Assessing the sustainable safety practices based on human behavior factors: an application to Chinese petrochemical industry

Sustainable and safe development is a key issue in petrochemical industry. However, many catastrophic accidents and irreversible environmental pollution have occurred in the petrochemical industry, most of which are caused by human behavior factor. The purpose of this paper is to evaluate and improve the sustainable safety performance of a petrochemical plant from the perspective of human behavior factors. Therefore, this study proposes an extended behavior-based safety (BBS) approach that combines the application of fuzzy analytic hierarchy process (FAHP) and fuzzy comprehensive evaluation (FCE) to assess sustainable safety performance of a petrochemical plant. The results show the importance of human behavior factors, and we should pay attention to the practical application of petrochemical industry. Finally, BBS management can reduce occupational injuries and accidents, which will enhance the sustainable safety development of the petrochemical industry.


Introduction
Petrochemical industry contains a large number of flammables, corrosive, toxic and explosive materials (Rollinson 2018). In the petrochemical industry, unsafe production may lead to environmental pollution, energy waste, climate change and oil price change, which are regarded as a great threat to sustainable safety development (Sharma et al. 2017). However, the importance of human factors has been fully established in the causes of occupational injuries and accidents (Baysari et al. 2008; Kelly and Efthymiou 2019; Karthick et al. 2020). One effective approach used to improve unsafe acts of persons is behavior-based safety (BBS) (Geller 2005;Navidian et al. 2015). There is no agreed concept of BBS, but it is usually considered as an approach aiming at safety intervening and modifying unsafe acts of persons (Geller 1999;Geller 2002). BBS has been successfully applied in various areas in North America, Asia and Europe for over the past years, such as mining (Hagge et al. 2017), the construction industry (Lee et al. 2019), robots (Scianca et al. 2021, the E-waste collection (Batoo et al. 2021) and the vehicle industry (Wang et al. 2018). The above researches demonstrate the effectiveness of BBS management on improving unsafe behaviors and sustainable safety performance. These researches provide support for the concentrate on participation in a BBS management. Furthermore, BBS management provides a procedural approach to construct sustainable safety performance gains and process control.
Although BBS has many successful cases, others point out that there is still room for improvement (Alkaissy et al. 2020). For instance, in the petrochemical and petroleum industry, the quantitative assessment and analysis of the behavior of employees is still scarcely considered. In order to reduce the occasional industry accidents, the study focused on the behavioral and assessment side of the employees, designed and implemented BBS program, which starts from the understanding and identification of potential unsafe behavior to assess the employees' safety performance (Norton et al. 2017). Therefore, to preferably solve the accidents and occupational injuries, the combination of a fuzzy analytic hierarchy process (FAHP), a BBS and fuzzy comprehensive evaluation (FCE) should be proposed. There are several reasons that we apply the combined method.
Analytic hierarchy process (AHP) is a decision-making method to analyze and solve complex problems based on mathematics and psychology (Nguyen et al. 2015). However, the determinants of sustainable safety performance evaluation contain intangible side of precise data. It becomes difficult for human to immediately describe and evaluate sustainable safety performance. Since people and preference judgments are usually vague and uncertain, they cannot assess their preference with an accurate numerical value (Pramanik et al. 2020). Additionally, the evaluating factors involve complex and vague internal relationships, leading to a fuzzy evaluation result (Mohsenzadeh et al. 2019). As a result, it is necessary to utilize a hybrid method to evaluate the sustainable safety performance involving linguistic variables (Ma et al. 2021). Fuzzy set is an effective and appropriate tool to tackle uncertain problems, while AHP, developed by Saaty, is a reasonable decision-making approach (Saaty 1980). As an extension of AHP, FAHP can solve the problem of fuzzy multi-criteria decision-making and it has been widely applied in various fields, such as project prioritization and selection (Shaygan and Testik 2019;Wang et al. 2019), capital investment (Khashei-Siuki and Sharifan 2020; Chien et al. 2021), brand preference , three-dimensional printers (Chen and Wu 2021), sustainable production and consumption (Shete et al. 2020;Goyal et al. 2021), manufacturing service (Gul et al. 2018;Hu et al. 2021), oil industry (Iqbal et al. 2021) and city construction (Lyu et al. 2020). So, the FAHP is appropriate for calculating the factors weights in the assessment model. To improve employees' sustainable safety performance, this paper develops a fuzzy management behavior-based assessment model by using FAHP, BBS and FCE (Chang 1992;Chang 1996).
The rest of this paper is organized as follows. In Section 2, we describe literature review and research gap. In Section 3 and 4, we describe the related definition of FAHP and FCE, respectively. In Section 5, taking a petrochemical plant as an example, the combined application of the FAHP, FCE and BBS management is explained. The main discussions are drawn in Section 6. The conclusions and limitations of the study are given in Sections 7 and 8, respectively.

