Poisoning and Asphyxiation Risk Assessment in a Steel Plant Based on Fuzzy Bayesian Network

Abstract

There is a lack of a quantitative assessment of the risk factors associated with poisoning and asphyxiation accidents in steel enterprises, especially the insufficient treatment of uncertainty in risk analysis. To address this issue, this work proposed a risk assessment method based on fuzzy Bayesian network (FBN), which established a risk assessment indicator system for poisoning and asphyxiation from four aspects, including human, material, environmental, and management factors, and illustrated the relationship between these risk factors through fault tree analysis (FTA). Taking a steel plant in China as an example, fuzzy set theory (FST) and expert surveys were combined to determine the prior probabilities and conditional probabilities of Bayesian network (BN) nodes. The results show that (i) the probability of poisoning and asphyxiation accidents in this steel plant is 74%; (ii) among the various influencing factors, defective or inadequate monitoring and alarm devices, isolation devices, equipment inspection systems, and toxic gas operation management are identified as the critical contributors; and (iii) this accident probability has decreased to 47% after rectification measures and reassessment. The findings of this research offer valuable insights for steel enterprises in preventing poisoning and asphyxiation accidents.

1. Introduction

Steel, as an indispensable basic material in modern society, is widely used in many key areas such as construction, automobiles, machinery, and energy, and its demand continues to remain high. Steel enterprises are industrial enterprises that mine, treat, smelt, or process ferrous metal ores into materials. Steel enterprises manufacture various products, including plates, profiles, tubes, and special steels. Although their production processes and product characteristics differ, they all face similar health and safety risks. In China, the steel industry is not only a pillar of the national economy but also an important force in promoting the process of industrialization, which contributes to economic growth and employment. Steel production is a high-risk industry because it is highly susceptible to safety accidents due to the inherent dangers of high temperatures, gas operations, and harsh working environments during the production process. Carrying out a safety risk assessment of typical accidents in steel enterprises is of great significance in promoting the transformation of safety risk management from after-the-fact rescue to before-the-fact.

Numerous scholars have undertaken progressive research into the safety management models and methodologies of steel enterprises. Based on resource eco-economics and eco-industry theory, a risk source model was established to analyze the risk factors in the production process of Chinese steel enterprises [1]. The relevant researchers comprehensively evaluated the health risks associated with occupational exposure to heavy metals within steelmaking installations of Iran steel plants [2]. Scholars analyzed historical accident data from steel enterprises and used hierarchical analysis and gray theory to construct an accident warning model for Chinese steel enterprises, validated through a case study [3]. Meanwhile, other researchers used the failure mode and effects analysis (FMEA) and risk priority number (RPN) to analyze risk factors in the Yazd steel plant’s departments, identifying hazardous activities and proposing improvement strategies [4]. Scholars used the fuzzy evaluation method combining the entropy weight method (EWM) and the ordinal relationship method (ORM) to establish a risk evaluation system for toxic gas leakage and assessed the overall condition of the Chinese metallurgical plant as “excellent” [5]. It can be seen that there are relatively few papers on the disaster risk assessment of typical accidents in steel enterprises.

Among traditional risk analysis methods, the Bayesian network (BN) is an effective causal analysis tool for uncertain knowledge in probabilistic systems [6]. As an analytical tool, BN can be combined with other analytical methods (bow-tie method, fault tree method, fuzzy theory, etc.), which can be used to perform risk analysis under conditions of imperfect historical data and make the assessment more effective in accident analysis [7,8,9]. Therefore, the BN model is widely used for risk analysis in various fields. Fuzzy logic was incorporated into the BN used for the safety assessment of oil and gas pipelines, and the most significant causes of oil and gas pipeline failures were found [10]. Researchers proposed a methodology to analyze Arctic shipping accident scenarios using BN and identified the most important causal factors for potential accident scenarios [11]. Combining the similarity aggregation method with the BN model resulted in more accurate and reliable tank accident prediction results [12]. BN is suitable for analyses with insufficient data information and many indicator factors that require probabilistic analysis and explanation of the inference process. Its shortcomings are that it needs to obtain part of the data with the help of subjective evaluation, which reduces the rigor and has complex analytical calculations. Overall, BN is an effective and widely used method for risk analysis.

