Journal of Loss Prevention in the Process Industries 50 (2017) 397–402

Contents lists available at ScienceDirect

Journal of Loss Prevention in the Process Industries

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

Lessons learned from analyzing a VCE accident at a chemical plant T

a,∗ b b b
Changlong Zhu , Jiayong Zhu , Li Wang , M. Sam Mannan
a
College of Safety Science and Engineering, Nanjing Tech University, Nanjing 210009, China
b
Mary Kay O'Connor Process Safety Center, Artie McFerrin Department of Chemical Engineering, Texas A&M University, College Station, TX 77843-3122, USA

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

Keywords: On June 5, 2017, there was a Vapor Cloud Explosion (VCE) at JinYu Petrochemical Co. Ltd. During unloading of
Vapor cloud explosion liquified natural gas, the gas leaked from the truck and formed a vapor cloud, then the vapor cloud met with the
Process safety management electrical sparks in the plant laboratory, and an explosion occurred. This incident caused significant casualties
Computational fluid dynamic and property loss. The main causes of the accident are demonstrated by a fishbone diagram. The major causes
Event sequence diagram
include (a) defective design, (b) noncompliance with standard operation procedures (SOPs), (c) the presence of
Monte Carlo methodology
ignition sources, (d) ineffective enforcement of safety rules, (e) inadequate design of emergency facilities, and (f)
delayed emergency response. The analysis reveals that process safety management (PSM) is a crucial factor in
the success of chemical plants, especially for small and median-scale companies in developing countries.
Dispersion phenomenon of liquefied gas is simulated by commercial Computational Fluid Dynamic (CFD)
software. The simulation results approximately agree with the real incident. The simulation indicates that a
visual and quantitative consequence analysis can provide general guidance for PSM. Finally, based on various
PSM conditions (good, normal, weak, and bad), the probabilities of VCE generated from leakage are obtained by
an event sequence diagram (ESD) and Monte Carlo methodology. By comparing the quantitative probability
values in four different PSM situations, the results show that emergency management associated with effective
PSM is crucial to avoid VCE incidents.

1. Introduction effect have been conducted (Cozzani et al., 2014; He et al., 2017;
Hemmatian et al., 2014; Vílchez et al., 2011; Villa et al., 2016). Because
A large number of catastrophic accidents have occurred in chemical of these studies, process safety management (PSM) is now recognized as
process industries in the past decades. The most common examples of essential for chemical plants. However, a few companies still lack of an
these accidents are Vapor Cloud Explosions (VCE). There were about awareness of process safety, especially in some developing countries
174 vapor cloud explosion (VCE) accidents worldwide that were re- (Khan et al., 2015; Knegtering and Pasman, 2009; Zhao et al., 2014;
ported between 1940 and 2010 (Zhu et al., 2015). For example, the Atkinson et al., 2015; Sharma et al., 2013; Li et al., 2014; Zhang and
liquid petroleum gas explosion in Mexico in 1984 and the explosion at Zheng, 2012). The objective of this paper is to analyze the potential
British Petroleum's Texas City refinery in 2005 were noteworthy in- causes of an incident occurred in Shangdong, China on 2017, as well as
cidents. These incidents revealed that effective safety management of to reveal the importance of PSM implementation and quantitative risk
hydrocarbons and reactive chemicals (e.g., hydrogen, methane, acet- assessment.
ylene, ethylene, and liquid petroleum gas) could substantially reduce
the risks in chemical plants. Additionally, past cases showed that VCE 2. Description of a VCE incident
accidents could occur in any part of the operating processes for ha-
zardous chemicals including preparation, utilization, storage, trans- On June 5, 2017, liquefied petroleum gas was released during the
portation and disposition. Defective design of facilities, imperfect unloading process at JinYu chemical plant in China's Shandong
equipment maintenance, human error in operations and lack of safety Province (Work Safety Committee of the State Council in China, 2017).
rule enforcement by management are important factors that can cause Since liquefied petroleum gas is a heavy gas, after it leaked, it settled
accidents. Much research has been done on the causes of VCE accidents near the ground. When the gas cloud encountered an ignition source,
(Chang and Lin, 2006; Konstandinidou et al., 2011). Furthermore, the explosion occurred. Ten people died, and 9 were injured in the
qualitative risk analysis, consequence analysis and study of the domino incident. The plant was damaged. Facilities, including 15 carrier

∗
Corresponding author.
E-mail address: [email protected] (C. Zhu).

