Journal of Global Scientific Research 7 (2) 2022/ 2094-2103

Contents lists available at www.gsjpublications.com

Journal of Global Scientific Research

in Chemical Engineering
journal homepage: www.gsjpublications.com/jgsr

Risk Assessment of Ammoina Storage Tank facility in a Fertilizer

Production Plant Based on Bayesian Approach: A Case Study
Mohammed Faraj Saeid1,2*, Ismail Hassan Abdilahi3, Azizan Ramli4
1Civil
Aviation Higher Institute (CAHI), Tobruk, Libya.
2AidStation, Heavy Oil Depot and Oil Berth, Brega Petroleum Marketing Company, Tobruk, Libya.
3Faculty of Chemical and Process Engineering Technology, College of Engineering Technology, Universiti Malaysia Pahang, Lebuhraya

Tun Abdul Razak, 26300 Gambang, Kuantan, Pahang, Malaysia.

4Faculty of Industrial Sciences and Technology, College of Computing and Applied Sciences, Universiti Malaysia Pahang, Lebuhraya Tun

Abdul Razak, 26300 Gambang, Kuantan, Pahang, Malaysia.

ARTICLE INFO ABSTRACT

Received: 4 Jan 2022, In this paper, Bayesian Network methodology based on interview questionnaire with
Revised: 7 Jan 2022, two experts was applied on an ammonia storage unit in a fertilizers production plant.
Accepted: 25 Jan 2022, Ammonia is stored in cold storage tank as liquefied gas at atmospheric pressure. This
Online: 21 Feb 2022 tank is susceptible to failures due to various abnormal conditions arising due process
failures. Root Cause Analysis (RCA) was used to identify all the failure modes, which
could result in the occurrence of the undesirable incident. Subsequently, the storage
Keywords: unit was found to belong to the “highly hazardous category” based on consequences
Ammonia tank failure, Bayesian analysis technique. Evaluation of probability of occurrence was applied based on
Network, Root Cause Analysis, conditional probability table (CPT) to obtain the occurrence probabilities of the
Consequence Analysis, Failure potential failures and the respective consequences. Subsequently, BN prediction
Prediction and Diagnosis analysis was used is useful tool to predict the future events and it useful tool in the
decision making, in addition BN diagnosis analysis was performed to determine the
posterior probabilities of basic events which can be used to find the system
vulnerabilities. The results showed that corrosion, overfilling, and faulty pressure
indicator are the most likely to cause the top event. Nevertheless, these situations
could be averted if the operators are trained and are sensitized to practice relatively
simple safety measures.

1. Introduction

Chemical process industries (CPIs) are very prone properly managed could escalate small mishaps
to catastrophic accidents as they contain large into catastrophic events [2]. It is very important to
quantities of highly flammable and explosive understand the hazards and risks associated with
chemicals. The operating conditions of intense, the the process; to carry out a risk assessment to
complex interactions between the system identify them and to take appropriate action to
components [1] and human and organizational eliminate or minimize hazards and risks;
failures issues which make CPI vulnerable to otherwise, a catastrophic accident may result. For
process deviations and failures, that if not example, process installations involve a large
___________
*Corresponding author:
E-mail addresses: Mohammed Faraj Saeid faraj.saeid@yahoo.com
2523-9376/© 2022 Global Scientific Journals - MZM Resources. All rights reserved.
Saeid, M. F. et al. Journal of Global Scientific Research (ISSN: 2523-9376) 7 (2) 2022 2095