In this work, a targeted risk assessment model for accidents in steel enterprises was proposed, integrating fault tree analysis (FTA) with fuzzy Bayesian network (FBN) methodology, to realize the quantitative assessment of this risk. A case study was carried out on a steel plant where forward and backward reasoning and sensitivity analysis were used to predict the probability of accidents, identify critical influencing factors, and suggest improvement measures.

2. Methodology

2.1. Statistics of Accidents in Steel Enterprises

The analyzed accident data are primarily sourced from the official website of China’s Ministry of Emergency Management. The time frame for these accidents spans from 1 January 2010 to 31 December 2023. This work focused on accidents involving steel enterprises, with accident levels designated as larger or above, as classified according to the ‘Regulations on Reporting and Investigation of Production Safety Accidents’. After rigorous screening and organization, 85 valid accident cases that met this study’s criteria were ultimately selected for analysis.

The 85 selected accidents for analysis were systematically categorized by the ‘Classification for Casualty Accidents of Enterprise Employees’ (GB 6441-1986) [13].

Table 1. Poisoning and asphyxiation accidents in some Chinese steel enterprises. (From the accident report posted on the official website of China’s Ministry of Emergency Management).

Date	Location	Cause of the Accident	Number of Deaths	Number of Injuries
3 August 2017	Ezhou Desheng Iron & Steel Co., Ltd. in Hubei Province.	Incorrect operation by personnel, failure to install monitoring and alarm devices, and poor lighting.	3	6
5 February 2018	Shaoguan Steel Group Company Limited in Guangdong Province.	Personnel operational errors.	8	10
13 August 2021	Nantong Tianwen Casting Co., Ltd. in Jiangsu Province.	Lack of strict safety management in inspection and maintenance operations.	3	2
22 August 2021	Malipo Tianxiong New Materials Co., Ltd. in Yunnan Province.	Failure to install detection and alarm devices, poor ventilation.	4	5

presents the poisoning and asphyxiation accidents in some Chinese steel enterprises. Figure 1 presents the analysis outcomes regarding the types of safety accidents in Chinese steel enterprises. As can be seen, when categorized by frequency, poisoning and asphyxiation accidents (39 cases) emerge as the most prevalent, followed by other explosions (13 cases), and burning hot (8 cases). Similarly, in terms of fatalities and injuries, these three accident types also rank among the top. This suggests that poisoning and asphyxiation accidents pose the most significant threat to the production process of steel enterprises. The main reasons for this are the frequent use of gas equipment and the risk of leakage, as well as the combined effects of poor operational management, low employee safety awareness, and poor ventilation in the operating environment. These factors lead to the accumulation of toxic and hazardous gases in the work area, increasing the risk of poisoning and asphyxiation. Therefore, steel enterprises must prioritize identifying and mitigating the risk factors that lead to such accidents in their production safety management.

2.2. Research Framework

To comprehensively evaluate the risk of poisoning and asphyxiation in steel enterprises, this work proposed a risk assessment model that integrated FTA and FBN. Figure 2 presents the research framework employed in this work. The model leverages the fault tree (FT) to deeply explore the logical relationships between risk factors and subsequently maps them onto a BN. This approach not only enhances the inference efficiency but also improves the model’s dynamic updating capabilities. Furthermore, the introduction of FST strengthens the objectivity of the assessment results.