Literature review and research gap
The petrochemical industry can be regarded as one of the most important aspects of sustainable and safe development because of its importance in economy, quality of life, society and environment. However, behavior-based safety management system has been proved to be a neglected field and has not been systematically applied to the petrochemical industry. In recent years, the application of technical tools such as human factors in sustainability safety assessment has attracted great attention from academia, policymakers, government agencies and major stakeholders in the chemical industry. In addition to industry differences, the design of sustainability safety indicators has been criticized for ignoring human factors at an early stage. The implementation of automation, information technology and artificial intelligence for safety in the petrochemical industry (quality monitoring, autonomous oil equipment, high-precision positioning, drilling control, etc.) has also increased the complexity of sustainable safety.
Despite the introduction of the state-of-the-art safety equipment, human behavior factors are still the main causes of accidents in petrochemical companies. The fundamental cause of occupational injuries is human factors, which are caused by complexity and the result of interaction between some factors (people and people, people and information). Automation system conveys some opaque information, involving many control rules and new information technologies. These interactions have produced the opacity and uncertainty of information in the system, which makes operators rely on indirect information and reduces the ability to deal with emergencies and to take further action. In addition, the complexity of enterprises and various work pressures ignore human behavior factors, and senior managers have to deal with various safety management pressures almost every day. This pressure has produced a strong desire to solve the problem in the shortest time at all costs. Managers tend to develop a guiding management style that neither accepts workers' opinions nor any corrective action. With the emergence of new technologies and complex relationships within and between organizations, sustainable and safe development is also becoming more and more complicated.
Nowadays, the chemical industry has become a very complicated system. In the complex and pressure environment (technology, environment, politics, industry and economy), modern companies must pay attention to human factors and meet the challenges of sustainable and safe development. This paper has two purposes: firstly, to analyze the role of the human factors approach in sustainable safety management systems, and secondly, to demonstrate the application of this approach to safety assessment in petrochemical companies. Finally, we hope that approaches based on human behavioral factors will contribute to the solution of occupational safety and sustainable and safe development, which is one of the major complex issues facing the modern chemical industry at present.

Fuzzy analytic hierarchy process
In this section, we introduce some definition and operational laws correlated with triangular fuzzy number (TFN).

Triangular fuzzy numbers
A fuzzy number A on R to be TFN if its membership function ∼ A (x) : R → [0,1] is equal to Eq. (1) (Chang 1996) where l≤m ≤ u, l and u can be expressed as the lower and upper values of the, respectively, support of ∼ A , and m for the modal value (as Fig. 1). When l = m = u, the TFN becomes a non-fuzzy number.

The operational laws of TFN
Let ∼ A = (l 1, m 1, u 1 ) and ∼ B = (l 2, m 2, u 2 ) be two TFN, then their operations are shown as follows (Kauffman and Gupta 1991):

The extent analysis FAHP method
The process of Chang's FAHP can be discussed as given below, and the fuzzy pairwise comparison matrix Ã = ã ij n×n can be mathematically defined as follows: where For S i , the fuzzy synthetic extent value, with respect to the ith object, is expressed as with The degree of possibility of S j = (l j, u j, m j ) ⩾ S i = (l i, u i, m i ) is denoted as (6) A −1 = l 1 , m 1 , u 1 −1 = l 1 ∕u 2 , m 1 ∕m 2 , u 1 ∕u 2  (l, m, u) To compare between S j and S i , it is required to calculate both V (S i ≥ S j ) and V (S j ≥ S i ). The minimum degree of possibility d(i) of V (S j ≥ S i ) for i, j = 1, 2, . . ., k is calculated as follows (see Fig. 2): If assumedd(A i ) = min V(S ≥ S i ), fori = 1, 2, , ⋯, k,then the weight vector is given as Here, A i (i = 1, 2, …, n)are n elements. The weight vectors are obtained by normalizing as follows: where W is a real number.

Questionnaire design and linguistic scales
In this study, TFN expresses a subjective pairwise comparison of decision-makers. The linguistic terms and graphical representation of triangular linguistic labels are shown in Table 1 and Fig. 3 (Kahraman et al. 2006).
Namely, the importance of one factor over another is divided into six levels.