number of pumps, compressors, separators, (FTA) , failure modes and effects analysis (FMEA),
complex pipe systems, storage tanks, etc. in the have been criticized for presenting accurate
congested area. A small mistake by the operator or results due to disadvantages such as being static
a problem in the process system could escalate (unable to update the probability of events), using
into such a disastrous event, as the process area is generic data failure (unable to handle data
congested with process equipment and pipe uncertainties) and being unable to consider
systems and has limited ventilation and escape complex dependences between system
routes. Case histories have shown that components [9]. Bayesian network is a scheme
catastrophic accidents have had a significant that shapes the relationship of probability
impact on people, the environment and society as between events. In this scheme, the nodes
they have resulted in fatalities and large financial represent the events, and the lines that link them
losses. For example, a vapor cloud explosion in the represent the causal relationship. Each node has
BP Texas City refinery in 2005 resulted in 15 its own probability table, which lists the frequency
fatalities, 180 injuries and $1.5 billion in losses of the occurrence of an event (it can be discrete or
[3]. The investigation fund that lack of risk continuous). Therefore, the BN consists of a graph
reduction measures and insufficient process associated with the probability tables. The
safety contributed to this catastrophic accident. evolution of a factor, such as the fact that one of
The U.S. CSB investigation on explosion of natural the nodes passes from an indecisive state to a
gas at the ConAgra foods processing facility in certain state, causes the evolution of all the factors
North Carolina in 2009 and the Kleen Energy linked to it. Consequently, the network
Power Plant Connecticut in 2020, reported failure recalculates the probabilities as the amount of
to follow inherently safer approach from fire and knowledge available increases [10]. Recently, in a
explosion hazard contributed to explosions [4]. In comparative study on FTA and BN in process
2010, 11 deaths and 17 injuries resulted from a safety analysis, explained the various modeling
fire and explosion caused due to blowout at the features of BN that help to incorporate multi-state
Macondo well [5]. Also, the environment and variables, dependent failures, functional
wildfire surrounding the Gulf of Mexico had uncertainties and expert opinions that are
catastrophic effects resulting from the continuous frequently encountered in safety analysis but
spill from the wellhead for 87 days. cannot be considered FTA [4]. A risk-based
accident model based on BN has been developed
These incidents have profoundly changed the for analysis of leakage failure of submarine oil and
perception of people and have contributed greatly gas pipelines [11]; the findings of their
to rising concern about emphasizing the safety of investigation have shown that BN can provide a
processes. more case-specific and realistic analysis of the
consequences compared to the bow-tie (BT)
Risk assessment is a methodology used to identify method, since it could consider common cause
possible events or future accidents that may failures and conditional dependence in the
occur, determine how likely they may occur, and accident evolution process. QRA conventional
evaluate the potential consequences. As a result, methods such as BT, FTA, and event tree analysis
the hazard can be assessed qualitatively and (ETA) are strictly one way, a reverse inference or
quantitatively for a specific failure scenario [6]. a two-way analysis may also be provided by BN,
The reliable risk analysis methodology plays a key not only to find the results of the causes, but also
role in the development of prevention plans and in to find the causes of the results [12].
the control of failures during CPI process
operations. The qualitative and quantitative A quantitative risk assessment has been
methods are two key aspects of risk [7], widely performed for an ammonia storage facility. The
applied in chemical industries. In spite of their storage facility was found to belong to the “highly
popularity as more adaptable and cost-effective hazardous category” based on qualitative Hazard
decision-making tools than qualitative methods Analysis (HAZAN) technique, which involved a
[8], Quantitative risk assessment (QRA) detailed Fire, Explosion and Toxicity Index studies
conventional methods (e.g. fault tree analysis (FETI) and Hazard and Operability (HAZOP)
Saeid, M. F. et al. Journal of Global Scientific Research (ISSN: 2523-9376) 7 (2) 2022 2096