2.3. Fault Tree of the Poisoning and Asphyxiation Accidents

FT is a specialized form of tree logic diagram that utilizes intuitive graphics to elucidate the underlying mechanisms of system failures. By integrating specified events, logic gates, and symbols, it effectively depicts the various causal relationships within a system [14]. Based on the investigation reports of 39 poisoning and asphyxiation accidents collected, the causes of accidents with a high frequency of occurrence were distilled as the basic causal elements leading to accidents. An in-depth analysis of the accident risk factors in steel enterprises was conducted, considering four critical aspects, namely, human, material, environmental, and management, under the guidance of the accident causation theory. Additionally, the risk factors specific to poisoning and asphyxiation accidents were identified by integrating the ‘Classification for Casualty Accidents of Enterprise Employees’ (GB 6441-1986) and the ‘Guidance Manual for Identification and Prevention of Larger Hazardous Factors in Metallurgical Industry’.

Table 2. Risk factors for poisoning and asphyxiation accidents.

Symbol	Description	Symbol	Description
T	Poisoning and asphyxiation accidents	X9	Defective monitoring and alarm devices
A1	Human factors	X10	Defective interlocking devices
A2	Material factors	X11	Defective safety warning signs at hazardous locations
A3	Environmental factors	X12	Improper explosion protection devices
A4	Management factors	X13	Poor lighting in the work area
B1	Unsafe behavior	X14	Poor ventilation of production sites
B2	Defective equipment, facilities, tools, accessories	X15	Cramped and cluttered workplaces
B3	Defective guards, safeties, signals, etc.	X16	Places where people gather in unsafe locations
B4	Poor production work environment	X17	Inadequate accountability for production safety
B5	Unreasonable placement of goods and equipment	X18	Inadequate dual prevention mechanisms
B6	Inadequate production safety regulations	X19	Inadequate equipment inspection management systems
B7	Inadequate education and training	X20	Inadequate system for reporting and investigating accidents
B8	Emergency management deficiencies	X21	Inadequate safety operating procedures
B9	Failure to implement safety management measures for key operations	X22	Inadequate safety training for gas workers
X1	Venturing into dangerous places	X23	Insufficient safety training for confined space operations
X2	Operating errors and ignoring warnings	X24	Insufficient security education at the three levels
X3	Working without a special operations certificate, etc.	X25	Flawed emergency preparedness plans
X4	Defective gas pipe isolation devices	X26	Inadequate emergency drills
X5	Defective gas drains	X27	Inadequate safety management of inspection and maintenance operations
X6	Defective dust collectors	X28	Inadequate safety management of toxic and hazardous gas operations
X7	Defective dispersal devices	X29	Inadequate safety management of confined space operations
X8	Defective fire-fighting equipment and facilities	—	—

presents the risk factors for poisoning and asphyxiation accidents. Figure 3 presents the poisoning and asphyxiation FT to further characterize the relationships between these risk factors.

2.4. Bayesian Network

(i)

Bayesian network fundamentals

BN, a directed acyclic graph (DAG) incorporating probabilistic analysis and graph theory, was extensively utilized in constructing security models that relied on uncertain knowledge [15]. In BN, nodes represent variables, while directional arrows depict direct causal relationships between interconnected nodes. The arrows point to the child nodes from the parent nodes, and the conditional probability tables (CPTs) assigned to the nodes indicate the dependencies between them [16,17]. BN has proven to be an efficacious approach to risk assessment, particularly in scenarios where data availability is limited [7,18]. BN facilitates both forward and backward reasoning by propagating evidence across the network and updating probabilities, thus identifying crucial influences that are vital to overall safety [19].

Assuming that there exists m mutually independent parent nodes X_i1, X_i2, …, X_ij, …, X_im for the child node X_i in the BN, the probability of X_i can be expressed as:

$$P(X_i)={\displaystyle \sum _{j=1}^m\left[P(X_i\left|X_{ij})P(X_{ij})\right.\right]}$$

(1)

where P(X_i) is the probability of the child node X_i, P(X_i∣X_ij) is the probability of X_i occurring given that the parent node X_ij has occurred, and P(X_ij) is the probability of the parent node X_ij.