Establishing the comparison matrix of the decision-makers
Integration of the opinions of multiple decision-makers is one of the characteristics of the AHP. In the application of this method, geometric mean operation is commonly used to integrate group opinions, and this operation satisfies Pareto principle (unanimity condition) and homogeneity condition (Chen et al. 2015). This study is composed of K decisionmakers (experts), who construct the group comparison matrix by comparing n factors in pairs. In the group comparison matrix, the TFNs can be expressed as As a result of these comparisons, a series of k matrices, ∼ A k = (ã ijk ) , can be obtained, where ã ijk = (l ij, m ij, u ij ) denotes an importance of factor i to j relatively, as evaluated by the decision-maker k.

The consistency tests
In order to obtain rational assessment results, the consistency must be tested. Consequently, a fuzzy comparison matrix is transformed into a crisp matrix. In this study, Chang's approach is used to defuzzify a fuzzy number and the fuzzy perception can be expressed reasonably (Chang et al. 2009). A TFN represented as ã ij = (l ij, m ij, u ij ) can be defuzzified to a crisp number as follows: where l ij = m ij − l ij × + l ij is interpreted as the left-end value of α-cut for a ij . Conversely, u ij = u ij − u ij − m ij × is interpreted as the right-end value of α-cut. λ indicates the degree of optimism among decision-makers, which is any numerical value from 0 to 1 [52]. When λ is equal to 0, the degree of optimism is maximum among decision-makers. Conversely, when λ is equal to 1, a decision-maker is highly pessimistic. Additionally, the decision-making circumstances fluctuate if α decreases. When α = 0, the degree of uncertainty is the maximum. Therefore, the above formula can be described as a fluctuating or stable condition, and its range is from 0 to 1. In a comparison matrix, all the factors of assessment model can be transformed from TFNs to crisp numbers (Hsu et al. 2016), which is shown as follows: where superscripts α and λ are for comparisons based on Eq. (19).
In order to check the consistency of the comparison matrix, the consistency index (CI) and the consistency ratio (CR) are given as [34] (19) Here, RI is the random index and λ max denotes the maximum eigenvalue. As shown in Table 2, if CR ≤ 0.1, the consistency of the judgment matrix is in the acceptable range.

Fuzzy comprehensive evaluation method
FCE is an application of the fuzzy set theory. FCE is a quantitative, reasonable and objective evaluation method, suggested by Zadeh (Zadeh 1965;Zadeh 1978). Although FCE has been widely carried out in many areas, there are few utilizations of FCE in the field of the BBS management. In this paper, we used FCE as a tool for sustainable safety performance assessment and its application steps are as follows:

Evaluation factor set
where m denotes the number of criteria layers and p is the number of sub-criteria layers.

Determining a fuzzy relationship matrix
In the evaluation factor set, a single factor is evaluated to determine its membership degree. The fuzzy relation matrix can be expressed as follows: where r ij is the fuzzy membership degree result of the ith factor belonging to the jth rank.

Calculating fuzzy comprehensive evaluation results
As described in section 4, the weight set W is obtained by using FAHP. Therefore, the FCE model can be established based on weight set W and fuzzy relationship matrix R as follows: The results of the FCE can be obtained based on the maximum membership principle.

The proposed framework
BBS management is mainly used to improve the safety behaviors performance of employees. However, in the lack of a large amount of data, it is necessary to integrate expert judgments of experience into the safety assessment. In order to evaluate the safety of complex management system, a flexible and comprehensive technique method needs to be provided. Therefore, our proposed method integrates the FAHP, multi-criteria technique, fuzzy comprehensive evaluation with BBS management. This method integrates the group consistent decision-making principle, which can not only select the suitable alternative but also scientifically quantify the sustainable safety performance of employees. In Fig. 4, we describe a flowchart of the assessment framework and safety behaviors analysis based on the combined application of FAHP, FCE and BBS management. As illustrated, FAHP, FCE and BBS processes are combined as a part of the safety management. The colored module shows the application of FAHP and FCE quantified safety behavior, which is implemented by the combination assessment of these two approaches.
More specifically, the FAHP is used for pairwise comparisons which calculate the local and global priority values and prioritize the safety behaviors factors. Next, the evaluation grade is calculated, and the conclusion of performance of employees' safety behavior is obtained by combining qualitative analysis and quantitative methods. According to the application of BBS program, the safety behavior of employees is guided and observed and intervenes in the chemic plant. Finally, after a period of BBS process, the assessment grade is again calculated about sustainable safety performance.