analysis. Subsequently, Fault Tree Analysis (FTA) their consequences using expert
was used to identify all the failure modes [13] interview questionnaire.
2. Evaluation of Probability of Occurrence:
A methodology of Bayesian Networks (BNs) based For the estimation of occurrence
dynamic failure assessment (DFA) was applied on probability of an event, it is proposed to
an ammonia storage unit in a specialized chemical use the Bayesian Networks technique.
industry. The study shows the use of DFA tools in This technique was chosen because of
conducting an effective failure assessment study its advantages against traditional ones,
and the method showed that it can be detected the like fault tree and event tree analysis.
vulnerable areas which help to prevent escalation Developing Bayesian Network model
of an abnormal event to a major accident [8]. will be into two parts, the first one is to
develop direct acyclic graph (qualitative
In process industries, a lack of formal data in part) by using Genie (Graphical network
conducting risk analysis is quite a hindrance and interface) software package. It is useful
has proved to be the biggest blight in conducting a for decision analysis and for graphically
proper safety assessment. Safety analysts need to representing the union of probability
use a large amount of subjective information in and networked occurrences.
their studies. The preferred tool must therefore be Particularly, Genie can be used for the
able to use this subjective information formally in analysis of Bayesian networks, or
order to achieve objective results [8]. directed acyclic graphs; the second part
is to create the probability of the
The aim of the present study is to analysis the variables or factors by using conditional
causes and consequences of ammonia tank probability table (quantitative part).
leakage in a fertilizers production plant based on 3. Bayesian networks analysis (prediction
Bayesian networks (BNs) methodology for the and diagnosis): In this step, the risk was
risk assessment. A BNs is a simple method that estimated and assessed by the network
gives the qualitative and quantitative analysis of with considering the failure
the factors as well as by using BNs method it can probabilities and the relevant
be simply to present the relationship between the consequences. Only two different states
factors and effects. of true and false were considered for the
risk node.
2. Research Methodology
Case Study: Ammonia Storage Tank
Ammonia is widely used as a precursor to other
agricultural and industrial processes products Ammonia surplus is stored in a large capacity
throughout the world. Approximately 80% of the tank in Fertilizers Industries. Special processes
ammonia produced by the industry is used in must be developed for the efficient discharge
agriculture as fertilizer and therefore, due to its and storage of hazardous chemicals, such as
high risk, the safety of ammonia storage must be ammonia, and several related parameters must
ensured, and preventive measures must be taken be checked and evaluated. In ammonia plants,
to prevent any explosion or fire. However, ammonia is produced from hydrogen and
ammonia production plants are still suffering in nitrogen-containing synthesis gas at a ratio of
this area due to a lack of uncertain and limited approximately 3:1. In addition to these
data [14]. The proposed approach is based on components, the synthetic gas also contains
quantitative and qualitative risk assessment limited inert gases, such as argon and methane.
practice. This approach goes through the The H2 source is demineralized water (steam)
following seven main steps: and hydrocarbons in natural gas. Atmospheric
air is the source of N2. The source of CO2 is the
1. Analysing the root causes that lead to hydrocarbons in the feed of natural gas [13].
occurrence the accidents scenarios and
Saeid, M. F. et al. Journal of Global Scientific Research (ISSN: 2523-9376) 7 (2) 2022 2097

Figure 1. Schematic sketch of ammonia storage facility. V: vent line; PI: pressure indicator; LI: level
indicator; TIC: temperature indicator (adapted from 13)

A schematic of a representative storage tank is happened, why it happened, and what can be done
shown in Fig. 1. The tank is 12 m long and 2.9 m in to prevent it from happening again. The analysis
diameter. The chemical is stored at ambient was implemented based on an interview by using
temperature and at a pressure of 13 kg/cm2 , questionnaire with two experts from BASF
under these conditions, it exists as a pressurized PETRONAS Chemicals Sdn. Bhd and Central
liquid. The tank is made of carbon steel and is Emergency & Fire Services Pengerang Integrated
supported by a saddle (3 no) at a height of 1 m. Complex (CEFS PIC) in Malaysia. Expert opinion is
The ground underneath is composed of small one of the most prevalent approaches to address
stones and soil, all around the storage section is possibilities of safety risks. Table 1 presents the
provided with a concrete path. The tank is root causes of ammonia leakage, all possible
equipped with a bayonet heater in which hot routes for the occurrence of the probable scenario,
water at 70-110 °C is used as a heating medium. commonly referred to as the top event. The factors
There is also a cooling system to prevent of ammonia storage tank leakage assessed under
overheating, which may lead to a build-up of material and operational failures, both of these
pressure. Ammonia is transferred through factors have a root cause. Material failure of the
insulated pipelines to the nearby sections [13]. tank is due to corrosion, wrong design and fatigue.
Overpressure and implosion can cause
3. Results and Discussion operational failure due to some factors such as
overfilling, cooling system failure and faulty
3.1 Root Cause Analysis (RCA) pressure indicator that lead to overpressure in the
system.
Root cause analysis (RCA) is a key part of risk
assessment and it is useful tool to identify what
Saeid, M. F. et al. Journal of Global Scientific Research (ISSN: 2523-9376) 7 (2) 2022 2098