When P(X_i) changes, perform backward reasoning to calculate the posterior probability of each node:

$$P(X_{ij}\left|X_i\right.)={\displaystyle \frac{P(X_i\left|X_{ij})P(X_{ij})\right.}{P(X_i)}}$$

(2)

(ii)

Transforming fault tree into Bayesian network

In the realm of risk assessment using BN, numerous experts employed a technique that involved mapping FT onto BN [20,21,22,23,24]. This approach leverages the FT’s proficiency in combining the logical relationships between risk factors while fully harnessing the deductive reasoning strengths of BN. The resulting network structure is designed to exhibit dynamic characteristics, enabling it to update the occurrence probabilities of basic events promptly based on the input of new data. The correspondence of an FT to a BN and the logic gate conversion are shown in Figure 4 and Figure 5 [25]. The mapping of FT to BN has the following main steps.

Structure mapping: Each event in the FT (including the bottom event, intermediate event, and top event) is mapped to a node in a BN. Among them, the bottom event (basic event) is converted to the root node of the BN; the intermediate event is converted to the middle node of the BN; and the top event is converted to the leaf node of the BN.

Probability mapping: According to the logical relationship between the events in the FT (e.g., “and gate” and “or gate”), the corresponding CPT is established in the BN.

2.5. Fuzzy Set Theory

FST, introduced by the eminent scholar Zadeh [26], has demonstrated significant advantages and applicability in tackling complex and uncertain problems. Since risk factors in the actual production environment are often ambiguous and uncertain, the description of risk levels such as “low”, “medium”, “high”, etc., often relies on the subjective judgment and empirical knowledge of experts. Fuzzy theory can effectively deal with this kind of ambiguity by defining fuzzy sets and fuzzy rules and transforming the fuzzy language of experts into numerical model parameters, to realize the quantification of risk factors. Various fuzzy number forms have existed, including trapezoidal, normal, and triangular fuzzy numbers. Triangular fuzzy numbers were commonly employed to furnish a relatively precise description of uncertainty and yield plausible outcomes [27]. Consequently, for the sake of computational convenience, this work employed the triangular fuzzy number as a functional representation to describe the node failure rate.

(i)

Triangular fuzzy numbers

In FST, considering an argument domain U, for any element x belonging to U, there exists a corresponding value F(x) ranging from 0 to 1. This F represents a fuzzy set on U, where F(x) signifies the degree of affiliation of x to F. Specifically, the triangular affiliation function can be formulated as follows. By definition, its membership function is:

$$F(x)=\left\{\begin{array}{l}0 x<a\\ {\displaystyle \frac{x-a}{b-a}} a\le x<b\\ 1 x=b\\ {\displaystyle \frac{c-x}{c-b}} b<x\le c\\ 0 x>c\end{array}\right.$$

(3)

where a is the minimum possible value of the node state, b is the most likely value of the node state, and c is the maximum possible value of the node state.

(ii)

Natural Language Variables

Seven fuzzy linguistic scales were defined to describe the failure probability of basic events [28,29].

Table 3. Correlation between natural language variables and triangular fuzzy numbers.

Natural Language Variables	Triangular Fuzzy Numbers
Very Low (VL)	(0.0, 0.0, 0.1)
Low (L)	(0.0, 0.1, 0.3)
Fairly Low (FL)	(0.1, 0.3, 0.5)
Medium (M)	(0.3, 0.5, 0.7)
Fairly High (FH)	(0.5, 0.7, 0.9)
High (H)	(0.7, 0.9, 1.0)
Very High (VH)	(0.9, 1.0, 1.0)

presents the correlation between these natural language variables and triangular fuzzy numbers.

(iii)

Expert scoring weights

Given the variations in expertise across different domains, the two critical influencing factors of expertise disparities, namely, ‘education’ and ‘years of experience’, were selected [9,16,25].

Table 4. Expert scoring weights.