Application of proposed research framework
The mathematical model has been applied to a state-owned petrochemical plant in China. This petrochemical plant includes an administrative leading department, facility management, petroleum processing department, quality control department, oil and gas gathering team, oil refining department, information management center, etc. Analyzing the occupational injury statistics of this company, it is found that the accidents are mainly caused by human factors. Detailed analysis shows that about 90% of industry accidents are caused by employees' unsafe behaviors. In the high-risk operation system, it is of great significant to determine the safety behavior factors and take necessary preventive measures to reduce the accidents occupational injuries. Furthermore, with the help of an experienced scholar and administrator, five groups of 30 experienced managers, supervisor, scholars, engineers and employees have been discussed and adjusted the safety behavior factors of petrochemical plant. In consequence, the model is built by five group of experts as shown in Fig. 5.
As shown in Fig. 5, the evaluation index system consists of three levels. The first level is the target layer. The second level is the criterion layer, which includes six factors (U 1 -U 6 ), such as employees' psychological behavior (U 1 ), process safety education (U 2 ), working atmosphere (U 3 ), management factors (U 4 ), equipment factors (U 5 ) and employees' comprehensive quality (U 6 ). The third level is the sub-criterion layer including 19 sub-factors (U 11 -U 63 ).

Weights are calculated using FAHP
To calculate the criterion weights, five groups of 30 experts are designated. The experts were organized for an indepth interview and asked to compare six criteria and 19 sub-criteria in the evaluation model. Pairwise comparisons, sourced from the expert evaluations on the importance of one criterion over another relatively, are used to establish the comparison matrix of every expert.
The geometric mean operation is utilized to obtain the representative comparison matrix of each expert group by Eq. (18). Three groups of factors of representative comparison matrix collected from the expert group are, respectively, shown in Table 3.
According to the environmental and conditional uncertainty, the experts can decide the α-cut subjectivity. In this paper, α = 0.5 is used to express that conditional and environmental uncertainty is stabilized, and λ = 0.5 denotes that the attitude of the experts is fair and rational. When α = 0.5 and λ = 0.5, the defuzzification given in Table 3 is expressed as follows: Similar to the calculation above, the crisp comparison matrix of all factors is obtained in Table 4. The next step is to calculate the consistency of the crisp comparison matrix. Firstly, by using Eqs. (21) and (22), CR = 0.0108 < 0.1. This shows that the crisp comparison matrix satisfies consistency and the weight distribution is reasonable.
The flowchart of the proposed assessment framework   We use Chang's extent analysis method and the fuzzy evaluation matrix of six factors (U 1 -U 6 ) in Table 3 to calculate the factors weights. Using Eq. (10) -(14) to calculate the value of fuzzy synthetic extent, the factors and sub-factors' weights are computed by the FAHP. Taking the factor weights of the pairwise comparison matrix in Table 3 as an example, the value of fuzzy synthetic extent is calculated as follows: The degree of possibility of every factor over the others is calculated by using Eq. (15), whose results are shown as follows:    Tables 5, 6 , 7, 8, 9, and 10.
By a similar operation, the weight vectors of sub-factors were determined. Finally, for all of the CR ≤0.1, the consistency in each crisp comparison matrix is accepted.

The BBS program for a petrochemical plant
As more and more managers realize the importance of BBS process, many Chinese companies have also implemented BBS procedures in their management process (Chen and Tian 2012;Zhang et al. 2017;Yu and Li 2019). Hence, to better understand the BBS practices to Chinese company, this paper selected a state-owned petrochemical plant in the east of China as a study case. The specific BBS program is shown as follows (Fig. 6): As illustrated, there are five steps in the BBS program, including the preparation period, design period, implementation period, intervention period and follow-up period. In addition, the detailed process includes expert interview, questionnaire design, collection of data and assessment of sustainable safety performance and so on.

Preparation Period
Firstly, in the BBS process, the first step is to assess the sustainable safety performance of employees, understand the current status of the petrochemical plant and review the existing organizational structure, plant history, personnel, and existing safety management system. Therefore, the existing employees of the petrochemical plant were surveyed using questionnaire on the sustainable safety performance, as shown in Fig. 5. Likert five-point scale is used to collect the questionnaire data (Likert 1932). In the first survey, a total of 824 questionnaires were received and 665 were available. Based on the questionnaire data, the fuzzy relationship matrix is obtained by using Eq. (24): where RA 1 , RA 2 , … , RA 6 denote the fuzzy relationship matrix of the first assessment. Taking the fuzzy relationship matrix RA 1 as an example, when "Corporate frequency and intensity of safety training" was considered, 64.51% of employees rated it "very high," 31.88% of employees rated it "high," 3.46% rated it "medium," 0% rated it "low" and 0.15% rated it "very low." Then, the first-layer fuzzy comprehensive evaluation result is obtained based on Eq. (26): Similar to the FA 1 calculation, we get According to Eq. (26) and the factors weights value, the assessment result set of second-layer fuzzy comprehensive is where B 1 is the fuzzy comprehensive results of the first evaluation. The results reveal that the probability of the sustainable safety performance very high is 0.3729, the probability of "high," "medium," "low" and "very low" is 0.4744, 0.1192, 0.0203 and 0.0131, respectively. The result shows that the sustainable safety performance is assessed as high based on the maximum membership principle. However, the employees' sustainable safety performance can be further improved in the future.

