You are on page 1of 15

Journal of Safety Research 82 (2022) 13–27

Contents lists available at ScienceDirect

Journal of Safety Research


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

Risk assessment of unsafe behavior in university laboratories using the


HFACS-UL and a fuzzy Bayesian network
Ziqi Li, Xiaolong Wang, Shiji Gong, Ninghao Sun, Ruipeng Tong ⇑
School of Emergency Management and Safety Engineering, China University of Mining and Technology-Beijing, Beijing 100083, China

a r t i c l e i n f o a b s t r a c t

Article history: Introduction: Risk assessment for unsafe behavior is an important task in the management of university
Received 26 July 2021 laboratories. Yet related research activities are still in the early stages. This paper attempts to deepen the
Received in revised form 25 October 2021 insight and provide a basis for further research. Method: As traditional methods are inadequate in terms
Accepted 15 April 2022
of quantitative assessment and uncertainty handling, this paper proposes a method to assess the risk of
Available online 27 April 2022
unsafe behavior in university laboratories using the human factor analysis and classification system for
university laboratories (HFACS-UL)-fuzzy Bayesian network (BN) approach. A BN structure was estab-
Keywords:
lished using the HFACS-UL model for the identification of factors influencing unsafe behavior. Using a
University laboratory safety
Risk assessment
fuzzy BN approach, parameters are learned based on prior knowledge and expert experience. The model
Fuzzy Bayesian network is then applied for inference analysis to identify the main risk factors. The key agents were also analyzed
University laboratories human factor along with meta-networks to determine further preventive and control measures. Results: Taking chem-
analysis and classification system istry laboratories of a university as an example, the results show that the probability of unacceptable
Unsafe behavior unsafe behavior in chemical laboratories is 86%, indicating that commitment and cooperation from dif-
ferent agents are required. Of the 24 risk factors, poor organizational climate, with a sensitivity value
of 24.1%, has the greatest impact on unsafe behavior. The most fundamental factor contributing to the
occurrence of unsafe behavior is inadequate legislation, which in turn results in unacceptable external
factors and inadequate supervision, thus forming the most likely causal chain. The functional department,
lab center director, and secondary faculty leadership team are the most critical agents. Conclusions:
Results from the chemistry laboratories demonstrate the credibility of the model. Practical applications:
This study may help provide technical support for risk management in university laboratories.
Ó 2022 National Safety Council and Elsevier Ltd. All rights reserved.

1. Introduction (U.S. Chemical Safety Board, 2018). The data suggest that univer-
sity laboratories are not in a low-risk environment, but rather they
The misconception that university laboratories are ‘‘lower risk” are even more dangerous than industrial laboratories because of
and ‘‘inherently safer” than industrial laboratories and factory the low investment and lax regulations on laboratory safety
operations persists inside and outside of academia (Olewski & (Schröder, Huang, Ellis, Gibson, & Wayne, 2016). It is therefore nec-
Snakard, 2017). Although the number of hazardous substances essary for agents to develop effective risk management practices to
may be small, university laboratories have multiple occupational ensure that laboratory safety is a priority.
hazards in their research and teaching activities, including chemi- In ISO 45001:2018, risk assessment is a mandatory requirement
cal, physical, behavioral, and psychological risk factors. The risks that provides guidance to risk managers. This assessment is a com-
associated with such hazards can pose a threat to the health and plete process consisting of risk identification, risk analysis, and risk
safety of agents if not managed properly. Accidents caused by poor evaluation. Currently, a series of studies on risk assessment-related
laboratory management are all major events (Peng, Li, & Zhou, issues (e.g., methods, applications) have been conducted in differ-
2019) and have claimed more than 70 lives worldwide since ent fields (e.g., energy, medicine, transportation) by various
2000, according to statistics. Chemical laboratories in the United researchers (Cattelani et al., 2020; Liu, Yang, Cui, & Yang, 2020).
States alone were involved in more than 260 accidents from Among these areas, there is limited research on risk assessment
2001 to 2018, with most of the accidents occurring in schools as it relates to education. According to the study, while risk assess-
ment and management has gradually become a popular topic in
university laboratory safety over the past decade, safety research
⇑ Corresponding author.
activities for laboratories within the education system are still in
E-mail address: tongrp@cumtb.edu.cn (R. Tong).

https://doi.org/10.1016/j.jsr.2022.04.002
0022-4375/Ó 2022 National Safety Council and Elsevier Ltd. All rights reserved.
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

the early stages (Yang, Reniers, Chen, & Goerlandt, 2019). There are the results of risk assessment. Therefore, the influences of unsafe
many aspects of occupational health and safety management in behavior were first identified using the HFACS-UL model, which
university laboratories that can be improved. can drill down into the hidden factors of an accident and classify
Statistics show that nearly half of laboratory accidents are the factors into the appropriate categories. Then, triangular fuzzy
caused by human factors. Human causes mainly involve violation numbers based on subjectivity and objective expert weights were
of operating procedures, improper operation or careless operation adopted to ensure the reliability of expert assessments to cope
(Zhu, Tang, Li, & Fang, 2020). For this reason, some scholars have with uncertainty more effectively. A BN model for unsafe behavior
shifted the focus of laboratory safety research to human behavior. in university laboratories was developed on this basis. Moreover,
Zhu et al. (2020) identified key factors for laboratory explosions key influencing factors were identified through inferential analysis
from four aspects: human, machine, environment, and manage- of the BN, and targeted risk control measures were proposed. These
ment, noting that human causes include lack of knowledge, poor measures were based on an analysis of the relationship between
operation, lack of training, and negative attitudes. The investiga- agents and key influences. This study can provide technical support
tion by Walters, Lawrence, and Jalsa (2017) found that safety for risk management in university laboratories.
awareness among students in chemistry laboratories is deficient The structure of the paper is as follows. Section 2 describes the
and education and training should be further enhanced. methodology used in this paper. The detailed procedure of the
Jorgensen (2017) developed an instrument to assess the level of developed risk assessment and control model is described in Sec-
unsafe behaviors in research laboratories. Significant differences tion 3. In Section 4, the model is applied to chemical laboratories
in the assessment of safety behaviors were measured through psy- at a university, and the case is discussed and analyzed. The relevant
chometric properties. Given that laboratory accidents are intrinsi- results are provided in Section 5.
cally related to unsafe human behavior (Peng et al., 2019; Shariff &
Norazahar, 2012; Wu, Liu, & Lu, 2007), this study proposed a risk
2. Objects and methods
assessment and control framework for unsafe behavior in univer-
sity laboratories, which is important for improving research effi-
2.1. Research objects
ciency and ensuring personnel safety.
A number of approaches to risk assessment and control are cur-
Behavioral safety research in university laboratories focuses on
rently available, and some have been used to manage laboratory
unsafe human behavior. According to the definition in GB6441-
safety in universities. Omidvari, Mansouri, and Nouri (2015) devel-
1986 ‘‘Classification for Casualty Accidents of Enterprise Staff and
oped a model based on the analytical hierarchy process (AHP) and
Workers,” human errors that can cause accidents are unsafe behav-
failure mode and effect analysis (FMEA) logic, which identified the
iors. Therefore, all actors that can cause human error within the
factors influencing fire risk in educational center laboratories. Ho
university laboratory system are the objects of concern in this
and Chen (2018) analyzed and improved the waste liquid disposal
study. They include laboratory operators (faculty and students),
process using a fuzzy theory-based FMEA approach in combination
executive leaders (such as deans), and laboratory safety adminis-
with real cases from university chemistry laboratories in Taiwan.
trators in faculties, departments, and research units within the uni-
However, little risk management research has been conducted
versity, as well as supervisors in functional departments
specifically on unsafe behavior in university laboratories. Although
responsible for laboratory management in the university (such as
Shariff and Norazahar (2012) proposed a laboratory at-risk behav-
directors of laboratory management offices). Relevant personnel
ior and improvement system (Lab-ARBAIS) to monitor and control
in higher-level local and national departments managing labora-
the high-risk behaviors of students in the laboratory, the role of
tory safety in universities are also involved. These agents may
other agents was neglected. Most of the researchers mentioned
cause unsafe behavior during operation or supervision because of
earlier used traditional risk management techniques to analyze
personal mistakes, which eventually lead to accidents. For exam-
laboratory risks from a qualitative or quantitative perspective,
ple, when experimenters are under tremendous research pressure,
but these methods are limited to static control management. BN,
they may develop psychological barriers such as frustration and
in contrast, offers significant advantages over traditional risk
discouragement, which in turn trigger misconduct; supervisors
assessment techniques. The BN is a causal analysis tool for uncer-
who develop laboratory management procedures bring their own
tainty inference that has been shown to be useful for acquiring and
attitudes or experiences into their work, which may also influence
merging information from multiple sources. It is widely used in the
them to objectively identify risk factors and form erroneous
field of uncertain knowledge expression and inference and can
judgements.
update the prior probabilities in the form of a likelihood function
Unsafe behavior in university laboratories reflects more than
(Cai, Min, Liu, & Feng, 2018; Ding, Ji, & Khan, 2020). Traditional
just the individual’s behavioral errors. Studies of unsafe behavior
BNs, however, require probability measures to assess uncertainty,
have found that many accidents are caused by unsafe human
but accurate information is sometimes lacking in the management
actions, while unsafe actions are triggered by defects in the safety
of university laboratories. Probabilistic estimation models that
management system (Yu, Cao, Xie, Qu, & Zhou, 2019). Therefore,
combine expert judgement and fuzzy set theory can be alternatives
taking unsafe actions in the laboratory alone as the object of study
when historical data are insufficient (Li, Wang, Wang, Shao, & He,
will inevitably leave out the safety behavior performance of each
2020; Li, Xu, & Shuai, 2020).
agent. Therefore, risk assessment of unsafe behavior in university
Safe practices are mandatory and are the responsibility of all
laboratories should focus on different subjects, including non-
agents engaged in academic activities within a laboratory, includ-
personnel factors such as the environment.
ing managers and researchers. Successful risk management is
dependent on effective control methods. The demand for safety
may not be satisfied when personnel (students, faculty, safety 2.2. Establishment of the HFACS-UL model
administrators) involved in related activities in the laboratory pos-
sess varying degrees of risk control skills. To address this issue, this The HFACS, proposed by Wiegmann and Shappell based on the
study developed a dynamic fuzzy BN model to assess unsafe Swiss cheese model (SCM), is an open system approach for analyz-
behavior risk in university laboratories, which is a primary task ing human factors. The system focuses on potential and apparent
for effective control. As the first step of risk management, risk iden- failures and the interrelationships between them. Specifically, it
tification and its comprehensiveness can have a great impact on provides a detailed delineation of the causes of human error at four
14
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

