Human Factors Analysis of China’s Confined Space Accidents from 2013 to 2022: Ensuring the Safe and Sustainable Development of Enterprises

Abstract

Confined space operations are inherently dangerous, leading to frequent accidents with serious consequences. This study utilizes an improved Human Factors Analysis and Classification System (HFACS) model to analyze the human factors contributing to confined space accidents, employing both chi-square tests and grey correlation analysis. The integration of these two analytical methods is essential for providing a comprehensive understanding of the causal relationships among human factors, thereby enabling more robust and validated conclusions. Initially, we identified human factors associated with confined space accidents based on the traditional HFACS model and the unique characteristics of confined space operations, resulting in the identification of 5 primary risk factors and 17 secondary risk factors. Subsequently, we employed chi-square tests and grey correlation analysis to examine the causal relationships among these human factors. The combined results of these methods reveal significant influence relationships within the HFACS model levels pertinent to confined space accidents, identifying 11 significant causal relationships and three paths of accident development. The primary aim of the statistical and correlation analyses is to identify and validate the significant causal relationships among the identified human factors, thereby enhancing our understanding of their impact on confined space accidents. The findings of this research are instrumental in reducing the risk of confined space accidents within enterprises, ultimately ensuring the safe and sustainable operation of production processes.

1. Introduction

In recent years, the number of confined spaces has increased rapidly with the acceleration of modernization. The frequency of confined space operation has increased greatly, and confined space accidents have gradually attracted people’s attention. Poisoning, suffocation, explosions, and other accidents in confined space have caused a huge social impact [1,2,3]. The characteristics and common problems of confined space accidents are very prominent, and are mainly reflected in the insufficient ability of enterprises to identify the risks of confined space operations, the inadequate approval procedures for the implementation of confined space operations, the imperfect management mechanism of confined space outsourcing operations, the failure to equip necessary protective supplies and emergency equipment according to regulations, etc. [4]. Most of these accident factors are related to unsafe acts. Therefore, it is necessary to analyze the human factors in the confined space accident. This research not only aims to enhance safety protocols but also contributes to the sustainable operation of enterprises by minimizing risks associated with confined space work.

The Human Factors Analysis and Classification System (HFACS) model is often used in human factors analysis. It was first used to investigate the influence of human factors on accidents in the aviation field, and then gradually used in accidents dominated by human factors such as in coal mines, the chemical industry, transportation, and elevators. For example, G. Lenné et al. [5] used the HFACS model to analyze 263 major mining accidents in Australia in 2007–2008 and found correlations between accident factors. Some factors can be used to predict the likelihood of other factors. Zarei et al. [6] proposed a hybrid dynamic human factor model based on HFACS, intuitionistic fuzzy set theory, and a Bayesian network to study the influence of human factors on accidents in the chemical industry. Zhang et al. [7] combined the causal classification framework of HFACS with the interaction model of contributing factors to analyze 396 traffic accidents in 28 provinces in China and found that human unsafe behavior is the key factor affecting traffic safety. In this paper, the HFACS model is used to identify the risk of the confined space operation process from the perspective of human factors, which is helpful for the subsequent safety risk analysis and risk control of the confined space operation.

Confined space operation risk has been threatening the health and safety of operators, and research on the confined space operation safety has been carried out continuously. Selman et al. [8,9] analyzed the fatal injuries caused by work-related injuries involving confined space, and the research showed that most of the deaths were caused by blind rescue. The main cause of death of the entrants was poisoning asphyxiation caused by toxic and harmful gases, and the death of rescue workers was mostly caused by toxic and harmful gases. Burlet-Vienney et al. [10,11,12] made a statistical analysis of 40 confined space accidents in Quebec, Canada, from 1998 to 2011. They found that enterprises neglected the management of contract assignment and how to integrate and prevent the risks of confined space operation in actual management. Enterprises lack specific operational guiding principles in training, rescue, and other aspects. Ross [13] elaborated on the hazards, risks, steps, and regulatory requirements of entering the confined space, and discussed the need for safety permission to enter the confined space, the relevant responsibilities of the entrants and their entourage, and the rescue requirements after an accident. Wang et al. [14] proposed a deep learning-based accident factors analysis model by using BERT-BiLSTM-CRF and CNN models. The model can extract the accident factors from a large number of confined space accident reports and carry out statistical analysis.

Current research on confined space risk analysis covers areas such as construction, petrochemicals, aviation, and shipping [15,16,17]. For example, Sakar et al. [18] analyzed the causes and potential consequences of confined space accidents onboard ships using the fuzzy bow-tie method. The findings suggest that high-level changes, including enhancements in industry practices, shifts in organizational culture, and clearer definitions of roles and responsibilities, are essential to reduce the likelihood of similar accidents reoccurring. Soner et al. [19] analyzed the causes of 101 confined space accidents onboard ships using the DEMATEL method. The results indicate that human error and weaknesses in organizational factors are the main causes of these accidents. Recent years have seen a significant increase in confined space accidents within manufacturing and trade enterprises, a sector that has been underexplored in previous risk analysis studies. While research in industries such as construction, petrochemicals, and aviation has provided useful insights, the frequent occurrence of confined space accidents in manufacturing and trade highlights the need for focused studies in this area to improve safety.