Design period
After the first safety evaluation was accomplished, a total of 30 experts were invited to the formation of a design team, including safety professionals, safety scholars, safety managers, administrative staff and supervisors. The design team participated in a two-day training meet to learn the basic theory of BBS management. Then, they met and discussed their idea at the petrochemical plant and developed the safety   Table 11. Five different behavior factors (coded as SO 1 -SO 5 ) are used to observe the safety behaviors performance of on-site employees, including personal protective equipment (SO 1 ), traffic management (SO 2 ), psychological performance (SO 3 ), facilities maintenance (SO 4 ) and daily performance (SO 5 ). Based on the above factors, the checklist includes 23 typical employees' behaviors, which can be classified by SO 1 -SO 5 .

Implementation period
The design team elect 35 experienced employees to set up the observation team. The role of the observation team is to maintain and monitor possible incidents during the BBS process. The observation team received a two-day safety training (ST) course in the fundamental principle and practice of BBS management. The training course included decisionmaking, unsafe behavior modification, how to manage others' resistance, observational and communicational skills, and providing personal feedback and effective checklist scoring (Dağdeviren and Yüksel 2008;Choudhry 2014;Wang et al. 2017). Safety training is shown in Table 12. Table 12 shows that the four-module course (coded as ST 1 -ST 4 ) is designed to train employees in the petrochemical plant. After the training period, the design team held a plant-wide launch activity to announce the BBS process. Everyone attends a one-day training meeting on guiding observations and providing feedback. Then, everyone was trained on how to make observations and convey feedback to their colleagues. Various training meetings ensure that each employee knows their role in the BBS process. When the BBS program was launched, employees began using both checklists for peer and self-observations.

Administrative leadership system
Administrative leadership system is one of the characteristics of China's state-owned company, and it can improve the willingness that everyone will accept the safety intervention, participate in observation and provide individual feedback. The administrative leaders require managers and supervisors to regularly check the employee discipline, the participation in the BBS process and maintenance of operation tools and equipment. Some leaders are also required to monitor the completion of the BBS process. The BBS management implemented shows that employees need support from their administrative leadership. In addition, these measures solve problems such as lack of communication in BBS process, low conference attendance and participation in observation (Brandhorst and Kluge 2021).

Observation
To ensure authentic and objective safety participation for employees, all observations are voluntary. Everyone is invited to participate in peer safety observations or selfobservations about 14 days. In addition, the BBS process requires each employee to complete two observations per day. The supervisors randomly observe the monitoring video of each employee every day for about 10-15 minutes. Supervisors and managers encourage their employees to spend the time to fulfill the observation, which can either be peer safety observations or self-observations. In every observation, the observers need to check all items in the safety behavior checklist and collect a behavior sample from each on-site employee to confirm whether it is safe or not. One receives a mental or a material reward for positively participating and supporting the BBS process, such as appreciation from a supervisor.

Feedback
BBS process studies have shown that safety behavior improves when management provides clear feedback on employee observation information (Favero et al. 2016). Therefore, safety observation team inspects information on  participation and observation every day, such as use video monitoring to recognize the unsafe behavior or potential hazards, by investigating the situation on the ground and working with the corresponding management to deal with any observed unsafe behavior or equipment problems. Furthermore, an observation report meeting is held twice a day: One is before the employees start work, and the other is before the employees leave work. The observation results are given as feedback to the employees on site during the report meeting on every day. Five working groups were identified as the highest rate of unsafe behavior. Employees of the five groups were asked to attend specific training courses such as watch accident videos. The goal of training courses is to improve their safety behavior and safety awareness, not punish these employees.