levels: organizational influences, unsafe supervision, preconditions BN is a useful risk management method that can easily deal
for unsafe acts, and unsafe acts. In general, the HFACS framework with incomplete data, so it has been widely used for risk analysis
involves a risk assessment that can be appropriately adjusted to and decision making in areas such as mine outburst hazards
specific industry characteristics. As a result, many scholars devel- (Wu, Xu, Zhou, & Qin, 2016), forest fires (Sevinc, Kucuk, & Goltas,
oped a fifth layer of factors, external factors, which introduced eco- 2020), and flood control operations (Chen, Zhong, An, Zhu, & Xu,
nomic, legal, and policy matters as complements to the HFACS. 2019). However, it is seldom utilized to assess the risk of unsafe
Excellent effects have been achieved in the construction (Xia, behavior in university laboratories. Therefore, this paper intro-
Zou, Liu, Wang, & Zhu, 2018), marine transportation (Zhang, duces BN as a suitable method to analyze the safety status of a lab-
Zhang, Goerlandt, Yan, & Kujala, 2019), and mining industries oratory. The rationale for BN is joint probability distributions and
(Liu et al., 2019). conditional independence, which is shown as follows:
Based on typical accident reports (2001–2018) of domestic and Y
n
 
foreign university laboratories and the analysis results of the acci- PðXÞ ¼ PðX1 ; X2 ;    ; Xn Þ ¼ P Xi jParentðXi Þ ð1Þ
dent research literature, it is found that external factors (such as i¼1
society) could lead to the increase of unsafe behaviors. Moreover,
where Parent(Xi) is the parent set of the variable Xi.
the results of existing university laboratory safety studies have
The prior probability of a variable can be updated dynamically
shown that external factors are also an important part of university
under the condition that new evidence U is given. The posterior
laboratory risk management that cannot be ignored (Dai, Fu, Liu, &
probability of the variable can then be obtained as follows:
Tong, 2019). Therefore, the original HFACS model was improved in
this study. Similar to the original HFACS model, the overall struc- PðX; UÞ PðX; UÞ
PðX1 ; X2    Xn jUÞ ¼ ¼ P ð2Þ
ture was decomposed top-down from five main levels of human PðUÞ X PðX; UÞ
error causes (including external influence, organizational influ-
ence, unsafe supervision, preconditions for unsafe behavior, and
2.4. Triangular fuzzy theory
unsafe behavior) and some adjustments were made in the sub-
items. The three newly added external influencing factors include
Fuzzy theory, first proposed by Zadeh (1965), employs fuzzy
inadequate legislation, design flaws, and social factors. In terms
numbers to represent imprecise values and uses membership func-
of legislation, although Chinese regulations and policies cover
tions to show uncertainties. Therefore, this study combines trian-
chemical safety, special equipment, occupational health, and many
gular fuzzy number theory with BN (namely, fuzzy BN) to form a
other fields (such as the Regulations on the Safe Management of
powerful problem-solving technique. The method has been applied
Hazardous Chemicals), the laws specifically for university labora-
to issues such as urban natural gas pipelines (Li et al., 2021) and
tories are still in a gap. As a result, there is no law to be followed
coal mine gas explosions (Li et al., 2020). Specifically, triangular
in the process of laboratory safety management. At the same time,
fuzzy number theory is used to convert uncertain linguistic vari-
the requirements for experimental facilities are gradually
ables to exact values in place of traditional BN probabilities. The
improved with the change of teaching methods in universities.
membership functions of the triangular fuzzy numbers are given
Laboratory construction is not only the purchase of reasonable
in Fig. 2 and in Eq. (3), where l, m, and u represent the lower bound,
equipment but also the overall planning and graphic design of
the most likely value, and the upper bound, respectively.
the laboratory. Basic conditions such as power supply, water sup-
8
ply, ventilation, and environmental protection need to be consid- > xl
< ml x 2 ½l; m
ered as well. In addition, some old university laboratories are l M ð xÞ ¼ xu
x 2 ½m; u ð3Þ
likely to be defective in their layout. In order to avoid greater losses > mu
:
0 others
from accidents, it is necessary to strengthen safety management by
regulating human behavior. In terms of social factors, university For the two triangular fuzzy numbers g M1 ¼ ðl1 ; m1 ; u1 Þ and
laboratory safety management should be linked with many gov- g
M2 ¼ ðl2 ; m2 ; u2 Þ, the relation is as in Eq. (4).
ernment departments and social institutions (such as environmen-
8
tal protection, public security, traffic, and research fund > e1M
M e 2 ¼ ðl1 þ l2 ; m1 þ m2 ; u1 þ u2 Þ
>
>
management) to create a good external environment for university >
<Me1Me 2 ¼ ðl1  l2 ; m1  m2 ; u1  u2 Þ
laboratory management. In summary, it makes sense to add the ð4Þ
>
> e
M1  Me 2 ¼ ðl1 l2 ; m1 m2 ; u1 u2 Þ
external factor to the analysis of unsafe behavior in university lab- >
>
:
oratories. The potential risk factors were finally identified, as kM e ¼ ðkl; km; kuÞ
shown in Fig. 1. This is a generic HFACS-UL model, so items can
be changed appropriately when performing risk assessments for
different laboratory scenarios. 3. Risk analysis process for university laboratories