https://doi.org/10.1016/j.jlp.2017.11.004
Received 22 October 2017; Received in revised form 6 November 2017; Accepted 12 November 2017
Available online 14 November 2017
0950-4230/ © 2017 Elsevier Ltd. All rights reserved.
C. Zhu et al. Journal of Loss Prevention in the Process Industries 50 (2017) 397–402

Fig. 1. Plant layout drawing.

vehicles of hazardous chemicals, 1 spherical tank, 2 vault tanks, pro- operating procedures and without supervision, the driver failed to
duction equipment, laboratory, control room, office buildings, sur- connect them properly. The worker on duty carelessly checked the spot
rounding enterprises and social vehicles were also damaged. In addi- and did not find out any problem. A leakage of liquefied gas began.
tion, 6 spherical tanks caught fire, and several pipeline systems After a period of 130 s following the release, a cloud of liquefied gas
collapsed. was ignited. A vapor cloud explosion occurred due to an ignition source
in the laboratory to the north. As a result, serious damage occurred
through a domino effect. Table 1 summarizes the timeline of this in-
2.1. Layout of the chemical plant and the leak area
cident.
The 86,955-square-meter plant had nine 200 m3 liquefied gas tanks,
which stored propane, isobutane and pentane oil, six 1000 m3 and six
3. The fishbone diagram of the accident
2000 m3 liquefied petroleum gas spherical tanks, eighteen 150 m3
pentane oil tanks, six 3000 m3 liquefied petroleum spherical tanks, six
The inadequate piping system was one of the reasons for this VCE
2000 m3 isooctane tanks and four 5000 m3 sulfuric acid storage tanks.
accident. The piping system was very complex. It was composed of
On the east side of these tanks, was the loading and unloading area for
several rotary joints, varying sizes of multi-section pipes, spherical
the tanker trucks. Fifty meters from the north side of the loading/un-
valves, flanges, quick connectors, and an anti-static device. Thus, any
loading area was the laboratory. The control room and the factory of-
small defect in a small component of the piping system could cause
fice building were located at the east side of the loading/unloading. The
failure to the loading operation. In addition, there were also several
layout of this plant is demonstrated in Fig. 1.
potential factors related to process safety: suitable supervision before
In the early morning of June 5, a tanker truck arrived at the un-
the loading operation, adequate connection procedures, effective
loading zone. The driver got out of the truck and tried to connect the
emergency management, frequent safety training for workers, etc.
omnidirectional loading arm to the tanker's discharge outlet. However,
Considering PSM and all the factors mentioned in this paper, the fish-
he failed to connect the arm to the outlet properly. Suddenly, large
bone diagram for this accident is shown in Fig. 3.
amount of liquefied gases began to leak. And the gases quickly spread
Through analysis of the fishbone diagram, we conclude that there
over the unloading zone. The location of leakage is shown in Fig. 2.
were four major reasons for this severe incident. The first factor was a
failure by workers to follow operation procedures and unloading pro-
2.2. The time sequence of the accident cedures. The second factor was a lack of equipment integrity manage-
ment and safety awareness. For example, there was no reliable leak
More specifically, at about 1:00 a.m. on June 5, 2017, the driver detection procedure or alarm instrument during unloading procedures
began to unload liquefied gas by connecting the omnidirectional so there was no immediate response. The third factor was a failure in
loading arm to the tanker's discharge outlet. Because of the complicated conducting consequence measurement of unexpected scenarios and

Fig. 2. The leakage location of the loading/unloading zone.