Table 1: Summary of the basic causes of ammonia leakage

Root Causes Basic cause of the hazard
Corrosion Corrosion caused by combination of oxygen, ammonia, stress and carbon steel
Overpressure Overpressure will happen if
a) Warm ammonia was injected in the tank.
b) Sudden mixing of ammonia solution and liquid ammonia when an oil layer
between these phases was broken up
Overfilling Error of level readings by operators in combination with failure of high level alarm,
then overfilling may occur
Fatigue Fatigue may occur because of the long lifetime of an ammonia storage tank
Faulty design Human error can cause faulty design
Implosion Vacuum causing tank implosion or collapse: the failure of the pressure transmitters
and a failing vacuum relief valve will cause the roof of the tank to partially collapse
Faulty pressure indicator Lack of maintenance
Cooling system failure cooling water blockage causes the cooling system failure

3.2 Consequences analysis The flammable limits of ammonia are 16-25% by

volume in air with an ignition temperature of
The objective of consequence analysis is to 651ºC. Ignition of such mixtures is improbable
quantify the harmful impacts in case of occurrence under ordinary conditions, it occurs in a confined
of the potential top event. Toxicity is the major space, it could result in an explosion. The
hazard associated with the accidental release of probability of this event is low and therefore
this chemical. Ammonia stored at pressure ammonia installations are not regarded as
possess considerable potential energy and should significant fire hazards.
a rupture of the primary container occur due to
corrosion or any other factor, the liquid will flash 3.3 Evaluation of Probability of Occurrence
into a vapour spontaneously as the
thermodynamic state of the ammonia adjusts itself At this step, a Bayesian Network representing the
to the diminished pressure. Since this adiabatic system was built in order to obtain the
flash evaporation takes place almost probabilities to occur leakage in the storage tank
instantaneously, most of the contents of the of ammonia. This network considers 2 different
ruptured pressure vessel will enter the possible ways of leakage of ammonia (Material
atmosphere either as a vapor or as a fine aerosol. and Operational failures) as shown in Fig. 2.
Saeid, M. F. et al. Journal of Global Scientific Research (ISSN: 2523-9376) 7 (2) 2022 2099

Figure 2. Direct acyclic graph of the BN

3.4 Conditional probability table (CPT) probability of occurrence of each failure scenario
with the respective consequence, and adding all of
Conditional probability table is the quantitative the risks. The average values of CPT were applied
part of the Bayesian network. The CPT values are to the GeNIe software package to release the final
extracted from the expert’s experience in the results as shown in the Fig. 3.
current study field by using detailed
questionnaire. Table 2 (a,b,c,d,e,f,g) present the

Figure 3. Bayesian network at normal conditions

Saeid, M. F. et al. Journal of Global Scientific Research (ISSN: 2523-9376) 7 (2) 2022 2100

Table 2: (a) Conditional probability of material failure

Corrosion True False
Fatigue True False True False
Faulty design True False True False True False True False
Material True 0.9 0.7 0.65 0.6 0.7 0.55 0.45 0.1
failure False 0.1 0.3 0.35 0.4 0.3 0.45 0.55 0.9

Table 2: (b) Conditional probability of overpressure

Overfilling True False
Faulty pressure True False True False
indicator
Cooling system True False True False True False True False
failure
Overpressure True 0.9 0.75 0.6 0.55 0.6 0.6 0.5 0.3
False 0.1 0.25 0.4 0.45 0.4 0.4 0.5 0.7

Table 2: (c) Conditional probability of ammonia leakage

Operational failure True False
Material failure True False True False
Ammonia True 0.8 0.6 0.7 0.1
leakage False 0.2 0.4 0.3 0.9

Table 2: (d) Conditional probability of operational failure

Implosion True False
Overpressure True False True False
Operational True 0.9 0.7 0.7 0.2
failure False 0.1 0.3 0.3 0.8