Item	Kind	Weight α	Item	Kind	Weight β
Education	Doctoral	1	Years of experience	More than 20 years	1
	Master’s degree	0.9		10 to 20 years	0.9
	Undergraduate	0.8		5 to 10 years	0.8
	Junior college	0.7		Less than 5 years	0.7

presents the expert scoring weights. The experts’ opinions are processed accordingly using Formula (4).

$$P={\displaystyle \frac{{\displaystyle \sum _{k=1}^n{\alpha }_k{\beta }_k{P}_k}}{{\displaystyle \sum _{k=1}^n{\alpha }_k{\beta }_k}}}$$

(4)

where α_k is the weight of the kth expert’s education, β_k is the weight of the kth expert’s years of experience, $P_k$ is the triangular fuzzy value transformed by the kth expert’s linguistic description of the probability of the node’s occurrence, and P is the fuzzy value after synthesizing the expert’s opinion on the node.

• (iv) Defuzzification

Defuzzification is the process of producing quantifiable results in fuzzy logic using Formula (5) [30]:

$$FPS={\displaystyle \frac{a+2b+c}{4}}$$

(5)

• (v) Fuzzy probability

The last step is to convert the fuzzy possibility (FPS) of the fuzzy event into fuzzy probability (FPR). Formula (6) is employed to convert the FPS to the FPR [31].

$$FPR=\left\{\begin{array}{l}{\displaystyle \frac{1}{{10}^K}} FPS\ne 0\\ 0 FPS=0\end{array}\right., K={({\displaystyle \frac{1-FPS}{FPS}})}^{{\displaystyle \frac{1}{3}}}\times 2.301$$

(6)

where K is a constant value, FPS is the fuzzy possibility, and FPR is the fuzzy probability of each node.

3. Case Study

3.1. Case Background

Baotou Iron and Steel (Group) Co., Ltd. in Inner Mongolia Autonomous Region of China was established in 1954, after 70 years of development, it has become the world’s largest rare earth industrial base and China’s important iron and steel industrial base, with two listed companies, Baotou Iron and Steel Co., Ltd. and Baotou Rare Earth Co., Ltd. in Inner Mongolia Autonomous Region of China. This work conducted a risk assessment of a steel plant under the Baosteel Group in 2023. The steel plant encompasses two steelmaking continuous casting operation zones and a raw material operation zone. Three production process flows are in operation: iron pre-treatment, converter, and continuous casting; iron pre-treatment, converter, refining, and continuous casting; and iron pre-treatment, converter, refining, vacuum degassing (VD), and continuous casting. Throughout these processes, various potential risks, including poisoning, asphyxiation, burns, explosions, electric shock, and fires, may arise due to various factors such as operational errors, equipment failures, and natural phenomena. If these risks are not adequately managed, they can easily result in casualties and property damage.

The developed assessment model was utilized to analyze the risk of poisoning and asphyxiation accidents within the steel plant. To ensure the assessment’s accuracy and comprehensiveness, a field study was conducted with six steel industry safety management experts. The experts thoroughly examined the plant, identifying and documenting hidden hazards by the standards and regulations of the steel industry. This work revealed numerous hidden hazards.

Table 5. The steel plant site assessment of some hidden hazards.

Description of Hidden Hazards	On-Site Photos	Description of Hidden Hazards	On-Site Photos
The exhaust fan was damaged in the desulfurization nitrogen valve room.		Confined space operators were not equipped with gas or oxygen alarms.
The slide gate valve lacked drainage at the bottom.		The de-dusting tank lacked a diffuser tube at the top.
Explosion-proof electrical appliances were not blocked under the gas pipelines.		CO alarms failed to be set at valve stations below gas lines.

presents a selection of these hazards.

3.2. Bayesian Network Modeling of Poisoning and Asphyxiation Accidents

(i)

Node probability calculation

The prior probability is the probability of an event occurring based on statistical data or empirical analysis [32]. To determine the prior probabilities of the root nodes, the opinions of six experts were solicited, who were tasked with evaluating the occurrence (True) status of the bottom event in the poisoning and asphyxiation accidents depicted in Figure 3. In this work, the impact of the experts’ educational backgrounds and years of relevant industry experience on the accuracy of the assessment was taken into account.