Previous studies on the risk analysis of confined space accidents often focused on identifying individual factors, lacking an in-depth exploration of the complex relationships among human factors, which resulted in an incomplete assessment of the multiple causes of accidents. Moreover, many studies failed to effectively combine quantitative and qualitative analyses, limiting accurate identification and prediction of risks. This research addresses these shortcomings by applying an improved HFACS model in conjunction with statistical analysis and grey correlation analysis to comprehensively identify and validate the causal relationships among human factors. This novel combination of the HFACS model with statistical analysis in the risk analysis of confined space accidents provides more reliable and comprehensive insights, ultimately contributing to effective risk reduction.

At present, the identification of accident causation factors in confined space operation still needs to be carried out, especially as the correlation between the factors causing accidents and the path leading to accidents are not clear. The purpose of this paper is to identify the human factors of confined space accidents, analyze the causal relationship between the human factors, and determine the accident path. This paper takes the cause of unsafe acts of confined space operators as the research object, and statistically analyzes the confined space accidents in China from 2013 to 2022. The improved HFACS model is used to identify the risk of the confined space operation process. The chi-square test, odds ratio analysis, and grey correlation analysis are used to determine the causal relationship between the human factors, the basic path of accident development, and the key factors that can block the development path of accidents.

2. Methods

2.1. Traditional HFACS Model

The HFACS model is proposed based on the “Swiss cheese” accident causation model [20]. As shown in Figure 1, the accident is directly triggered by human unsafe acts. An unsafe act is an easily identifiable explicit error. The reasons behind unsafe acts need to be determined by analyzing the hidden errors [21].

This model provides a structured framework to categorize and analyze human factors in accidents systematically [22]. It helps identify underlying issues at various levels, such as organizational influences, unsafe supervision, and preconditions for unsafe acts, which are prevalent in the high-risk, confined space environments typical of manufacturing and trade industries [20,23,24]. By using HFACS, we can address both immediate human errors and deeper systemic issues, making it highly applicable for confined space risk assessment.

The HFACS model is mainly used to identify the causes of unsafe acts. The dangerous act factors can be identified by tracking the work process. It abandons the traditional individual investigation method. It analyzes and interprets the causes of unsafe acts from the comprehensive perspective of individuals and organizations, so as to achieve the purpose of formulating corresponding prevention and treatment measures before accidents [25]. The HFACS model makes a detailed study of the causes of human error on four levels: unsafe acts, preconditions for unsafe acts, unsafe supervision, and organizational influence. Each higher level will affect the next lower level, which effectively solves the problem of combining the analysis of human factors with practical application in the analysis of accident causation. The traditional HFACS model is shown in Figure 2.

2.1.1. Organizational Influence

(1)

Organizational process: Policies and rules that govern and define organizational activities, such as work organization, work procedure standards, etc.

(2)

Organizational climate: Factors that can affect the effectiveness and motivation of a business, such as organizational structure, culture, policies, etc.

(3)

Resource management: Decisions made at the management level regarding the allocation and maintenance of business resources, such as personnel, funds, facilities, etc.

2.1.2. Unsafe Supervision

(1)

Inadequate supervision: The supervisor did not maintain good and effective communication with the operator during the operation, and did not provide professional advice and emergency measures.

(2)

Planned inappropriate operation: Insufficient working time, excessive tasks or workload and unreasonable work scheduling.

(3)

Failed to correct problem: Supervisors allow the operation to continue even if they are aware of deficiencies in personnel allocation, training, and other related safety aspects.

(4)

Supervision violations: Supervisors intentionally violate existing regulatory procedures, such as allowing unqualified personnel to operate, failing to follow effective operational authorization procedures, etc.

2.1.3. Preconditions for Unsafe Acts

(1)

Environmental factors

Physical environment: Working environment (such as lighting, temperature, humidity, oxygen content, toxic and harmful gas content, etc.). It also refers to the operator’s environment, such as the limited size of the work area, the shape of the room, etc.

Technological environment: This includes various conditions such as operation, control and display of the device to be operated.

(2)

Condition of operators

Adverse mental states: Mental fatigue, loss of understanding of the situation, empiricism, adventurism, weak safety awareness, etc.

Adverse physiological states: Illness, lack of oxygen, physical fatigue, hallucinations, disorientation, or colds.

Physiological/mental limitations: Activity requirements exceed the ability of the person, such as visual limitations, lack of rest, physical adaptation, etc.

(3)

Personnel factors

Personal readiness: Pre-deployment training and education are insufficient to ensure that staff are fit for duty.