Follow-Up Period
At the end of the intervention phase, the questionnaires used in the preparation period were again distributed to the employees to evaluate the changes of their safety behaviors. During this period, employees' behaviors were continuously observed and intervened. In the second survey, a total of 853 questionnaires have been collected, among which 726 replies were usable, with an effective reply rate of 85.11%.    Worker takes the correct operation posture Workers use appropriate access and egress Workers are not allowed to discard anything from a height A rough assessment of the safety behavior of workers on site Improve traffic safety Fill out work permits correctly Improve the quality of work Improve traffic safety Safety interaction (ST 4 ) Increase communication with other sites Improve interaction between employees Ask safety questions on site Shaping safety behavior By using Eq. (26), the factors weights value, the result set of second-layer comprehensive evaluation is as follows: where B 2 represents the fuzzy comprehensive results of the second evaluation.

Comparative analysis of FCE results following BBS intervention
The result of the second evaluation reveals that the probability of the sustainable safety performance of very high is 0.5030; the probability of "high," "medium," "low" and "very low" is 0.4302, 0.0644, 0.0011 and 0.0011. The result showing the sustainable safety performance is assessed as very high. The twice assessment results are shown in Fig. 7. (0.1764, 0.1714, 0.1685, 0.1627, 0.1651, 0.1559 Figure 7 illustrates that the sustainable safety performance of the second assessment is better than that of the first. During the implementation of BBS, there are no examples of rejecting safety interventions. This intervention has a significant influence on the management systems, raising safety climate and increasing safety communication. The results show that BBS is a sustainable safety management. In this intervention process, it is not necessary to stop working, but to take measures to correct employees' unsafe acts. The fact is that unsafe acts can be regarded as a social and psychological phenomenon. Especially in recent years, evaluation methods in group psychology and cognitive psychology have become more and more popular (Garzón 2008). Previous studies have shown that it is feasible to combine management, mathematics and psychology and that interdisciplinary insights can be obtained. The advantage of fuzzy mathematics is that it can integrate human thinking and objective data into the process of cognitive psychology. This integration can more richly describe human factors, sustainable safety development and BBS management. Previous safety management studies usually emphasized external factors and rarely considered the possibility of cognitive psychological effects. In addition to the external factors, BBS intervention is mainly affected by internal psychological factors, such as value congruence, psychological identity and cognitive process. Psychological identity is closely related to safety climate, safety communication and safety culture. This means that BBS intervention is more likely to succeed in a plant with a positive safety climate. Value consistency is likely to lead to satisfaction and recognition, so as to improve sustainable safety performance. However, the workforce is so diverse that workers, managers and supervisors are not also always consistent in the implementation of the BBS intervention. Therefore, employees may be too hasty to perfunctory BBS intervention. On the other hand, in the process of BBS intervention, safety ability is a key regulatory factor between safety behavior and sustainable safety performance. This means that unsafe act is not only a problem of psychological motivation, but may also be a problem of ability. There are some activity-based training programs to help employees improve their safety ability. As mentioned above, BBS intervention and psychological identity are two complementary methods to improve unsafe act. Many factors in the evaluation model (Fig. 5) are related to the dimension of psychological identity. Therefore, psychological identity also is the supplement of BBS intervention and explains the mixed effect from the perspective of psychology.
With the improvement in psychological identity, employees experience fewer conflicts and are more willing to cooperate with the BBS intervention. Thus, managers are motivated to pursue higher safety performance because it is in line with their psychological identity and personal interests. Employees are more driven by internal psychological identity than external punishment or reward. Psychological identity is relatively stable. Therefore, when the BBS program is implemented in plant with consistent psychological identity, its safety performance is more likely to be sustainable. In addition, the consistency of psychological identity can also provide as a feedback positive group norms. Positive group norms can reduce work conflicts and violations because all employees expect to work continuously and safely under occupational pressure (Goh et al. 2015). To sum up, psychological identity, willingness to pursue sustainable safety performance, positive group norms and human factors are crucial to the success of BBS intervention.  Figures 8 and 9 show the local and global weights of the research factors and sub-factors obtained by applying the FAHP techniques. Obviously, each factor with higher weight value is more important than the decision-making process.