2.3. Bayesian network To improve the effectiveness and reliability of risk assessment
and control of unsafe behavior in university laboratories, a new
The Bayesian network, also known as the belief network or cau- method based on fuzzy BN was explored. Fig. 3 illustrates the anal-
sal network, is a graphical model for describing the dependencies ysis process of the method. The first step is risk identification,
among data variables (Pearl, 1986). It has the functions of statisti- which depends on HFACS-UL to reveal potential risk factors and
cal decision making, causal analysis, and prediction, which can cause-and-effect relationships. The second step is to construct a
solve problems such as posterior probability and maximum poste- BN that converts entities into nodes and determine the probability
rior hypothesis. A BN structure is a directed acyclic graph (DAG), of risk factors using fuzzy theory. The third step is calculation and
where each node represents an attribute or random variable, and risk analysis, involving risk evaluation and providing risk control
the arcs between nodes show probabilistic dependencies among strategies. Key risk factors are acquired through fuzzy probability
attributes (data variables). Meanwhile, the conditional probability calculations and sensitivity analysis using the advanced reasoning
table (CPT) of Bayesian nodes is used to describe the probability of capabilities of BN. The risk control strategy is based on the analysis
a quantitative relationship between the dependent variable and associated with agents and risk factors, and then specific measures
the parent nodes. are designed for different agents. The results of the study can pro-
15
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Fig. 1. Modified HFACS-UL method for risk identification of unsafe behavior in university laboratories.

tors (such as work environment and safety climate; Qiao, Liu, Li,
Luo, & Wan, 2018). However, university laboratories are complex
social-psychological systems whose risk management is somewhat
unique in terms of material and equipment as well as personnel
composition. Moreover, the major risk factors vary in different
types of laboratories (e.g., biology, chemistry, physics). Therefore,
there is a need to identify risk factors and the relationships among
them in a contextualized manner. The unique risks include those
Fig. 2. Membership function for triangular fuzzy numbers.
arising from the different knowledge, attitudes, and experiences
of agents and the different teaching or research purposes of the
vide an objective reference for risk management in university lab- laboratory. Risk factors can be obtained from a priori knowledge
oratories and help reduce or prevent laboratory accidents or involving three sources: the analysis of typical accidents in univer-
injuries. sity laboratories, the analysis of factors influencing unsafe behavior
from previous research, and the empirical judgement of experts. Of
these, the expert’s empirical knowledge is also useful in revising
3.1. Risk identification the risk list to form the basis for subsequent analysis.
Therefore, to better classify risk factors accurately, based on the
The occurrence of unsafe behavior is influenced by various fac- improved HFACS-UL framework, four experts involved in safety
tors. Combining different industries, the present research on influ- management in university laboratories were consulted. The
encing factors can be broadly divided into the following three experts have more than 10 years of experience in laboratory safety
categories: personal factors, organizational factors, and other fac- management. They mapped the risk factors obtained from different
16
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Fig. 3. Framework for laboratory risk assessment and control.

sources into HFACS-UL for checking. The classification categories of there are more than 20 nodes in a BN, no fewer than three experts
risk factors are described in Table 1. should be selected to make a judgement (Banuls, Turoff, & Hiltz,
2013). This is because the opinions of multiple experts can avoid
3.2. Construction of the BN model bias or error and make subjective assessments more scientific. In
practice, however, the personal experience and knowledge of each
Based on risk identification, connections can be made among expert can produce different perceptions. Therefore, the use of a
the identified risk factors, resulting in a system of laboratory risk weighting factor is necessary to characterize the competence and
assessment indicators. In the process of BN learning, the aggrega- quality of the experts for the purpose of improving consistency
tion of a priori knowledge is crucial. Therefore, each indicator and reliability. Generally, the arithmetic average method can be
was converted to a node of the BN, and the state of each node adopted to obtain relatively reasonable weighting results (Laal
was determined to establish the initial BN. The BN model devel- et al., 2020; Li et al., 2020). With reference to laboratory risk man-
oped from the HFACS-UL is shown in Fig. 3. The probability of each agement practices, this study divided experts’ personal informa-
node can be identified in a number of ways, for example, by relying tion into three categories, including years of experience,
on accident statistics. With limited historical data on safety acci- education level, and professional status. The weighting criteria of
dents in university laboratories, however, this study utilized a expert selection can be seen in Table 2.
combination of expert experience and fuzzy theory to assist in esti-
mating conditional probabilities. 3.2.2. Expert assessment and fuzzy judgement
Risk assessment of laboratory safety in universities is a multi-
3.2.1. Expert selection and weighting indicator decision problem. To obtain an impartial assessment,
It is important to ensure the reliability of subjective judgements we collected experts’ linguistic descriptions of the indicators to
from experts when the conditional probabilities of BN nodes are be assessed. Such descriptions, integrated with fuzzy set theory,
obtained according to the experts. Banuls et al. suggested that if may help risk assessment overcome possible ambiguities (Yazdi,
17
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Table 1
Classification categories of risk factors.

Level Risk category Description of risk category


External influences Inadequate legislation The legal guarantee system for university laboratory safety is imperfect or has gaps, and safety supervision lacks
(L1) (R1) strong regulations
Design flaws (R2) Deficiencies in the design of the laboratory building layout, equipment layout, water supply and drainage,
electrical, etc.
Social factors (R3) Economic, political, safety culture and other social factors
Organizational Organizational process The procedures for binding and regulating agents in the organization are insufficient and there are gaps in the
influences (L2) (R4) operational process, such as conflicting operational requirements
Organizational climate Individual behavior is influenced by poor organizational structure, culture or policies, including unclear division of
(R5) responsibility for safety, inadequate risk management procedures, low management commitment
Resource management Inappropriate management, allocation and maintenance of the organization’s human, financial and equipment
(R6) resources
Unsafe supervision Supervisory violations The supervisor intentionally violates orders, instructions, rules or regulatory procedures, such as allowing teachers
(L3) (R7) to purchase lab materials without permission
Inadequate supervision Supervisors or organizers fail to provide sufficient professional guidance, training or oversight, which makes it
(R8) impossible to identify and control risks
Planned inappropriate Failure of supervision to effectively assess the risks related to the behavior of inexperienced personnel who are
operations(R9) allowed to perform tasks that are beyond their capabilities
Failure to correct Oversight fails to correct problems and still allows them to persist, even though they are obvious
problems (R10)
Preconditions for Conditions of individuals Adverse physiological states (R11): Physiological events that lead to low physical functioning, such as lack of sleep
unsafe acts (L4) (I1) and physical fatigue
Adverse mental states (R12): Guided by the personality of the stakeholder, or the presence of psychological
problems, mental fatigue, poor vigilance, intellectual deficiency
Environmental factors Physical environment (R13): Including weather, lighting, noise, hazardous gases, lack of hazard signs, and excessive
(I2) clutter, etc.
Technological environment (R14): Failure of equipment or devices, defective PPE, or lack of safety protection
devices
Personal factors (I3) Personal readiness (R15): Managers disregard rules of personal competence and individuals exist with inadequate
rest or training
Crew resource mismanagement (R16): Including various inappropriate coordination, communication and
teamwork issues
Unsafe acts (L5) Violations (R17) Disregard of rules and instructions, or violation of procedures or regulations due to unskilled operation
Errors (I4) Skill-based Errors (R18): Technical errors due to improper execution procedures, inadequate training or poor
operational skills
Perceptual Errors (R19): Perceptions and awareness of things deviate from reality
Decision Errors (R20): Errors of judgement due to poor information, knowledge or experience, such as using the
wrong tools or responding incorrectly to emergencies