398
C. Zhu et al. Journal of Loss Prevention in the Process Industries 50 (2017) 397–402

Table 1 diameter and 0.2 m thick. The tank diameters on the truck were about
Timeline of this accident. 3 m. Two tanker trucks were 5 m apart. The leakage was in the loading
arm joint and at an angle of 0°along the horizontal direction. The
Time The accident development
leakage diameter was 30 mm. The following assumptions are used in
00:00:00 The operations at the loading/unloading zone were busy. the simulation. The wind speed was assumed to be very low (∼1 m/s)
More than 10 tankers entered the site almost at the same during leakage, and the atmospheric stability class was D. The simpli-
time.
fied model was established and simulated using commercial CFD soft-
00:57:20 A tanker truck arrived at the 11th unloading station.
00:57:30 The driver got out of the tanker. ware. Fig. 5 shows the simulation of liquefied gas dispersion based on
00:57:40 The driver performed the discharge operation. the above conditions. The simulation results were in approximate
00:58:00 The worker on duty began to check the spot. agreement with the real incident. Thus, the CFD simulation results
00:58:40 The unloading was proceeding. could provide a good guideline for prediction and consequences ana-
00:59:10 The connection between the loading arm and the tanker's
lysis of gas leakage.
discharge outlet broke away and the leakage occurred.
00:59:10–01:01:20 No effective emergency response was executed. After 130s, the gas concentration on the ground was almost 10% at
01:01:20 Flare and blast burst out at the north side of the loading/ the north laboratory, which was 50 m from the tanker trucks. The
unloading area. concentration was in the range of the liquefied gas explosion limit scale
after 01:01:20 Death, injuries, damages, and domino effects occurred.
(1.5～10%). Because there was no explosion-proof switch in the la-
Until 16:00 Fire was extinguished.
boratory, sufficient electrical spark ignition energy was generated and a
detonation scenario was inevitable.
PSM on the operating processes. Therefore, no contingency plan and By comparing the simulation results with the real incident, it is
emergency response were implemented when the liquefied gas leakage revealed that the consequence risk analysis using CFD modeling can be
occurred. The last was the lack of an explosion-proof plant design. The used for making important recommendations for accident prevention
facilities could have prevented or reduced the consequences of the in- and emergency response. Thus, it is necessary to encourage managers to
cident if an explosion-proof electrical switch had been installed at the implement quantitative risk analysis as well as PSM.
laboratory.
4.2. Comparative analysis on probability from leakage to VCE
4. Comparison of analysis results on accident consequences
When the flammable liquefied gas leaked out, a shut off valve
4.1. Comparison of on-site leakage and dispersion with CFD software should have been activated. If the response did not work, the effective
simulation measures of plugging should have been taken. If these responses did not
work, a mist water spray system should have been activated to reduce
During the leakage at the piping connection, the video surveillance the concentration of hazardous gas. If this had happened, the diluted
at the plant recorded the event. The videos of leakage taken at different gas would not have been ignited by the static charge. The level (good,
times are shown in Fig. 4. At the beginning of the leakage, the liquefied normal, weak and bad) of PSM directly determined the probability of
gas was clustered between two tankers. After 2s, the gas cloud spread VCE. The probability values of VCE calculated here is more specific
distinctly from the bottom of the two tankers towards the east and west than the data provided in other studies (Moosemiller, 2011; Zhu et al.,
sides. After 6s, the dispersion traveled distinctly towards the north and 2012).
south areas. After 10s, the gas cloud spread out and covered the sur- Four PSM cases were listed in Table 2. The first case is with a good
rounding areas. After 130s, the explosion occurred at the north side of PSM. The explosion occurred after 130s of leakage. If the laboratory
the tankers. had been designed to prevent the buildup of a static charge and other
As shown in the video recording, the tanker trucks were 4 m in ignition sources, the vapor cloud would have spread out and become
height and 8 m in length. The front and rear wheels were all 1 m in less concentrated, though it could have been ignited by other ignition

Fig. 3. Fishbone diagram of the accident.

399
C. Zhu et al. Journal of Loss Prevention in the Process Industries 50 (2017) 397–402

Fig. 4. The video screenshots of the leakage scenario.

Fig. 5. The simulation of diffusion process of liquefied gas.

sources outside of the chemical plant. In this first case, the allowable dynamic event sequence diagram is used to establish the series process
maximum time to react before the explosion occurred (Tmax) is as- (Zhou et al., 2016; Luo and Hu, 2013), as shown in Fig. 6. The prob-
sumed to be twice as long as the above mentioned incident (130 s). abilities under four cases are calculated by the Monte Carlo method.
Tmax is set to 260 s. As shown in Fig. 6, the event for the leakage alert is 1,000,000 simulations are performed to reduce sample bias and error.
denoted by de1. Then, the leakage is detected by a device or an op- The probabilities of occurrences from leakage to VCE are listed in
erator. The event of making the decision to control the leakage is de- Table 3.
noted by de2. The event of activating the mitigation system (e.g., water The results show that a good PSM system could significantly reduce
curtain) to reduce the hazardous material concentration is denoted by the probability of VCE by more than 80%. Therefore, a series of
de3. The event of the teams to complete the plugging is denoted by de4. emergency responses are important for preventing incidents. The
The time required for event de1, de2, de3, de4 is designated by variables emergency management of every plant should be carefully considered.
t1, t2, t3, t4, respectively. These variables are assumed to be a normal Deficiencies in emergency management should be revised and im-
distribution. The second case is with regular PSM. Tmax is 260s, and the proved. Emergency responses such as emergency response plans,
time for de1, de2, de3, de4 is set 20% longer than the first case. The emergency infrastructure and emergency procedure training should be
third case is set to be with weak PSM. Tmax is 260s, and the decision- considered in chemical processing plants. Additionally, it is paramount
making time for de1, de2, de3, de4 is all 50% longer than the first case. to establish an effective safety culture in the organization.
The fourth case is based on bad PSM. Because of inadequate compiance
audits, Tmax is 130 s (same as the above mentioned incident) and de-
cision-making time de1, de2, de3, de4 is all 100% longer than the first 5. Lessons from the causes and consequences of the incident
case. The mean and variance of de1, de2, de3, de4 about the above cases
are shown in Table 2. From the perspective of process safety management, the lessons
In order to analyze the above cases in a consistent manner, a learned in this case should include at least the following aspects:

Table 2
Parameters of normal distributions for emergency action durations.