Table 6. Expert information and weighting.

Expert Number	Education	Weight α	Years of Experience	Weight β	Combined Weight
Expert 1	Doctoral	1	More than 20 years	1	1
Expert 2	Master’s degree	0.9	More than 20 years	1	0.9
Expert 3	Master’s degree	0.9	10 to 20 years	0.9	0.81
Expert 4	Undergraduate	0.8	10 to 20 years	0.9	0.72
Expert 5	Undergraduate	0.8	5 to 10 years	0.8	0.64
Expert 6	Junior college	0.7	5 to 10 years	0.8	0.56

presents the experts’ information and their respective weight allocation.

Table 7. The expert fuzzy evaluation and comprehensive calculation result of the root nodes.

Root Nodes	Expert 1	Expert 2	Expert 3	Expert 4	Expert 5	Expert 6	P	FPS	FPR
X1	H	H	H	H	H	FH	(0.676, 0.876, 0.988)	0.854	0.053
X2	H	VH	H	VH	VH	VH	(0.822, 0.961, 1)	0.936	0.115
X3	M	M	M	M	FL	FL	(0.248, 0.448, 0.648)	0.448	0.003
X4	H	VH	VH	FH	VH	VH	(0.795, 0.932, 0.984)	0.911	0.087
X5	H	H	H	FH	H	FH	(0.645, 0.845, 0.972)	0.827	0.043
X6	FH	FH	M	FH	FH	FL	(0.417, 0.617, 0.817)	0.617	0.011
X7	FH	H	VH	VH	VH	VH	(0.775, 0.916, 0.978)	0.896	0.075
X8	H	FH	H	FH	H	H	(0.630, 0.830, 0.965)	0.814	0.039
X9	VH	VH	H	VH	VH	H	(0.841, 0.970, 1)	0.945	0.128
X10	M	FH	H	H	H	H	(0.575, 0.775, 0.916)	0.760	0.027
X11	H	H	H	FH	H	M	(0.621, 0.821, 0.948)	0.803	0.036
X12	H	FH	VH	VH	VH	VH	(0.779, 0.920, 0.981)	0.900	0.078
X13	M	FL	M	M	M	M	(0.261, 0.461, 0.661)	0.461	0.004
X14	M	H	FH	H	H	H	(0.579, 0.779, 0.918)	0.764	0.028
X15	FH	FH	FH	FH	M	FL	(0.424, 0.624, 0.824)	0.624	0.011
X16	M	FL	M	FL	M	M	(0.230, 0.430, 0.630)	0.430	0.003
X17	FH	FL	FH	FH	FH	M	(0.398, 0.598, 0.798)	0.598	0.010
X18	FH	FH	FH	FH	M	M	(0.448, 0.648, 0.848)	0.648	0.013
X19	VH	VH	H	VH	FH	VH	(0.810, 0.941, 0.986)	0.920	0.096
X20	FL	FL	M	M	M	M	(0.218, 0.418, 0.618)	0.418	0.003
X21	FH	FH	FH	FH	FH	M	(0.476, 0.676, 0.876)	0.676	0.016
X22	FH	VH	VH	VH	H	VH	(0.786, 0.921, 0.978)	0.902	0.080
X23	FH	H	H	H	H	FH	(0.633, 0.833, 0.966)	0.816	0.040
X24	FL	FH	FH	M	FH	FH	(0.383, 0.583, 0.783)	0.583	0.009
X25	H	FH	H	H	H	H	(0.661, 0.861, 0.981)	0.841	0.048
X26	H	FH	H	H	M	H	(0.606, 0.806, 0.939)	0.789	0.033
X27	FH	H	M	H	H	H	(0.587, 0.787, 0.926)	0.772	0.029
X28	VH	H	VH	FH	VH	VH	(0.799, 0.934, 0.984)	0.913	0.089
X29	H	VH	FH	VH	VH	VH	(0.787, 0.926, 0.983)	0.906	0.083

presents the results of the analysis, using Formulas (4)–(6).