Table 3
Table 2 Judgement of triangular fuzzy numbers.
Weighting criteria for expert selection.
Linguistic description Aggregation of fuzzy numbers
Series Description Score
High (H) (0.8, 0.9, 1)
Education level PhD 4 Medium-High (MH) (0.6, 0.7, 0.8)
Master 3 Medium (M) (0.4, 0.5, 0.6)
Bachelor 2 Medium-Low (ML) (0.2, 0.3, 0.4)
Less than a Bachelor 1 Low (L) (0, 0.1, 0.2)
Job title Professor, Senior Engineer 4
Associate Professor, Engineer 3
Assistant Professor, Assistant Engineer 2 The opinions of different experts are aggregated according to
Teaching Assistant, Technician 1
the linear opinion pool method after the linguistic variables are
Years of experience >20 5 used to determine the corresponding fuzzy numbers, as shown in
15–20 4
10–15 3
Eq. (5).
5–10 2
X
n
<5 1 Mj ¼ ðqij  wi Þ; j ¼ 1; 2;    ; k ð5Þ
i¼1

where Mj is the aggregated fuzzy number of nodes, wi is the combi-


nation weight, qij is the language evaluation given by expert i about
Daneshvar, & Setareh, 2017). In fuzzy judgements, linguistic prob- node j, k is the number of experts, and n is the number of nodes.
abilities are ultimately quantified as triangular fuzzy numbers
through a suitable affiliation function. Fuzzy numbers provide a
3.2.3. Fuzzy processing
more realistic representation of human judgements. According to
The purpose of the defuzzification process is to convert the
Saaty and Ozdemir (2003), the usual ability of experts to judge is
overall fuzzy number of the aggregate into fuzzy possibilities.
between five and nine. Therefore, we introduce five linguistic vari-
Specifically, the aim is to use the exact value as a representation
ables, namely, high, medium–high, medium, medium–low, and
of the fuzzy number. There are many techniques for defuzzifica-
low. The qualitatively described triangular fuzzy number judge-
tion, including the center of area (CoA), maximizing and minimiz-
ments are shown in Table 3.
ing fuzzy sets, and the mean of maxima (MoM). Among them, the
18
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

area averaging method is not only simpler to calculate but will also ment. The aim is to describe how the model developed in this
not cause the loss of fuzzy information, which is very suitable for research can support risk management in university laboratories.
the results of the triangular membership function to defuzzy. The
process is as follows. 4.1. Background

l þ 2m þ u
M ¼ ð6Þ The university’s chemistry laboratories undertake experimental
4 teaching tasks for chemistry-related majors across the school, with
After obtaining the fuzzy values M* of the nodes, the probability nearly 100 types of experiments and nearly 70,000 people involved
information of the nodes is normalized to satisfy the uniformity of annually. A wide variety of chemicals are involved in the labora-
the probabilities. The prior probability Pj of the relevant node is tory, including corrosive chemicals and flammable agents. They
eventually obtained. may cause accidents and hazards in the process of chemical stor-
age, use, and disposal of waste chemicals. A number of risk factors
M were identified in previous safety self-audits, as shown in Table 4.
pj ¼ Pn  ð7Þ
j¼1 M Hence, it is necessary to conduct a risk assessment of agents’
unsafe behaviors and to reasonably analyze the influencing factors
and control measures.
3.3. Implementation of the BN model Four experts from the university in related fields were invited to
participate in the evaluation to determine the severity of each risk
3.3.1. Inferential analysis factor. Table 5 shows the specific information and weighting values
Based on the establishment of the BN, the prior probability from of the experts.
processing is input for inferential computation and analysis. BN-
based reasoning involves causal induction, reverse inference, and 4.2. Determination of the topology and parameters
sensitivity analysis. Causal induction can be applied to reason
about the likelihood of unsafe behaviors occurring in university Following the expert analysis, the factors included in the model
laboratories, and help managers determine and prevent risks in and the established influencing relationship were considered that
advance by the magnitude of the probability of the outcome. Rev- were applicable to this case. The statistical analysis was then per-
erse inference can analyze the root causes of unsafe behaviors. The formed on a sample of 98 accident cases that occurred in university
probability of each contributing cause for a given result helps man- laboratories at home and abroad between 2001 and 2018. The
agers to rank risk factors. These two inferences rely on formulas (1) probability of partial root nodes was determined after further cal-
and (2). Sensitivity analysis is performed to represent the relation- culations, as shown in Table 6. Accident information and data were
ship between parameters and output probabilities. It is helpful to obtained from the Ministry of Emergency Management of the Peo-
understand whether nodes are faulty and to determine which fac- ple’s Republic of China, the U.S. Chemical Safety Board (CSB), and
tors have a significant impact on unsafe behavior. The higher the related research papers (Fu, Liu, & Liu, 2018). The probability of
sensitivity value, the more important the risk factor is, and man- other follow nodes was determined based on expert judgement.
agers should control it in a more targeted manner. This study uses In Table 7, the status and expert evaluation results of the 15 root
the algorithm proposed by Kjærulff and van der Gaag (2000) to nodes are shown. In addition to the root node, there are eight inter-
analyze the sensitivity of each parent node. The algorithm creates mediate nodes whose states are described in Table 8. The com-
a function for each parameter that represents the output probabil- bined effect of intermediate nodes eventually results in unsafe
ity of the parameter under study. acts, with states described in terms of acceptability and
unacceptability.
3.3.2. Risk control analysis After determining the prior probability of the root node, the
Agents have a direct or indirect influence on unsafe behavior in expert is asked to determine the probability of all other nodes of
university laboratories. Therefore, it is necessary to propose risk the BN and to give the value of the conditional probability. How-
control strategies, from the perspective of stakeholders, combined ever, because of length limitations, only the fuzzy CPT for I2 is
with risk factor analysis. In line with the risk control approach pro- given in Table 9 and Table 10.
posed by Wang et al., the relationship between agents and risk fac- For the BN in Fig. 4, the CPT of the node ‘‘Unsafe supervision” is
tors is presented in this study as follows. the most complex. The node has six parent nodes, meaning that its
CPT has 384 combinations, which makes it difficult for experts to
Risk distribution ¼ social network make accurate judgements. Therefore, it is simplified, assuming
risk distribution network ð8Þ that when ‘‘Failure to correct problems” or ‘‘Planned inappropriate
operations” are not acceptable, ‘‘Supervisory violations” are severe,
From a mathematical point of view, Eq. (8) can be viewed as a
or ‘‘Inadequate supervision” often occurs, ‘‘Unsafe supervision” will
matrix operation, where the elements are based on an organiza-
certainly be generated, so that CPT can be minimized to 72 combi-
tional structure and a distribution of responsibilities. If there is a
nations. This simplification, already applied by Li et al., is feasible.
relationship between the elements, the relationship is set to ‘1’;
otherwise, ‘0.’ On this basis, the social network and the risk distri-
4.3. Results and analyses
bution network are formed. A social network represents the inter-
relationship of different agents (cooperation or reporting), a risk
4.3.1. Causal induction
distribution network represents the relationship between risk fac-
Using the BN model and risk probability assessment method
tors and agents, and the management of risk factors needs to be
developed in Section 3.2, GeNIe software (developed by the deci-
handed over to agents.
sion systems laboratory at the University of Pittsburgh) was used
to aid in the inference. Based on the expert’s initial description of
4. Case study the fuzzy assessment, the probability of the risk event could be
obtained for each intermediate variable. Ultimately, the risk prob-
This paper takes chemistry laboratories in a university as an ability of unsafe behavior in the chemical laboratory was derived,
example in which a fuzzy BN was applied to conduct a risk assess- which is shown in Fig. 5. The unacceptable probability of unsafe
19
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Table 4
Risk factors for unsafe behavior in chemical laboratories.

Risk factors Site risk description Site photos


Poor physical environment Stacking of debris

Excessive debris on lab benches

Poor technical environment Incorrect pressure gauge reading by the fire extinguishing apparatus

Lab personnel not equipped with personal protective equipment

Violations Failure to wear protective gloves when weighing hazardous chemicals

Long hair left untied in the lab

Table 5
Expert information.

