Bayesian
[Accessed 22 May 2020]. Networks (DBNs). 2018. ArXiv Prepr ArXiv181209027.
[12] Wright H. Survey: COVID-19 impacts raw material availability for pellets | [42] Kocian A, Massa D, Cannazzaro S, Incrocci L, Di Lonardo S, Milazzo P, et al.
Biomassmagazine.com. Biomass Mag; 2020. http://biomassmagazine.com/articles Dynamic Bayesian network for crop growth prediction in greenhouses. Comput
/17056/survey-covid-19-impacts-raw-material-availability-for-pellets. [Accessed Electron Agric 2020;169:105167. https://doi.org/10.1016/j.
23 May 2020]. compag.2019.105167.
[13] Bairamzadeh S, Pishvaee MS, Saidi-Mehrabad M. Multiobjective robust [43] Cai B, Liu Y, Liu Z, Chang Y, Jiang L. A multiphase dynamic bayesian network
possibilistic programming approach to sustainable bioethanol supply chain design methodology for the determination of safety integrity levels. Bayesian networks
under multiple uncertainties. Ind Eng Chem Res 2016;55:237–56. https://doi.org/ reliab. Eng. Springer; 2020. p. 217–37.
10.1021/acs.iecr.5b02875. [44] Ramírez PAP, Utne IB. Use of dynamic Bayesian networks for life extension
[14] Gold S, Seuring S. Supply chain and logistics issues of bio-energy production. assessment of ageing systems. Reliab Eng Syst Saf 2015;133:119–36. https://doi.
J Clean Prod 2011;19:32–42. org/10.1016/j.ress.2014.09.002.
[15] Hong BH, How BS, Lam HL. Overview of sustainable biomass supply chain: from [45] Li M, Liu Z, Li X, Liu Y. Dynamic risk assessment in healthcare based on Bayesian
concept to modelling. Clean Technol Environ Policy 2016;18:2173–94. approach. Reliab Eng Syst Saf 2019;189:327–34. https://doi.org/10.1016/j.
[16] Iakovou E, Karagiannidis A, Vlachos D, Toka A, Malamakis A. Waste biomass-to- ress.2019.04.040.
energy supply chain management: a critical synthesis. Waste Manag 2010;30: [46] Christopher M, Peck H. Building the resilient supply chain. 2004.
1860–70. https://doi.org/10.1016/j.wasman.2010.02.030. [47] Sajid Z, Khan F, Zhang Y. Integration of interpretive structural modelling with
[17] Sundaram S. Biorefineries and chemical processes: design, integration and Bayesian network for biodiesel performance analysis. Renew Energy 2017;107.
sustainability analysis. Green Process Synth 2015;4:65–6. https://doi.org/10.1016/j.renene.2017.01.058.
[18] Sajid Z, Zhang Y, Khan F. Process design and probabilistic economic risk analysis of [48] Amundson J, Faulkner W, Sukumara S, Seay J, Badurdeen F. A bayesian network-
bio-diesel production. Sustain Prod Consum 2016;5:1–15. https://doi.org/ based approach for risk modeling to aid in development of sustainable biomass
10.1016/j.spc.2015.10.003. supply chains. In: Bogle IDL, fairweather MBT-CACE. 22 Eur. Symp. Comput. Aided
[19] Lozano-Moreno JA, Maréchal F. Biomass logistics and environmental impact process Eng., vol. 30. Elsevier; 2012. p. 152–6. https://doi.org/10.1016/B978-0-
modelling for sugar-ethanol production. J Clean Prod 2019;210:317–24. https:// 444-59519-5.50031-9.
doi.org/10.1016/j.jclepro.2018.10.310. [49] Rauch P. Developing and evaluating strategies to overcome biomass supply risks.
[20] Konkol D, Górniak W, Świniarska M, Korczyński M. Algae biomass in animal Renew Energy 2017;103:561–9. https://doi.org/10.1016/j.renene.2016.11.060.
production. Algae biomass charact. Appl. Springer; 2018. p. 123–30. [50] Sajid Z, Khan F, Zhang Y. A novel process economics risk model applied to a
[21] Chiaramonti D, Prussi M, Ferrero S, Oriani L, Ottonello P, Torre P, et al. Review of biodiesel production system. Renew Energy 2018;118:615–26. https://doi.org/
pretreatment processes for lignocellulosic ethanol production, and development of 10.1016/j.renene.2017.11.022.
an innovative method. Biomass Bioenergy 2012;46:25–35. [51] Williams PRD, Inman D, Aden A, Heath GA. Environmental and sustainability
[22] Mood SH, Golfeshan AH, Tabatabaei M, Jouzani GS, Najafi GH, Gholami M, et al. factors associated with next-generation biofuels in the US: what do we know?