The probabilities of the intermediate nodes are determined by the CPT generated by the logical gate assignment in the FTA model. Taking intermediate node B₄ (poor production work environment) as an example, Figure 3 shows that B₄ is associated with its parent nodes X₁₃ (poor lighting in the work area) and X₁₄ (poor ventilation of production sites) as ‘or’ gates. According to the logic gate mapping algorithm depicted in Figure 5, the conditional probabilities of the intermediate node B₄, P (B₄ = True|X₁₃ = True, X₁₄ = True), P (B₄ = True|X₁₃ = True, X₁₄ = False), and P (B₄ = True|X₁₃ = False, X₁₄ = True), are all 1, and P (B₄ = True|X₁₃ = False, X₁₄ = False) has a value of 0. B₄ was found based on Formula (1):

$$\begin{array}{c}P(B_4=True)=P(X_{13}=True)P(X_{14}=True)+P(X_{13}=True)\\ P(X_{14}=False)+P(X_{13}=False)P(X_{14}=True)\\ =0.03\end{array}$$

(7)

$$P(B_4=False)=1-P(B_4=True)=0.97$$

(8)

(ii)

Forward reasoning

Forward reasoning refers to the process of determining the probability of a risk event occurring in a given state, where the risk factors are already identified, utilizing the conditional probability distribution among the nodes. The probability for all intermediate and leaf nodes was calculated using the above process, and these data were then inputted into the Genie 2.3 software.

Figure 6 shows the forward reasoning results of poisoning and asphyxiation accidents in the steel plant. It can be found that the probability of poisoning and asphyxiation accidents occurring in the steel plant stands at 74%, highlighting the potential risk associated with such accidents.

(iii)

Backward reasoning

Backward reasoning involves the calculation of the posterior probability of each root node based on the conditional probability distribution among the nodes under the known occurrence of a risk event. In this work, with the assumption that the poisoning and asphyxiation accidents have occurred in the steel plant, the probability of the target node T (poisoning and asphyxiation accidents) was set to 100%. By applying backward reasoning, the posterior probability of each node was obtained.

Figure 7 presents the prior and posterior probability of root nodes. It can be found that five risk events have a particularly significant impact on poisoning and asphyxiation accidents in the steel plant. These are X₉ (defective monitoring and alarm devices), X₂ (operating errors and ignoring warnings), X₁₉ (inadequate equipment inspection management systems), X₂₈ (inadequate safety management of toxic and hazardous gas operations), and X₄ (defective gas pipe isolation devices). All of these risk events have posterior probabilities exceeding 12%. Therefore, it is essential to focus on these five risk events, prioritize the development of stringent improvement measures, and reduce the probability of these five risk events occurring.

• (iv) Sensitivity analysis

Sensitivity analysis serves as a method to identify crucial factors contributing to accidents by assessing the degree of influence that other nodes have on the target node. Here, a sensitivity analysis was conducted on node T (poisoning and asphyxiation accidents) to identify the critical factors leading to such accidents.

Figure 8 presents the outcomes of the sensitivity analysis of partial root nodes for poisoning and asphyxiation accidents. It can be found that several risk factors significantly influence the prevention of poisoning and asphyxiation in the steel plant. Specifically, X₉ (defective monitoring and alarm devices), X₂ (operating errors and ignoring warnings), X₁₉ (inadequate equipment inspection management systems), X₂₈ (inadequate safety management of toxic and hazardous gas operations), and X₄ (defective gas pipe isolation devices) have been identified as major contributors, with their impact diminishing in the order stated. Therefore, attention should be focused on these five risk factors, and targeted prevention and control measures should be taken to ensure the safe and stable operation of the steel plant.