Serial number Education level Job title Years of experience Weight score Weight value
Expert 1 PhD Professor Over 20 years 13 0.2955
Expert 2 PhD Professor 19 years 12 0.2727
Expert 3 PhD Senior Engineer 11 years 11 0.2500
Expert 4 Master Engineer 8 years 8 0.1818

influential factors at each level include inadequate legislation at


Table 6
Probability of a few root nodes. the external influence level (80% probability of inadequacy) and
social factors (74% probability of influence); organizational climate
Bayesian nodes State Probability
at the organizational influence level, which has an unacceptable
Violations Yes 0.4184 probability of 64%; inadequate supervision at the unsafe supervi-
No 0.5816
sion level, which has a 47% probability of recurrence; individual
Skill-based errors Many 0.2245
Little 0.7755 readiness at the personnel factor level (inadequate probability of
Failure to correct problems Acceptable 0.8367 62%); and crew resource mismanagement (unacceptable probabil-
Unacceptable 0.1633 ity of 59%) among personnel factors. All of these factors need to be
Technological environment Adverse 0.3571 focused on and controlled by appropriate measures to reduce the
Good 0.6429
Resource management Irrational 0.1224
occurrence of unsafe behaviors.
Rational 0.8776

4.3.2. Reverse inference


By using the reverse inference function of the posterior proba-
behavior is 86%. According to the checklist for laboratory safety bility to identify the most likely factors leading to the outcome, risk
inspection in higher education released by the Ministry of Educa- management can be carried out more precisely. Accordingly, the
tion in China, the chemistry laboratory of the university conducted node ‘‘unsafe behavior” is set as the evidence node following the
four self-inspections in the past year, with 104 items. A total of 344 rule, that is, the probability value of the unacceptable state of
violations and errors were found, accounting for 82.7% of the total, unsafe behavior is assumed to be 100%. The posterior probability
as shown in Table 11. This is consistent with the results of the lab- distribution is then derived (see Fig. 6).
oratory’s previous self-examination of safety hazards and verifies The probability of unsafe supervision L3 occurred with the high-
the reasonableness of the constructed model. Specifically, the more est probability value of 94.34% from the posterior probability,
20
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Table 7
Evaluation results of the 15 root nodes.

Parent node State Fuzzy probability distribution Fuzzy probability value Prior probability
R1 Extremely imperfect (0.4545, 0.5545, 0.6545) 0.5545 0.3160
Imperfect (0.7500,0.8500,0.9500) 0.8500 0.4845
Relatively perfect (0.1500,0.2500,0.3500) 0.2500 0.1425
Perfect (0,0.1000,0.2000) 0.1000 0.0570
R2 Slight (0.6864,0.7864,0.8864) 0.7864 0.4701
General (0.4909,0.5909,0.6909) 0.5909 0.3532
Serious (0.0955,0.1955,0.2955) 0.1955 0.1169
Fatal (0,0.1000,0.2000) 0.1000 0.0598
R3 Huge impact (0.4500,0.5500,0.6500) 0.5500 0.2902
Large impact (0.7455, 0.8455, 0.9455) 0.8455 0.4461
Small impact (0.2500,0.3500,0.4500) 0.3500 0.1846
No impact (0.0500,0.1500,0.2500) 0.1500 0.0791
R4 Totally acceptable (0,0.1000,0.2000) 0.1000 0.0759
Partially acceptable (0,3636,0.4636,0.5636) 0.4636 0.3517
Unacceptable (0.6545,0.7545,0.8545) 0.7545 0.5724
R5 Totally acceptable (0,0.1000,0.2000) 0.1000 0.0794
Partially acceptable (0.2545,0.3545,0.4545) 0.3545 0.2816
Unacceptable (0.7045,0.8045,0.9045) 0.8045 0.6390
R7 Severe (0,0.1000,0.2000) 0.1000 0.0587
Serious (0.2000,0.3000,0.4000) 0.3000 0.1760
Relatively serious (0.4000,0.5000,0.6000) 0.5000 0.2933
General (0.7045,0.8045,0.9045) 0.8045 0.4720
R8 Often (0.7500,0.8500,0.9500) 0.8500 0.4710
Occasionally (0.4545,0.5545,0.6545) 0.5545 0.3073
Rarely (0.2000,0.3000,0.4000) 0.3000 0.1663
Never (0,0.1000,0.2000) 0.1000 0.0554
R9 Totally acceptable (0.5091,0.6091,0.7091) 0.6091 0.3702
Partially acceptable (0.6364,0.7364,0.8364) 0.7364 0.4475
Unacceptable (0.2000,0.3000,0.4000) 0.3000 0.1823
R11 Severe (0,0.1000,0.2000) 0.1000 0.0525
Serious (0.2000,0.3000,0.4000) 0.3000 0.1575
Relatively serious (0.5045,0.6045,0.7045) 0.6045 0.3174
General (0.8000,0.9000,1.0000) 0.9000 0.4726
R12 Severe (0,0.1000,0.2000) 0.1000 0.0545
Serious (0.2000,0.3000,0.4000) 0.3000 0.1634
Relatively serious (0.4364,0.5364,0.6364) 0.5364 0.2921
General (0.8000,0.9000,1.0000) 0.9000 0.4901
R13 Totally acceptable (0.0364,0.1364,0.2364) 0.1364 0.0938
Partially acceptable (0.7136,0.8136,0.9136) 0.8136 0.5594
Unacceptable (0.4045,0.5045,0.6045) 0.5045 0.3469
R15 Extremely inadequate (0.4955,0.5955,0.6955) 0.5955 0.3075
Inadequate (0.6955,0.7955,0.8955) 0.7955 0.4108
Adequate (0.3455,0.4455,0.5455) 0.4455 0.2301
Extremely adequate (0,0.1000,0.2000) 0.1000 0.0516
R16 Totally acceptable (0,0.1000,0.2000) 0.1000 0.0749
Partially acceptable (0.35,0.45,0.55) 0.4500 0.3372
Unacceptable (0.6864,0.7864,0.8864) 0.7864 0.5879
R19 Often (0.3091,0.4091,0.5091) 0.4091 0.2228
Occasionally (0.6591,0.7591,0.8591) 0.7591 0.4134
Rarely (0.4091,0.5091,0.6091) 0.5091 0.2772
Never (0.0591,0.1591,0.2591) 0.1591 0.0866
R20 Often (0.8000,0.9000,1.0000) 0.9000 0.4626
Occasionally (0.5455,0.6455,0.7455) 0.6455 0.3318
Rarely (0.2000,0.3000,0.4000) 0.3000 0.1542
Never (0,0.1000,0.2000) 0.1000 0.0514

which is the most likely factor leading to unsafe behavior. Of this, 100%, the probability value of node L1 is the largest at this point.
the probability of frequent occurrence of inadequate supervision is The probability of unacceptable external influence is 82.57%, as
the highest, at 47.1%, which deserves focused attention. Therefore, shown in Fig. 7(a). Then, the inference is continued in reverse for
this factor should have a priority to be checked when there are the node external influence. Assuming that the probabilities of
unsafe behaviors in university laboratories. L1, L3, and L5 negative situations occurring are all 100%, the proba-
When it is clear that unsafe supervision occurs, that is, the prob- bility value of node R1 inadequate legislation is calculated to be the
ability of occurrence of this node is 100%, the reverse inference is largest, with a value of 83%, as shown in Fig. 7(b). Following the
continued, and other risk factors are calculated. Assuming that above analysis, the most likely causal chain leading to unsafe
the unacceptable state probability value of unsafe supervision L3 behaviors is {R1 ? L1 ? L3 ? L5}, namely, {inadequate legisla-
and the occurrence probability value of unsafe behavior L5 are both tion ? unacceptable external factors ? unsafe supervision ? un-

21
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Table 8 4.4. Discussion on risk control strategies


Status of the eight intermediate nodes.