Lignocellulosic biomass to bioethanol, a comprehensive review with a focus on Environ Sci Technol 2009;43:4763–75.
pretreatment. Renew Sustain Energy Rev 2013;27:77–93. [52] Zhang Y, Hu Q, Xiao D, Liu X, Shen Y. Spatial-temporal change of agricultural
[23] Zhang K, Pei Z, Wang D. Organic solvent pretreatment of lignocellulosic biomass biomass and carbon capture capability in the mid-south of hebei province. In: Li R,
for biofuels and biochemicals: a review. Bioresour Technol 2016;199:21–33. Monti A, editors. L. Alloc. Biomass crop. Challenges oppor. With chang. L. Use.
https://doi.org/10.1016/j.biortech.2015.08.102. Cham: Springer International Publishing; 2018. p. 159–87. https://doi.org/
[24] Adams P, Bridgwater T, Lea-Langton A, Ross A, Watson I. Chapter 8 - biomass 10.1007/978-3-319-74536-7_9.
conversion technologies. In: Thornley P, adams PBT-GGB of BS. Academic Press; [53] Ayyub BM. Elicitation of expert opinions for uncertainty and risks. CRC Press;
2018. p. 107–39. https://doi.org/10.1016/B978-0-08-101036-5.00008-2. 2001.
[25] Sajid Z, Khan F, Zhang Y. Process simulation and life cycle analysis of biodiesel [54] Larkin P, Gracie R, Shafiei A, Dusseault M, Sarkarfarshi M, Aspinall W, et al.
production. Renew Energy 2016;85. https://doi.org/10.1016/j. Uncertainty in risk issues for carbon capture and geological storage: findings from
renene.2015.07.046. structured expert elicitation. Int J Risk Assess Manag 2019;22:429–63.
[26] Sadhwani N, Liu Z, Eden MR, Adhikari S. Simulation, analysis, and assessment of [55] Hemming V, Burgman MA, Hanea AM, McBride MF, Wintle BC. A practical guide
CO2 enhanced biomass gasification. In: Kraslawski A, Turunen I, 23rd, editors. Eur. to structured expert elicitation using the IDEA protocol. Methods Ecol Evol 2018;9:
Symp. Comput. Aided process eng., vol. 32. Elsevier; 2013. p. 421–6. https://doi. 169–80.
org/10.1016/B978-0-444-63234-0.50071-3. [56] Schmidt O, Gambhir A, Staffell I, Hawkes A, Nelson J, Few S. Future cost and
[27] Gautam P, Kumar S, Lokhandwala S. Chapter 11 - energy-aware intelligence in performance of water electrolysis: an expert elicitation study. Int J Hydrogen
megacities. In: Kumar S, Kumar R, Pandey A, editors. Curr. Dev. Biotechnol. Energy 2017;42:30470–92.
Bioeng. Elsevier; 2019. p. 211–38. https://doi.org/10.1016/B978-0-444-64083- [57] Arlot S, Celisse A. A survey of cross-validation procedures for model selection. Stat
3.00011-7. Surv 2010;4:40–79.
[28] Papadokonstantakis S, Johnsson F. Biomass conversion technologies - D3.1 Report [58] Govindarajan M. Text mining technique for data mining application. Proc world
on definition of parameters for defining biomass conversion technologies. 2017. Acad Sci Eng Technol 2007;26:544–9. Citeseer.
Gothenburg. [59] Geroldinger A, Lusa L, Nold M, Heinze G. On resampling methods for model
[29] Pandey A, Mohan SV, Chang J-S, Hallenbeck PC, Larroche C. Biomass, biofuels, assessment in penalized and unpenalized logistic regression. ArXiv Prepr
biochemicals. Biohydrogen. Elsevier; 2019. ArXiv210107640. 2021.
[30] Leduc S, Schwab D, Dotzauer E, Schmid E, Obersteiner M. Optimal location of [60] Marcot BG, Hanea AM. What is an optimal value of k in k-fold cross-validation in
wood gasification plants for methanol production with heat recovery. Int J Energy discrete Bayesian network analysis? Comput Stat 2020:1–23.