Weight analysis
The global weights of all sub-factors are the results of multiplying the local weights of all sub-factors with the weights of the factor to which it belongs. In Fig. 8, we rank the local weights of all factors from large to small. We realize that U 1 (employees' psychological behavior) is the highest percent of effect on the sustainable safety performance rank among all factors (U 1 ranks highest in the local weight), followed by U 2 (process safety education), U 3 (working atmosphere), U 5 (equipment factors), U 4 (management factors) and U 6 (employees' comprehensive quality). Thus, U 1 (employees' psychological behavior) and U 2 (process safety education) are the key behavior factors that must be considered for the employees. This means that the safety managers should mainly pay attention to the psychological and educational issues. Psychological pressure may have negative effects on an individual physical and mental health, such as increased distractibility, lower concentration, and more prone to burnout. When employees experience general psychological pressure, they are unlikely to initiate accident reporting or safely use equipment. If employees face greater psychological pressure, the plant will face more lost working days, absenteeism and lower safety performance.
In addition, U 2 (process safety education) is also considered to be one of the most important factors for the practice of sustainable safety development in the chemical industry. Process safety education is generally defined as learning safety principles and operating disciplines to prevent major accidents and casualties in the process industry (Mkpat et al. 2018). Process safety education aims to improve the understanding of process safety principles, promote safety knowledge sharing and improve technical level (Nesheim and Gressgård 2014). In addition to improving safety performance, process safety education indirectly promotes many areas of safety culture. It sustains industry reliability, improves productivity and enhances the sustainable safety development of the chemical industry. Consequently, these plants effectively convey safety information through meetings and training, quickly solve safety problems and regard safety training as an investment. It is also observed that safety climate significantly affected sustainable safety performance. The traditional safety climate only focuses on the physical aspects of safety, and now it has been extended to the psychological safety atmosphere, which focuses on the psychological aspects of health and safety (Yaris et al. 2020). A positive safety climate gives priority to psychological safety, encourages psychological-oriented safety behavior and pays attention to sustainable safety performance. Hence, these three factors are considered to be the key to achieving sustainable safety development.
In Fig. 9, U 21 (frequency and intensity of safety training) is considered to be the most influential factor in the global weight part, followed by U 12 (personal psychological quality), U 31 (performance of workers in daily safety behavior) and U 11 (personal attention at work). Conversely, U 44 (corporate reward and punishment of safety production), U 42 (improvement degree of safety regulations), U 43 (corporate supervision of safety production) and U 41 (improvement degree of safety production plans) have less influence. U 21 (Safety training) is actually formed and developed for major accidents in the chemical process industry. Accidents in this dangerous industry can have serious consequences for the personnel, surrounding environment and plant assets. In terms of preventing major incidents, safety training has developed various tools, approaches and procedures aimed at removing human errors and technical design defect, as well as safety management systems. The dangers of the chemical industry led to complexity tasks, including the rate of information change, information diversity and increase in information volume. These factors demand the employees to make greater efforts that lead to greater stress and may lead to burnout, unsafe acts, fatigue and incidents. For U 12 (personal psychological quality) and U 31 (performance of workers in daily safety behavior), high psychological quality and safety behavior can hold a positive attitude toward stress, and employees will not experience any negative results.
Furthermore, according to the idea of self-regulation theory, for U 11 (personal attention at work), attention mechanism and behavior process can guide individuals to develop toward the safe direction of goal setting. For U 44 , Komaki et al. (1978) proposed a method of rewarding employees to reduce unsafe acts in the industry. Rewards and punishments create a psychological environment that encourages employees to modify their target behavior. However, this theory does not explain why rewards or punishments modify a behavior and what makes it work. Among U 42 (improvement degree of safety regulations), U 43 (corporate supervision of safety production) and U 41 (improvement degree of safety production

Analysis of the recordable incidents in UK
The incidence of manufacturing-related accidents and injuries also remains high. Therefore, in any realistic prospect of safety intervention in this complex industry, the understanding of the root cause of the accident is still indispensable. Pickup et al. (2020) uses the personal diary method to record and analyze real-time data of safety incidents (Pickup et al. 2020). This study qualitatively explores the safety-related events recorded by employees for the first time, so as to identify the perceived potential risk and find human errors. Taking a car manufacturing site in the UK as an example, Pickup et al. recorded and analyzed 176 incidents and classified them according to the event type in Table 13. The most common type of incident is unsafe acts, accounting for 38.07%, followed by unsafe conditions Fig. 10 Distribution of initial causes for each event (Dakkoune et al. 2018) (28.41%). The proportion of near misses (N=46) is high compared with accidents (N=13), accounting for 26% of the data, and exceeds the official data reported in the same time period (N=12 and N=3, respectively). Therefore, exploring the causes of accidents is still an indispensable part of injury reduction and safety improvement, especially within complex manufacturing and chemical industries. This study reveals that a more comprehensive taxonomy and model need to be integrated into accident investigation and analysis, so as to explore the diversity of human error and cognitive performance at the personal level. Meanwhile, human factors describe the correlation between safety climate and unsafe act, including safety culture and manager quality. Consequently, the study makes BBS intervention possible in daily unsafe acts, thus clarifying a new view of human factors at the system and individual levels.