Events case1(Tmax = 260s) case2(Tmax = 260s) case3(Tmax = 260s) case4(Tmax = 130s)

μ(min) σ(min) μ(min) σ(min) μ(min) σ(min) μ(min) σ(min)

de1 1 0.5 1.2 0.5 1.5 0.5 2 0.5

de2 1 0.5 1.2 0.5 1.5 0.5 2 0.5
de3 2 0.5 2.4 0.5 3 0.5 4 0.5
de4 2 0.5 2.4 0.5 3 0.5 4 0.5

400
C. Zhu et al. Journal of Loss Prevention in the Process Industries 50 (2017) 397–402

Fig. 6. The event sequence diagram from leakage to VCE.

Table 3 quickly. Staff trusted the reliability of the equipment and ignored
Probabilities from leakage to VCE under four cases. the possibility of accidental leakage, which also indicated a lack of
risk consciousness.
Cases case1 case2 case3 case4
(d) Emergency Planning and Response: Initial emergency response was
Probability 0.1440–0.1450 0.3250–0.3260 0.4965–0.4975 0.9970–0.9980 slow. There was basically no emergency management procedure,
and the emergency training was not effectively organized. The
ability to respond dynamically with emergency skills and resilience
(a) Operating Procedures: The operation procedures for staff and dri- in different situations was not sufficient.
vers should be clearly defined. In this case, the unloading process of (e) Process Hazard Analysis: Human errors, lack of detail in proce-
drivers wasn't supervised by a manager, as it should have been. dures, the inability to unload tanker trucks quickly, and 24/7 op-
Also, the procedures for unloading should be modified so that dri- eration fatigue all together multiplied the risk exponentially. These
vers can detect mechanical deficits. Detailed operating procedures combined risks were not fully considered with qualitative and
can be established. In this case, there was no confirmation proce- quantitative methods. In addition, the unloading areas were too
dure to make sure that the connection was good. The risk of mis- close to the tanks. The domino effect risk was also underestimated.
takes in operating procedures had been underestimated. (f) Compliance Audits: The laboratory and control room near the un-
(b) Mechanical Integrity: The reliability of connection components loading area was not explosion-proof, nor were the electrical and
decreased with time. Similarly, the emergency facilities became laboratory equipment. The facilities were not designed to prevent
ineffective with irregular maintenance and time. This dynamic risk an explosion.
was not fully considered. Mechanical integrity management was not
effectively implemented.
6. Conclusions
(c) Training: Staff lacked an awareness of the risk of combustion and
explosion of liquefied petroleum gas. Even when the leak was un-
VCE accidents often occur in chemical plants in developing coun-
controlled, the staff still wasted of time and did not evacuate
tries. These lessons we learned from this accident should be understood

401
C. Zhu et al. Journal of Loss Prevention in the Process Industries 50 (2017) 397–402

by mangers. In this study, a severe VCE incident that occurred in a Disclaimer

chemical plant was described and analyzed. According to the incident
fishbone diagram, chemical plants should implement PSM to ensure The views and opinions expressed in this article are those of the
process safety. The comparison of real and simulation results show that authors and do not necessarily represent those of others.
CFD modeling can provide a good baseline for quantitative con-
sequence risk analysis. Also, the modeling results can provide good
guidance for PSM. Finally, based on the event sequence diagram and Acknowledgments
reasonable assumptions, the probabilities of incident occurrence from
gas leakage to VCE under distinctive PSM conditions of good, regular, This research is sponsored by the Jiangsu Government Scholarship
weak and bad were calculated by Monte Carlo methods. The quanti- for Overseas Studies. Authors are also very thankful to Tencent video
tative results showed that the good PSM, using well-functioning (https://v.qq.com/x/page/h0513tcwqe7.html) for providing some in-
emergency management can significantly reduce the likelihood of a formation about this accident.
VCE incident.

Appendix

Types of events, conditions and gates in an ESD and their iconic representations.