Res 2008;32:1080–91. [61] COVID-19 United JH. States cases by county - Johns Hopkins coronavirus resource
[31] Ekşioğlu SD, Acharya A, Leightley LE, Arora S. Analyzing the design and center. Johns Hopkins Univ; 2020. https://coronavirus.jhu.edu/us-map. [Accessed
management of biomass-to-biorefinery supply chain. Comput Ind Eng 2009;57: 10 May 2020].
1342–52. [62] Cathey L. Government coronavirus response: trump declares a national emergency,
[32] Giarola S, Bezzo F, Shah N. A risk management approach to the economic and says he “likely” will get tested - ABC News 2020 [)], https://abcnews.go.com/Poli
environmental strategic design of ethanol supply chains. Biomass Bioenergy 2013; tics/government-coronavirus-response-live-updates-trump-declares-national/story
58:31–51. https://doi.org/10.1016/j.biombioe.2013.08.005. ?id=69580277. [Accessed 10 May 2020].
[33] Shabani N, Sowlati T. A hybrid multi-stage stochastic programming-robust [63] Train R. Coronavirus lockdown in the USA: which states and what measures have
optimization model for maximizing the supply chain of a forest-based biomass been applied? AsCom. 2020. https://en.as.com/en/2020/04/07/other_sports/158
power plant considering uncertainties. J Clean Prod 2016;112:3285–93. https:// 6293073_218312.html. [Accessed 11 May 2020].
doi.org/10.1016/j.jclepro.2015.09.034. [64] Schumaker E. Here are the states that have shut down nonessential businesses -
[34] Ojha R, Ghadge A, Tiwari MK, Bititci US. Bayesian network modelling for supply ABC News. ABC News; 2020. https://abcnews.go.com/Health/states-shut-essent
chain risk propagation. Int J Prod Res 2018;56:5795–819. ial-businesses-map/story?id=69770806. [Accessed 11 May 2020].
15
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
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
Citation: Sajid, Z.; Khan, M.K.;
and storage tanks of flammable chemicals. In this regard, the Computational Fluid Dynamics (CFD)
Rahnama, A.; Moghaddam, F.S.;
models are more appropriate, as these models provide three-dimensional (3D) modeling of all
Vardhan, K.; Kalani, R.
Computational Fluid Dynamics
sequences of events in an accident. In this study, CFD is used to model VCE in two industrial
(CFD) Modeling and Analysis of accidents: the Amuay refinery disaster (happened in 2012) and the Indian Oil Corporation’s (IOC)
Hydrocarbon Vapor Cloud Jaipur terminal (2009). This work studies 3D CFD modeling of flammable cloud explosion in the
Explosions (VCEs) in Amuay real-time configuration for both accidents. FLACS (FLame ACceleration Simulator), a CFD software,
Refinery and Jaipur Plant Using is used to simulate the loss of hydrocarbon containment, cloud formation, and explosion in both
FLACS. Processes 2021, 9, 960. industrial case studies. The ignition locations and grid sizes were varied to analyze their influence on
https://doi.org/10.3390/pr9060960 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,
Academic Editor: Chi-Min Shu
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
Received: 14 April 2021
and quantitative risk assessment of the Amuay refinery accident and Jaipur fire. The simulation of
Accepted: 26 May 2021
Published: 28 May 2021
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-
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
sion (VCE) simulation; Amuay refinery; Jaipur fire
published maps and institutional affil-
iations.
1. Introduction
Fuels are a significant energy source for power generation and transportation. In
Copyright: © 2021 by the authors. general, fuels such as petrol, natural gas, diesel, and petroleum gas have high demand.
Licensee MDPI, Basel, Switzerland. The need for energy is pushing every country to develop refinement facilities for crude
This article is an open access article oil processing and refining. The proliferation of processing industries has increased the
distributed under the terms and potential hazards to the safety of people and process facilities [1]. Today, process safety
conditions of the Creative Commons intermingles in the design, production, and maintenance of every processing facility. How-
Attribution (CC BY) license (https:// ever, despite all safety guidelines and measures, accidents are happening. The magnitude
creativecommons.org/licenses/by/ of these accidents ranges from minor damage to equipment or personnel to catastrophic
4.0/).
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.
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.
Yes Fire
No Release to
atmosphere
No
No Fire
Yes
Yes VCE
Figure 2. An event tree of VCE.
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
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
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.
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.
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.
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
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]