Table 2: (e) Conditional probability of toxic gas release

Ammonia leakage True False
True 0.85 0.1
Toxic gas release False 0.15 0.9

Table 2: (f) Conditional probability of fire

Ammonia leakage True False
Fire True 0.6 0.1
False 0.4 0.9

Table 2: (g) Conditional probability of explosion

Ammonia leakage True False
Explosion True 0.55 0.2
False 0.45 0.8

Table 3: Definitions of CPT values

0.1 highly unlikely to happen
0.3 Unlikely to happen
0.5 Equally likely to happen or not to happen
0.7 Likely to happen
0.9 Highly likely to happen
0.2, 0.4, 0.6, 0.8 Intermediate values (e.g the value of 0.2 is between highly
unlikely and unlikely to happen)
Saeid, M. F. et al. Journal of Global Scientific Research (ISSN: 2523-9376) 7 (2) 2022 2101

3.5 Bayesian Network analysis (prediction and It is assumed that the conditional probability of
diagnosis) corrosion, implosion and overfilling are 100%
(failure has occurred) which is led to increase in
BN prediction analysis the probability distribution of some factors such
as material failure probability raised up to 71%,
BN predictive analysis is using forward prediction operational failure probability increased to 84%
of the Bayesian network. The analysis was and overpressure probability raised up to 71%. as
performed by changing some of conditional a result of this assumption, increasing in the
probability values with 100% to predict the probability of toxic gas release, fire and explosion
probability distribution of failure factors before an consequences up to 64%, 46% and 45%
accident occur. BN prediction analysis is useful respectively, as indicated in the Fig. 4.
tool to predict the future events in the decision
making.

Figure 4. BN prediction analysis

BN diagnosis analysis 53% of corrosion, fatigue, and fault design,

respectively. While the factors of operational
BN diagnosis analysis is a backward diagnosis of failure were 69% of overpressure and 55% of
the Bayesian network to determine the posterior implosion in the storage tank, Furthermore, the
probabilities of basic events which can be used to probability Causes of overpressure were 61% of
find the system vulnerabilities (weakness). The overfilling, 61% of faulty pressure indicator and
probability of ammonia leakage assumed to be 41% of cooling system failure. Likewise, the
100% (ammonia leakage occurred). It can be probability of the consequence occurrence
observed that when the probability of operational increased as 85% of toxic gas release, 60% of fire
and material failures be 81% and 78% could and 55% of explosion. BN diagnosis analysis is
result in the occurrence of ammonia leakage shown in Fig. 5.
accident scenario. In addition, the probability
causes of material failure were 74%, 54%, and
Saeid, M. F. et al. Journal of Global Scientific Research (ISSN: 2523-9376) 7 (2) 2022 2102

Figure 5. BN diagnosis analysis

The results show that corrosion, overfilling, and faulty pressure indicator are the most likely to cause the
ammonia leakage accident scenario.

Table 3. shows a comparison of the probability of the factors, consequence and ammonia leakage scenario
occurrence during the normal conditions of the ammonia storage tank with that ones in prediction and
diagnosis analysis.

Table 3: Comparison of the likelihood failure factors

Factors Normal BN Prediction BN Diagnosis BN

Ammonia leakage 62% 72% 100%
Material failure 63% 71% 78%

Operational failure 67% 84% 81%

Overpressure 63% 84% 69%

Implosion 50% 100% 55%

Corrosion 70% 100% 74%

Fatigue of the tank 50% 50% 54%

Fault design 50% 50% 53%

Overfilling 60% 100% 61%

Fault pressure indicator 60% 60% 61%

Cooling system failure 40% 40% 41%

Consequences of the failures

Toxic gas release 57% 64% 85%

Fire 41% 46% 60%

Explosion 42% 45% 55%

Saeid, M. F. et al. Journal of Global Scientific Research (ISSN: 2523-9376) 7 (2) 2022 2103