3.3. Probability of Accident after Rectification

The steel plant has accorded significant attention to the hidden hazards identified through the experts’ on-site investigation and has implemented corrective measures accordingly. Rectification is the process of eliminating or controlling the discovered potential safety hazards, through taking a series of effective measures to make the production environment, equipment and facilities, personnel behavior, etc., meet the requirements and standards of safe production. The rectification process usually includes the steps of identifying and evaluating potential hazards, formulating rectification programs, implementing rectification measures, and verifying the effects of rectification. Despite the fact that some hidden hazards involving complex factors such as management systems and fire-fighting equipment are difficult to fully rectify in a short period of time, the remaining hidden hazards have been basically addressed. Figure 9 presents the rectification status of some hidden hazards.

To evaluate the effect of the rectification, the six experts were invited to conduct a survey and to score the steel plant again. Figure 10 presents the comparison of the prior probability of root nodes before and after rectification. According to the results of forward reasoning, the probability of poisoning and asphyxiation accidents has decreased to 47%. This significant decline not only underscores the effectiveness of the rectification work but also validates the reliability of the established assessment indicator.

The following recommendations were proposed for some of the critical influencing factors that contribute to poisoning and asphyxiation accidents in the steel plant.

Regarding operating errors and ignoring warnings: establish standardized procedures for recording and responding to alarm information, with corresponding emergency response plans based on alarm levels; organize regular safety training to ensure employees understand and master safety protocols and emergency handling measures.

For monitoring and alarm devices: locate them in potential toxic gas release areas such as pumps, valves, and unloading ports; arrange gas alarms reasonably with adequate spacing for comprehensive monitoring; establish a multi-tiered alarm mechanism with distinct levels including warnings, alarms, and emergency alarms; install the alarm controller independently from production process control instrumentation in a manned area.

To optimize equipment inspection management: establish a standardized inspection process with a comprehensive plan, schedule, and clear inspection content; strengthen training and oversight of inspection personnel, including a periodic evaluation mechanism; intensify preventive maintenance measures, including detailed files and maintenance records; implement real-time monitoring and analysis of equipment operating conditions.

4. Conclusions

This work established a risk assessment indicator system to quantify the probability of poisoning and asphyxiation accidents in a Chinese steel plant, identifying critical risk factors and significantly reducing accident risk through rectification measures. The main conclusions are as follows:

(i) A risk assessment indicator system was established for poisoning and asphyxiation accidents, encompassing human, material, environmental, and management factors, providing a framework that can be utilized for risk assessment in similar enterprises. The application of the FTA and FBN methodology to a steel plant in China provided a quantitative assessment of the probability of poisoning and asphyxiation accidents, which are found to be 74%.

(ii) Through a detailed analysis of the BN results, these critical risk factors that significantly contribute to the occurrence of poisoning and asphyxiation accidents were pinpointed. Among the various influencing factors, defective or inadequate monitoring and alarm devices, isolation devices, equipment inspection systems, and toxic gas operation management are identified as critical contributors to the high accident probability. This finding not only provides a direct target point for risk prevention and control in steel plants, filling the gap of the FBN assessment method in the field of accident risk research in steel enterprises, but also enriches the theoretical practice of industrial safety management and demonstrates the unique advantages of FBN in the risk assessment of complex systems.

(iii) The probability of poisoning and asphyxiation accidents decreased to 47% after rectification measures and reassessment. This significant decline not only underscores the effectiveness of the rectification work but also validates the reliability of the established assessment indicator. Considering the steel plant’s specific conditions, further risk prevention and control measures were proposed, including enhancing the inspection and maintenance of equipment, refining the safety management system, and intensifying employee safety training. It provides theoretical support and practical guidance for safety production in the steel industry.

(iv) The construction of the FBN model relies on a series of assumptions and prior knowledge, which may deviate from the actual situation and affect the accuracy of the assessment results. In addition, all possible risk factors, such as changes in the external environment, are not considered in the model, which may also lead to incomplete assessment results. Future research can explore how to integrate data from different channels, such as internal corporate monitoring data, industry reports, and government regulatory data, to improve the comprehensiveness and accuracy of risk assessment. At the same time, big data and machine learning technologies can be introduced to automatically identify and extract key risk factors and reduce the impact of human intervention and subjectivity.