Risk category State Combined with the actual situation of laboratory safety man-
External influences (L1) Acceptable/Unacceptable agement in this school, the relevant agents are divided into six cat-
Organizational influences (L2) Acceptable/Unacceptable egories: student, teacher, laboratory safety manager, laboratory
Unsafe supervision (L3) Yes/No director, laboratory center director, secondary faculty leadership
Preconditions for unsafe acts (L4) Acceptable/Unacceptable team, functional department, school leader, and education depart-
Conditions of individuals (I1) Acceptable/Unacceptable
Environmental factors (I2) Acceptable/Unacceptable
ment. The basic safety responsibilities of the above agents are
Personal factors (I3) Acceptable/Unacceptable listed in Table 12.
Errors (I4) Many/Little The four experts mentioned earlier were then consulted to
gather their judgements on the existence of factor relationships.
The relative value of the association between agents and risk fac-
safe behaviors}. The most fundamental factor for unsafe behavior tors was determined based on the calculation of Eq. (8), which
in the laboratory is also identified as the imperfection of the rele- means that the relevant agents with primary responsibility for
vant legislative policies. the key risk factors were identified, and the results are shown in
Fig. 9.
According to Wang et al.’s risk classification criteria, high
4.3.3. Sensitivity analysis impact levels are indicated in red, and low impact levels are indi-
In GeNIe software, the nodes represented by the five levels are cated in green. Ultimately, in conjunction with the Pareto principle,
selected as target nodes, meaning that L1
L5 are set as targets. the relevant agents ranked in the top 20% of impact levels are con-
Then, sensitivity analysis is performed to obtain the sensitivity dis- sidered key agents, and risk control strategies are proposed based
tribution, as shown in Fig. 8. on this.
From Fig. 8, it can be seen that nodes R5 (organizational cli- In the impact matrix above, the functional department and the
mate), R1 (inadequate legislation), and R8 (inadequate supervision) lab center director are the most critical agents, while the secondary
have the top three sensitivity values among the 24 risk factors, faculty leadership team has an impact level close to that of the crit-
with R5 having the largest sensitivity value of 24.1%. This implies ical agents. This result actually corresponds to reality, as functional
that organizational climate, legislative system, and supervision departments usually perform very specific tasks. For example, the
have the greatest influence on unsafe behaviors, and regulators security department is responsible for fire safety in the laborato-
need to devote more energy to them. In fact, the results are in line ries, and the laboratory equipment and assets department is
with the reality of laboratory safety management in Chinese uni- responsible for all laboratory matters (including safety). The func-
versities. In addition, nodes with a higher degree of sensitivity, tional department represents university-level management, the
which should not be neglected, include R2 (design defects), R3 (so- secondary faculty leadership team represents faculty-level man-
cial factors), R17 (violations), R6 (resource management), R7 (super- agement, and the lab center director is the basic manager of the
vision violations), R10 (failure to correct problems), and R9 labs. Therefore, appropriate risk control measures are considered
(inappropriate operational plans). These factors involve external from the above three relevant agents.
factors and unsafe supervision. This result is basically consistent According to the results in Section 4.3, for the risk factor for
with the reverse inference result. poor organizational climate, the functional department should first

Table 9
Fuzzy CPT results for I2.

Environmental factors
R13 R14 Unacceptable Acceptable
Expert 1 Expert 2 Expert 3 Expert 4 Expert 1 Expert 2 Expert 3 Expert 4
Totally acceptable Good (0, 0.1, 0.2) (0, 0.1, 0.2) (0, 0.1, 0.2) (0, 0.1, 0.2) (0.8, 0.9, 1) (0.8, 0.9, 1) (0.8, 0.9, 1) (0.8, 0.9, 1)
Adverse (0.8, 0.9, 1) (0.8, 0.9, 1) (0.8, 0.9, 1) (0.8, 0.9, 1) (0, 0.1, 0.2) (0, 0.1, 0.2) (0, 0.1, 0.2) (0, 0.1, 0.2)
Partially acceptable Good (0, 0.1, 0.2) (0.2, 0.3, 0.4) (0.2, 0.3, 0.4) (0.4, 0.5, 0.6) (0.8, 0.9, 1) (0.8, 0.9, 1) (0.6, 0.7, 0.8) (0.6, 0.7, 0.8)
Adverse (0.8, 0.9, 1) (0.8, 0.9, 1) (0.8, 0.9, 1) (0.8, 0.9, 1) (0, 0.1, 0.2) (0, 0.1, 0.2) (0, 0.1, 0.2) (0, 0.1, 0.2)
Unacceptable Good (0.2, 0.3, 0.4) (0.2, 0.3, 0.4) (0.6, 0.7, 0.8) (0.4, 0.5, 0.6) (0.4, 0.5, 0.6) (0.4, 0.5, 0.6) (0.2, 0.3, 0.4) (0.4, 0.5, 0.6)
Adverse (0.8, 0.9, 1) (0.8, 0.9, 1) (0.8, 0.9, 1) (0.8, 0.9, 1) (0, 0.1, 0.2) (0, 0.1, 0.2) (0, 0.1, 0.2) (0, 0.1, 0.2)

Table 10
Fuzzy CPT results for I2.

Environmental factors
R13 R14 Fuzzification Defuzzification Normalization
Unacceptable Acceptable Unacceptable Acceptable Unacceptable Acceptable
Totally acceptable Good (0, 0.1, 0.2) (0.8, 0.9, 1) 0.1 0.9 0.1 0.9
Adverse (0.8, 0.9, 1) (0, 0.1, 0.2) 0.9 0.1 0.9 0.1
Partially acceptable Good (0,1773,0.2773,0.3773) (0.7136,0.8136,0.9136) 0.2773 0.8136 0.2542 0.7458
Adverse (0.8, 0.9, 1) (0, 0.1, 0.2) 0.9 0.1 0.9 0.1
Unacceptable Good (0.3364,0.4364,0.5364) (0.35,0.45,0.55) 0.4364 0.45 0.4923 0.5077
Adverse (0.8, 0.9, 1) (0, 0.1, 0.2) 0.9 0.1 0.9 0.1

22
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Fig. 4. Initial BN model for unsafe behavior in university laboratories.

propose laboratory safety regulations and then supervise their In terms of violations and errors, the laboratory may implement
implementation. In addition, safety education and training for a safety access system and strengthen the standardized manage-
teachers and students should be increased to raise the safety ment of personnel operations. Safety education content is neces-
awareness of the personnel involved. Wirth, Foreman, Friedel, sary to strengthen its relevance and warning, in order to improve
and Andrew (2020) have demonstrated the importance of provid- the professional knowledge and operational skills of operators.
ing adequate education and training to laboratory workers. This Finally, it is necessary to focus on the physical and mental health
is because a positive safety culture in the laboratory can create a of the experimental operators, including students and teachers.
positive safety climate. The secondary faculty leadership team For example, experimental tasks and breaks can be reasonably
should develop more detailed rules based on the school-level sys- arranged, or regular psychological guidance and regulation can
tem so that safety responsibilities can be put in place. The lab cen- be provided by the university’ medical department.
ter director is expected to promote and regularly check the
implementation of plans and systems and guide laboratories in
their safety management. To address the risk factor for inadequate 5. Research limitations and future prospects
supervision, key agents should strengthen the team at the univer-
sity level, the faculty level, and the laboratory center level. Job In this paper, a systematic risk assessment and control model of
safety training, performance assessment and rewards, and job unsafe behavior in university laboratories was proposed, and the
evaluation should be improved. In response to inadequate legisla- credibility of the model was well verified within a university
tion, the university should work with the education sector to chemistry laboratory, but there are still some limitations. First, it
improve the legal system from a national level. is unable to address the shortage of statistical data at the root.
In addition, other risk factors that cannot be ignored need to be Now there are no institutions that systematically collect and pub-
controlled as well. For social factors, the government and schools lish data on university laboratory accidents. Although this study
should pay real attention to laboratory safety management. uses typical accident cases from the past 20 years as a sample, it
Emphasis can be placed on policy development and financial sup- is still not an adequate sample size to draw more general conclu-
port. Meanwhile, safety knowledge should also be promoted to the sions that may be applicable to all types of university laboratories.
public in various forms to create a good social safety climate. To Second, the acquisition of conditional probabilities in the model
solve the problem of design flaws in the laboratory, laboratory construction process relies mainly on domain experts, which may
infrastructure construction and renovation can be carried out in limit the application of the model.
accordance with national standards and requirements according Nevertheless, the method can be extended to university labora-
to the scale of schools. Government regulators also need to do their tories of different scales, and is important for risk management of
part. In terms of resource management, the whole life cycle man- unsafe behaviors to a certain extent. Our subsequent research will
agement shall be conducted. A complete management chain needs focus on information acquisition in various university laboratories.
to be formed from material procurement to use and then to dis- And further optimize the research related to unsafe behavior by
posal, and a management dynamic ledger should be established. combining the application of machine learning (Peng et al., 2019)
23
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Fig. 5. Entire BN model for unsafe behavior in university chemistry laboratories.