4. Conclusion Bayesian network approaches. Reliability

Engineering and System Safety, 96, 925-932 (2011).
[5] US National Commission on BP Deepwater Horizon
A qualitive and quantitative risk assessment were Oil Spill and Offshore Drilling. Deepwater: The gulf oil
performed for an ammonia storage tank using disaster and the future of offshore drilling (2011).
Bayesian network approach. Based on the root Available at
cause analysis, the storage section was found to http://www.oilspillcommission.gov/final-report/
Accessed on 5 January 2021.
belong to “high” hazard category. Since ammonia
[6] Abbassia, R. Bhandaria, J. Khana,b, F. Garaniyaa, V.&
is stored under pressurized conditions, accidental Chai, Sh. Developing a Quantitative Risk-based
rupture due to corrosion would lead to Methodology for Maintenance Scheduling Using
instantaneous dispersion of ammonia into the Bayesian Network. Chemical Engineering
surrounding atmosphere. The release of ammonia Transactions. 48.6. (2016).
[7] Han Z., Weng, W., Comparison study on qualitative
into the atmosphere resulting from rupture of the and quantitative risk assessment methods for urban
storage vessel was identified as one the top or natural gas pipeline network. J. Hazard. Mater. 189,
unwanted event. Overfilling and faulty pressure 509 518 (2011).
indicator failures are also considered as most https://doi.org/10.1016/j.jhazmat.2011.02.067.
[8] Roy, A., Srivastava, P., Sinha, S., Risk and reliability
likely to cause ammonia release into the
assessment in chemical process industries using
atmosphere. Bayesian methods. Rev. Chem. Eng. 30, 479-499
(2014). https://doi.org/10.1515/revce- 2013-0043.
5. Acknowledgements [9] Zarei, E., Azadeh, A., Aliabadi, M.M., Mohammadfam,
I., Dynamic safety risk modeling of process systems
The authors would like to acknowledge the using bayesian network. Process Saf. Prog. 36, 399-
407 (2017). https://doi.org/10.1002/prs.11889.
Faculty of Chemical and Process Engineering [10] Zerouali, B., Hamaidi, B. Predictive analysis for risk of
Technology, University Malaysia Pahang for fire and explosion of LNG storage tanks by fuzzy
providing the needed facilities in finishing the Bayesian network. Life Cycle Reliab Saf Eng 9, 319–
present work. 328 (2020). https://doi.org/10.1007/s41872-019-
00105-z.
[11] Li, X., Chen, G., Zhu, H., 2016. Quantitative risk
6. References analysis on leakage failure of submarine oil and gas
pipelines using Bayesian network. Process Saf.
[1] Roy, A., Srivastava, P., Sinha, S., Dynamic failure Environ. Prot. 103,163-173 (2016).
assessment of an ammonia storage unit: A case study. https://doi.org/10.1016/j.psep.2016.06.006.
Process Saf. Environ. Prot. 94, 385-401(2015). [12] Shan, X., Liu, K., Sun, P. L., Risk analysis on leakage
https://doi.org/10.1016/j.psep.2014.09.004. failure of natural gas pipelines by fuzzy Bayesian
[2] Al-Shanini, A., Ahmad, A., Khan, F., Accident modelling network with a bow-tie model. Sci. Program. 1-11
and analysis in process industries. J. Loss Prev. Proc. (2017). https://doi.org/10.1155/2017/3639524.
32, 319-334 (2014). [13] Roy, P., A. Bhatt, B. Kumar, Sarvjeet Kaur and C.
https://doi.org/10.1016/j.jlp.2014.09.016. Rajagopal. Consequence and risk assessment: Case
[3] U.S. Chemical Safety and Hazard Investigation Board study of an ammonia storage facility. ARCH.
(CSB). Investigation Report: refinery explosion and ENVIRON. SCI. (2011), 5, 25-36.
fire, Bp Texas City, Texas, March 23, 2005, Report No. [14] Prem Baboo, Ammonia Storage Tank Pre -
2005-04-I-TX. (2007). Commissioning. International Journal of Engineering
[4] Khakzad, N., Khan, F. and Amyotte,P. Safety analysis Research & Technology (IJERT). Vol. 8 Issue 09.
in process facilities: Comparison of fault tree and ISSN: 2278-018. (2019).