Symbol Annotation

Initial event: Beginning event of the ESD

Random delay: The delay time is a random number.

Action state event: The likelihood of each action state is expressed by probability or random number.

Comment event: Providing information of the development of event sequence.

Termination event:An end state of the ESD

Time condition: The direction of the event sequence is influenced by event time.

Action correctness condition: The direction of the event sequence is influenced by event state.

Output AND gate: The occurring of one input event will lead to multiple output events occurring.

Input AND gate: The output event will not occur until all input events occurs.

References based on event sequence diagram. Reliab. Eng. Syst. Saf. 114, 36–44.
Moosemiller, M., 2011. Development of algorithms for predicting ignition probabilities
and explosion frequencies. J. Loss Prev. Process Industries 24, 259–265.
Atkinson, G., Coldrick, S., Gant, S., Cusco, L., 2015. Flammable vapor cloud generation Sharma, R.K., Gurjar, B.R., Wate, S.R., Ghuge, S.P., Agrawal, R., 2013. Assessment of an
from overfilling tanks: learning the lessons from Buncefield. J. Loss Prev. Process accidental vapour cloud explosion: lessons from the indian oil corporation Ltd. ac-
Industries 35, 329–338. cident at Jaipur, India. J. Loss Prev. Process Industries 26, 82–90.
Chang, J.I., Lin, C., 2006. A study of storage tank accidents. J. Loss Prev. Process Vílchez, J.A., Espejo, V., Casal, J., 2011. Generic event trees and probabilities for the
Industries 19, 51–59. release of different types of hazardous materials. J. Loss Prev. Process Industries 24,
Cozzani, V., Antonioni, G., Landucci, G., Tugnoli, A., Bonvicini, S., Spadoni, G., 2014. 281–287.
Quantitative assessment of domino and NaTech scenarios in complex industrial areas. Villa, V., Paltrinieri, N., Khan, F., Cozzani, V., 2016. Towards dynamic risk analysis: a
J. Loss Prev. Process Industries 28, 10–22. review of the risk assessment approach and its limitations in the chemical process
He, B., Jiang, X., Yang, G., Xu, J., 2017. A numerical simulation study on the formation industry. Saf. Sci. 89, 77–93.
and dispersion of flammable vapor cloud in underground confined space. Process Saf. Work Safety Committee of the State Council in China. (2017). Available at: http://www.
Environ. Prot. 107, 1–11. chinasafety.gov.cn/newpage/Contents/Channel_4977/2017/0616/289880/content_
Hemmatian, B., Abdolhamidzadeh, B., Darbra, R.M., Casal, J., 2014. The significance of 289880.htm. Accessed 16.6.2017.
domino effect in chemical accidents. J. Loss Prev. Process Industries 29, 30–38. Zhang, H., Zheng, X., 2012. Characteristics of hazardous chemical accidents in China: a
Khan, F., Rathnayaka, S., Ahmed, S., 2015. Methods and models in process safety and risk statistical investigation. J. Loss Prev. Process Industries 25, 686–693.
management: past, present and future. Process Saf. Environ. Prot. 98, 116–147. Zhao, J., Suikkanen, J., Wood, M., 2014. Lessons learned for process safety management
Knegtering, B., Pasman, H.J., 2009. Safety of the process industries in the 21st century: a in China. J. Loss Prev. Process Industries 29, 170–176.
changing need of process safety management for a changing industry. J. Loss Prev. Zhou, J., Reniers, G., Khakzad, N., 2016. Application of event sequence diagram to
Process Industries 22, 162–168. evaluate emergency response actions during fire-induced domino effects. Reliab. Eng.
Konstandinidou, M., Nivolianitou, Z., Kefalogianni, E., Caroni, C., 2011. In-depth analysis Syst. Saf. 150, 202–209.
of the causal factors of incidents reported in the Greek petrochemical industry. Zhu, C., Jiang, J., Yuan, X., 2012. Study on ignition probability of flammable materials
Reliab. Eng. Syst. Saf. 96, 1448–1455. after leakage accidents. Procedia Eng. 45, 435–441.
Li, Y., Ping, H., Ma, Z., Pan, L., 2014. Statistical analysis of sudden chemical leakage Zhu, Y., Qian, X., Liu, Z., Huang, P., Yuan, M., 2015. Analysis and assessment of the
accidents reported in China between 2006 and 2011. Environ. Sci. Pollut. Res. 21, Qingdao crude oil vapor explosion accident: lessons learnt. J. Loss Prev. Process
5547–5553. Industries 33, 289–303.
Luo, P., Hu, Y., 2013. System risk evolution analysis and risk critical event identification

402