Table 11 Table 12
The overall situation of four safety self-inspections of chemical laboratories in the Agents and related responsibilities.
past year.
Agents Responsibility
Problem category Number of Percentage of
Student Compliance with regulations and conducting
problems problems
experiments.
Safety education 3 0.7% Teacher Laboratory teaching management.
Safety responsibility system 8 1.9% Lab Safety Manager Overall inspection of safety work in all laboratories,
Regulations 6 1.4% guidance on the implementation of rules and
Experimental site 32 7.7% regulations, organization of safety training, stopping
environment and correcting violations.
Safety facilities 23 5.5% Laboratory Director Enforcement of rules and regulations, standard
Violations and errors 344 82.7% management and safety inspection of laboratory.
Laboratory Centre Assistance to the faculty-level leadership team in
Director safety management, focusing specifically on the
and artificial intelligence (You, Li, Li, & Xu, 2021) in the field of implementation of all laboratory safety.
safety management and risk assessment. Meanwhile, expert sys- Secondary Faculty Overall planning, organization, leadership and
tem technology is used to develop an expert evaluation knowledge Leadership Team supervision of laboratory safety throughout the
school.
base.
Functional Construction and implementation of a university-
Department wide laboratory safety management system, safety
culture and safety rules and regulations.
6. Conclusions School Leader General direction and coordination of laboratory
safety at the university level.
In this study, the HFACS model was improved by considering Ministry of Education Development of laboratory safety policy and
planning, guidance and supervision of safety in all
the actual situation in university laboratories, and the risk factors schools.
were identified by a detailed decomposition of the causes of
human errors at five levels: external influence, organizational
influence, unsafe supervision, preconditions for unsafe behavior, behavior in university laboratories was assessed in terms of three
and unsafe behavior. Furthermore, the BN method was introduced, aspects: causal induction, reverse inference, and sensitivity analy-
and the fuzzy BN model of unsafe behavior in university laborato- sis. Finally, according to the assessment results, evidence-based
ries was constructed, using expert experience and fuzzy theory in risk control measures were proposed from the perspective of key
conjunction with laboratory accident cases. The risk of unsafe agents.

24
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Fig. 6. Posterior probability distribution.

(a) (b)

Fig. 7. Results of reverse inference (a) when the L3 and L5 negative scenarios occur at 100% confluence and (b) when the L1, L3 and L5 negative scenarios occur at 100%.

25
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Fig. 8. Sensitivity analysis results.

Fig. 9. Influence matrix among agents and risk factors.

The paper applied the constructed risk assessment and control From the theoretical application, this paper combined HFACS
methodology to a university chemistry laboratory and calculated and BN methods to study unsafe behavior in university laborato-
the probability distribution of each node in the network model ries, broadening the research in the field of campus safety. The
using the fuzzy BN parameter learning function. The causal induc- HFACS-UL and fuzzy BN models are established and modified
tion results showed that there was an unacceptable probability of based on the reality of university laboratories. It provides a new
86% unsafe behavior in chemistry laboratories. Factors with a high perspective for exploring risk management in university laborato-
impact include inadequate legislation, social factors, poor organi- ries where little research has been done.
zational climate, inadequate supervision, poor personal prepared- Practically, the proposed risk management model for university
ness, and poor resource management; mitigation of these factors laboratories appears to be a reliable and useful tool. It not only
requires commitment and cooperation at all levels (state, school deepens the insight into the risk factors of unsafe behavior in uni-
executive leadership, teachers and students, etc.) to reduce unsafe versity laboratories, but its findings also provide a basis for devel-
behavior. The posterior probability results indicated that unsafe oping effective countermeasures. Therefore, this study is important
supervision is the most likely factor leading to unsafe behavior. for improving the efficiency of research and ensuring the safety of
The most likely causal chain for unsafe behavior was {inadequate laboratory personnel. It also brings some enlightenment to
legislation ? unacceptable external factors ? unsafe supervi- improve the safety management of university laboratories.
sion ? unsafe behavior}. Inadequate legislation and policies were
the most fundamental factors. In addition, the sensitivity analysis
results showed that poor organizational climate, inadequate legis- Acknowledgements
lation, and inadequate supervision have the greatest impact on
unsafe behavior and that regulators need to devote more effort This research was supported by the National Natural Science Foun-
to addressing these factors. To avoid unsafe behavior, risk control dation of China (No. 52074302); the Second Batch of New Engi-
measures are proposed for the functional department, lab center neering Research and Practice Projects of the Ministry of
director, and secondary faculty leadership team. Education (No. E-AQGABQ20202706); the Chinese Association of
26
Z. Li, X. Wang, S. Gong et al. Journal of Safety Research 82 (2022) 13–27