Analysis of the recordable accident data in France
Accidents can cause disastrous damage to environment, human health and economy in the chemical industry. In order to prevent such incidents in the industry, it is necessary to review and analyze the past accident data. ARIA (Analysis, Research and Information on Accidents) database is considered to be one of the important technical accident databases in Europe. ARIA is a huge database managed and maintained by the French Ministry of ecology, sustainable development and energy. The database records more than 43000 accidents occurred in the world and in France. Dakkoune et al. (2018) collected and selected 169 safetyrelated events in the French database ARIA. These safetyrelated events occurred between 1974 and 2014. Dakkoune et al. (2018) analyzed the causes and consequences of these events. According to the type of event and the cause of the event, the risk distribution is shown in Fig. 10.
Obviously, the main initial cause of events in the chemical industry sector is operator error (about 40% of events). The other initial causes contain technical and physical causes, as well as human and organizational causes, which are, respectively, classified as the following uncontrolled or unexpected reactions, technical failure, insufficient risk analysis and corrosion, etc. In natural causes, most risks are less than 4%, although natural phenomena are often difficult to predict and generally destructive.
The chemical industry needs to pay more attention to and prevent these risks related to human factors. Because human behavior is complex and uncontrollable, it interacts with other external factors, such as equipment, colleagues, management and environment. In order to reduce accidents and injuries related to human factors, the research of approaches to improve the unsafe act of employees is crucial. In fact, it is difficult to design and apply digital systems to eliminate all human errors in daily work. On the other hand, a BBS intervention process needs to be proposed to identify and improve unsafe behaviors. For the government, this BBS approach can also be used as a starting point for developing safety management strategy, in order to prevent and reduce the number of safety-related events in the chemical industry.

Conclusions
This paper proposes a comprehensive sustainable safety performance assessment framework based on BBS management, FAHP and FCE, which included 6 factors and 19 subfactors. The evaluation results further show the importance of BBS management to the implementation and achievement of sustainable safety development in the petrochemical plant. The results of this study can be concluded as follows: (1) Due to its unique complexity and danger, petrochemical industry is the most prone to major accidents in the world, so it is urgent to improve the performance of employees' safety behavior. However, the traditional BBS research mainly focuses on the behavior analysis after the occurrence of an accidents, because there is no tool to collect employees' behavior data from on site.
In addition, the essence of employees' safety behavior evaluation involves many processes in the petrochemical plant. Therefore, we proposed the model overcome this limitation and it can record, monitor and assess employees' safety behaviors. In practical application, with the collection of behavioral data, the model is used to quantitatively evaluate the sustainable safety performance of employees.
(2) The model systematically combines the knowledge and experience of the expert team to calculate the weight of safety behavior factors. Weight ranking can also provide safety strategy to managers in a reasonable, scientific and effective manner. In addition, in order to reduce industry accidents, the petrochemical plant is becoming more and more employee-centered and strive to improve their sustainable safety performance.
In conclusion, the model determines the sustainable safety performance of employees based on calculated weight and obtained data, and its data sources are both objective and subjective. These evaluation results can help employees realize that safety is everyone's responsibility and safety is above all else. (3) To prevent potential risky behaviors of employees, this paper applies a more professional and systematic approach to assess sustainable safety performance of employees. BBS management will ensure that employees have adequate control and improved of safety behavior and psychological health. The BBS process emphasizes the operations management, planned main-tenance, correct use of equipment and risk analysis. This requires to develop guidelines for safety behavior, creation of unified teams with opened safety communication and shared responsibility for employee safety, and leaders and managers to consider the BBS program as a benefit rather than a burden.

Limitations of this study
Due to state-owned company and policy constraints, the intervene time is only 14 days. The assessment data and frequency of employee behaviors are limited. As a result, the proposed BBS management lacks some sustainability in collecting employee's behavior data. Moreover, the case study was implemented in state-owned company and mostly involved dangerous petrochemical experiments at the time of the study. In some ways, the BBS process may be not allowed and interrupted. Finally, the company culture and the nature of the company may influence the sustainability and effectiveness of the employees-focused BBS management and such influence needs further research.
Author contribution All authors have contributed to the study. Junqiao Zhang is the main contributor to writing the manuscript. Xuebo Chen checked the grammar and language of this manuscript and collected data in the petrochemical plant. Qiang Qu analyzed and computed the data.
Funding This study was supported by the National Science Fund of China (71571091, 71771112) Availability of data and material All data generated or analyzed during this study are included in this published article. More detailed data may be provided by the author upon reasonable request and permission

Consent for participate Not applicable
Competing interests The authors declare no competing interests