Higher Education Special Project on Laboratory Management mines. Resources Policy, 59, 210–216. https://doi.org/10.1016/j.
resourpol.2018.07.003.
Research (No. 2020SYD02).
Saaty, T. L., & Ozdemir, M. S. (2003). Why the magic number seven plus or minus
two. Mathematical and Computer Modelling, 38(3–4), 233–244. https://doi.org/
Potential conflicts of interest 10.1016/S0895-7177(03)90083-5.
Schröder, I., Huang, D. Y. Q., Ellis, O., Gibson, J. H., & Wayne, N. L. (2016). Laboratory
safety attitudes and practices: A comparison of academic, government, and
The authors have no conflicts of interest relevant to this article industry researchers. Journal of Chemical Health and Safety, 23, 12–23. https://
to disclose. doi.org/10.1021/acs.chas.8b23106.
Sevinc, V., Kucuk, O., & Goltas, M. (2020). A Bayesian network model for prediction
and analysis of possible forest fire causes. Forest Ecology and Management, 457.
References https://doi.org/10.1016/j.foreco.2019.117723 117723.
Shariff, A. M., & Norazahar, N. (2012). At-risk behaviour analysis and improvement
Banuls, V. A., Turoff, M., & Hiltz, S. R. (2013). Collaborative scenario modeling in study in an academic laboratory. Safety Science, 50(1), 29–38. https://doi.org/
emergency management through cross-impact. Technological Forecasting and 10.1016/j.ssci.2011.06.008.
Social Change, 80(9), 1756–1774. https://doi.org/10.1016/j. The U.S. Chemical Safety Board (CSB) (2018). CSB releases laboratory incident data
techfore.2012.11.007. (Jan. 2001 - Jul. 2018). https://www.csb.gov/assets/1/17/csb_laboratory_
Cai, B., Min, X., Liu, Y., & Feng, Q. (2018). Availability-based engineering resilience incident_data.pdf?16376 (accessed 30 May 2020).
metric and its corresponding evaluation methodology. Reliability Engineering & Walters, A., Lawrence, W., & Jalsa, N. K. (2017). Chemical laboratory safety
System Safety, 172, 216–224. https://doi.org/10.1016/j.ress.2017.12.021. awareness, attitudes and practices of tertiary students. Safety Science, 96,
Cattelani, L., Chesani, F., Palmerini, L., Palumbo, P., Chiari, L., & Bandinelli, S. (2020). 161–171. https://doi.org/10.1016/j.ssci.2017.03.017.
A rule-based framework for risk assessment in the health domain. International Wang, T., Gao, S., Liao, P., Ganbat, T., & Chen, J. (2019). A stakeholder-based risk
Journal of Approximate Reasoning, 119, 242–259. https://doi.org/10.1016/j. assessment and intervention framework for international construction projects:
ijar.2019.12.018. A meta-network perspective. International Journal of Managing Projects in
Chen, J., Zhong, P. A., An, R., Zhu, F., & Xu, B. (2019). Risk analysis for real-time flood Business, 14(2), 345–368. https://doi.org/10.1108/IJMPB-08-2018-0166.
control operation of a multi-reservoir system using a dynamic Bayesian Wiegmann, D. A., & Shappell, S. A. (2003). A human error approach to aviation
network. Environmental Modelling & Software, 111, 409–420. https://doi.org/ accident analysis. Farnham, UK: Ashgate Publishing Limited.
10.1016/j.envsoft.2018.10.007. Wirth, O., Foreman, A. M., Friedel, J. E., & Andrew, M. E. (2020). Two discrete choice
Dai, Y., Fu, J., Liu, H., & Tong, L. (2019). Application of HFACS and 24Model in experiments on laboratory safety decisions and practices. Journal of Safety
accident analysis of university laboratory. Experimental Technology and Research, 75, 99–110. https://doi.org/10.1016/j.jsr.2020.08.005.
Management, 36(10), 259–264. https://doi.org/10.16791/j.cnki.sjg.2019.10.063. Wu, J., Xu, S., Zhou, R., & Qin, Y. (2016). Scenario analysis of mine water inrush
Ding, L., Ji, J., & Khan, F. (2020). Combining uncertainty reasoning and deterministic hazard using Bayesian networks. Safety Science, 89, 231–239. https://doi.org/
modeling for risk analysis of fire-induced domino effects. Safety Science, 129. 10.1016/j.ssci.2016.06.013.
https://doi.org/10.1016/j.ssci.2020.104802 104802. Wu, T., Liu, C., & Lu, M. (2007). Safety climate in university and college laboratories:
Fu, J., Liu, H., & Liu, W. (2018). Cause analysis and administrative measures of fire Impact of organizational and individual factors. Journal of Safety Research, 38(1),
and explosion accidents in university laboratories. Journal of Jilin Institute of 91–102. https://doi.org/10.1016/j.jsr.2007.01.003.
Chemical Technology, 35(5), 87–92. https://doi.org/10.16039/j.cnki.cn22- Xia, N., Zou, P. X. W., Liu, X., Wang, X., & Zhu, R. (2018). A hybrid BN-HFACS model
1249.2018.05.021. for predicting safety performance in construction projects. Safety Science, 101,
Ho, C., & Chen, M. (2018). Risk assessment and quality improvement of liquid waste 332–343. https://doi.org/10.1016/j.ssci.2017.09.025.
management in Taiwan University chemical laboratories. Waste Management, Yang, Y., Reniers, G., Chen, G., & Goerlandt, F. (2019). A bibliometric review of
71, 578–588. https://doi.org/10.1016/j.wasman.2017.09.029. laboratory safety in universities. Safety Science, 120, 14–24. https://doi.org/
Jorgensen, E. F. (2017). Development and psychometric evaluation of the research 10.1016/j.ssci.2019.06.022.
laboratory safe behavior survey (RLSBS). Journal of Chemical Health and Safety, Yazdi, M., Daneshvar, S., & Setareh, H. (2017). An extension to fuzzy developed
24(5), 38–43. https://doi.org/10.1016/j.jchas.2017.01.005. failure mode and effects analysis (FDFMEA) application for aircraft landing
Kjærulff, U., & van der Gaag L. C. (2000). Making sensitivity analysis system. Safety Science, 98, 113–123. https://doi.org/10.1016/j.ssci.2017.06.009.
computationally efficient, in: Boutilier, G., Goldszmidt M. (Eds.), Proceedings You, M., Li, S., Li, D., & Xu, S. (2021). Applications of artificial intelligence for coal
of the Sixteenth Annual Conference on Uncertainty in Artificial Intelligence. mine gas risk assessment. Safety Science, 143. https://doi.org/10.1016/j.
Morgan Kaufmann, San Francisco, pp. 317-325. ssci.2021.105420 105420.
Laal, F., Pouyakian, M., Jafari, M. J., Nourai, F., Hosseini, A. A., & Khanteymoori, A. Yu, K., Cao, Q., Xie, C., Qu, N., & Zhou, L. (2019). Analysis of intervention strategies
(2020). Technical, human, and organizational factors affecting failures of for coal miners’ unsafe behaviors based on analytic network process and system
firefighting systems (FSs) of atmospheric storage tanks: Providing a risk dynamics. Safety Science, 118, 145–157. https://doi.org/10.1016/j.
assessment approach using Fuzzy Bayesian network (FBN) and content ssci.2019.05.002.
validity indicators. Journal of Loss Prevention in the Process Industries, 65. Zadeh, L. A. (1965). Fuzzy sets. Information and Control, 8(3), 338–353. https://doi.
https://doi.org/10.1016/j.jlp.2020.104157 104157. org/10.1016/S0019-9958(65)90241-X.
Li, M., Wang, H., Wang, D., Shao, Z., & He, S. (2020). Risk assessment of gas explosion Zhang, M., Zhang, D. i., Goerlandt, F., Yan, X., & Kujala, P. (2019). Use of HFACS and
in coal mines based on fuzzy AHP and Bayesian network. Process Safety and fault tree model for collision risk factors analysis of icebreaker assistance in ice-
Environmental Protection, 135, 207–218. https://doi.org/10.1016/j. covered waters. Safety Science, 111, 128–143. https://doi.org/10.1016/j.
psep.2020.01.003. ssci.2018.07.002.
Li, X., Zhang, Y., Abbassi, R., Yang, M., Zhang, R., & Chen, G. (2021). Dynamic Zhu, C., Tang, S., Li, Z., & Fang, X. (2020). Dynamic study of critical factors of
probability assessment of urban natural gas pipeline accidents considering explosion accident in laboratory based on FTA. Safety Science, 130. https://doi.
integrated external activities. Journal of Loss Prevention in the Process Industries, org/10.1016/j.ssci.2020.104877 104877.
69. https://doi.org/10.1016/j.jlp.2020.104388 104388.
Li, Y., Xu, D., & Shuai, J. (2020). Real-time risk analysis of road tanker containing Ruipeng Tong is currently a Professor at the School of Emergency Management and
flammable liquid based on fuzzy Bayesian network. Process Safety and Safety Engineering, China University of Mining and Technology-Beijing. His area of
Environmental Protection, 134, 36–46. https://doi.org/10.1016/j. expertise focuses on behavioral safety management, occupational health, accidents
psep.2019.11.033. analysis and prevention, and risk assessment.
Liu, C., Yang, S., Cui, Y., & Yang, Y. (2020). An improved risk assessment method
based on a comprehensive weighting algorithm in railway signaling safety Ziqi Li is a postgraduate at the School of Emergency Management and Safety
analysis. Safety Science, 128. https://doi.org/10.1016/j.ssci.2020.104768 104768. Engineering, China University of Mining and Technology-Beijing. Her research
Liu, R., Cheng, W., Yu, Y., Xu, Q., Jiang, A., & Lv, T. (2019). An impacting factors interests focus on behavior-based safety management and risk assessment and
analysis of miners’ unsafe acts based on HFACS-CM and SEM. Process Safety and management.
Environmental Protection, 122, 221–231. https://doi.org/10.1016/j.
psep.2018.12.007. Xiaolong Wang is a PhD student of Safety Science and Engineering at the School of
Olewski, T., & Snakard, M. (2017). Challenges in applying process safety Emergency Management and Safety Engineering, China University of Mining and
management at university laboratories. Journal of Loss Prevention in the Technology-Beijing. His research interests focus on accidents analysis and pre-
Process Industries, 49, 209–214. https://doi.org/10.1016/j.jlp.2017.06.013. vention, behavior-based safety management and risk assessment.
Omidvari, M., Mansouri, N., & Nouri, J. (2015). A pattern of fire risk assessment and
emergency management in educational center laboratories. Safety Science, 73, Shiji Gong is a postgraduate at the School of Emergency Management and Safety
34–42. https://doi.org/10.1016/j.ssci.2014.11.003. Engineering, China University of Mining and Technology-Beijing. His research
Pearl, J. (1986). Fusion, propagation, and structuring in belief networks. Amsterdam, interests focus on behavior-based safety management and accident analysis and
Holland: Elsevier Science Publishers. prevention.
Peng, T., Li, C., & Zhou, X. (2019). Application of machine learning to laboratory
safety management assessment. Safety Science, 120, 263–267. https://doi.org/ Ninghao Sun is a postgraduate at the School of Emergency Management and Safety
10.1016/j.ssci.2019.07.007. Engineering, China University of Mining and Technology-Beijing. His research
Qiao, W., Liu, Q., Li, X., Luo, X., & Wan, Y. (2018). Using data mining techniques to interests focus on behavior-based safety management and occupational health.
analyze the influencing factor of unsafe behaviors in Chinese underground coal

27

You might also like