Crew resource management: This refers to insufficient cooperation between teams, lack of effective communication, and difficulty in ensuring work status.

2.1.4. Unsafe Acts

(1)

Errors

Decision errors: This refers to when the implementation of the action plan does not meet the requirements of the current situation, which is divided into process error, choice error, and problem-solving error.

Skill-based errors: Ability-based action errors, such as incorrect allocation of attention, memory errors, etc.

Perceptual errors: Errors caused by insufficient perception of information in the current situation, such as incorrect understanding of visual and spatial information leading to incorrect judgment.

(2)

Violations

Routine: This refers to violations that have become habitual due to their long duration and high frequency, which may lead to accidents but are often accepted by most people or supervisory organizations.

Exceptional: This refers to random violations that are unrelated to personal behavior or organizational management systems. Exceptional violations are not related to typical patterns of personal behavior and are difficult to predict.

2.2. Confined Space Accident Causation Model Based on Improved HFACS Model

The traditional HFACS model is a human factor analysis model based on the background of aviation accidents. In the process of analyzing confined space accident investigation reports, many accident causation factors in the HFACS model cannot accurately describe the cause of the confined space accidents [23,26]. Applying the traditional HFACS model directly cannot analyze the confined space accidents completely and accurately. Therefore, according to the process risk of confined space operations and the characteristics of confined space accidents, this paper identifies the specific manifestations of various human factors in confined space accidents, and establishes an accident causation model suitable for confined space accidents based on the traditional HFACS model [25].

In the improved HFACS model, we have considered external factors caused by economic, legal, and other factors, and added an “External safety level” hierarchy that directly influences organizational influence. External factors such as safety input, policies and regulations, and government supervision play a crucial role in the management level of confined space operations. Especially in recent years, in order to prevent confined space accidents, the government and relevant departments have increased supervision efforts, and introduced a number of policies, regulations, and standards to standardize the confined space operation process. Therefore, we choose to add “External safety level” as the supplementary level of HFACS model, and analyze in depth the causes of human errors and violations from five levels: external safety level, organizational influence, unsafe supervision, preconditions for unsafe acts, and unsafe acts. In order to make the accident causation model better match the characteristics of confined space accidents, the “Organizational climate” is changed to “Poor safety culture”, “Inadequate supervision” is changed to “Inadequate supervision of operation process”, “Supervision violations” is changed to “Supervision violations of safety management personnel”, and “Crew resource management” is changed to “Operation team states”. The established confined space accident causation model has 5 levels (primary risk factors) and 17 personal factors (secondary risk factors). The confined space accident causation factors and their descriptions are shown in

Table 1. The confined space accident causation factors and their descriptions.

Primary Risk Factors	Secondary Risk Factors	Descriptions
Level 1: External safety level (X1)	Political and economic factors (X11)	The policies and regulations related to the standardized management of confined space operations are unreasonable or incomplete, the supervision of government departments is weak, and the investment of safety funds is insufficient.
Level 2: Organizational influence (X2)	Improper management processes (X21)	The implementation of the main responsibility for safety production is not in place, the responsibility system for safety production is not perfect, the division of safety management responsibilities is not clear, the hidden danger investigation and management system is not in place, the confined space operation safety management system or the operation safety operation procedures are not established/implemented, the confined space operation approval system is not established/implemented, and risk identification is not carried out in the confined space.
	Poor safety culture (X22)	The enterprise operates illegally, the safety awareness of management personnel is poor, a “safety-first” corporate atmosphere has not been formed, and the reward and punishment incentive system is missing or incomplete.
	Improper resource management (X23)	The enterprise is not equipped with full-time safety management personnel or rescue personnel, and is not equipped with complete safety protection supplies. Emergency rescue equipment failure or lack, lack of safety production qualifications, insufficient funds, etc.
Level 3: Unsafe supervision (X3)	Inadequate supervision of operation process (X31)	Supervisors and operators do not establish effective communication. Supervisors do not provide professional guidance for emergency response. Operators do not detect the gas environment in a limited space during operation.
	Incomplete work plan (X32)	Unreasonable working time arrangement, excessive working tasks or workload, uneven distribution of personnel tasks, unclear division of responsibilities, etc.
	Failed to correct problem (X33)	Supervisors allow work when they find that the operator is not adequately prepared. For example, personnel who do not wear protective equipment in accordance with the regulations are not corrected in time, and operators are allowed to continue working in inappropriate environmental conditions.
	Supervision violations of safety management personnel (X34)	Supervisors intentionally violate existing regulations and procedures by allowing unqualified personnel to operate, failing to conduct pre-operation safety education and skills training, failing to perform effective job approval procedures, and failing to disclose safety technology. For example, supervisors falsify the monitoring records of the gas environment in the confined space, and allow operators to continue their work without re-testing the gas after the operation is interrupted.
Level 4: Preconditions for unsafe acts (X4)	Physical environment (X41)	Excessive toxic and harmful gases, insufficient oxygen concentration, insufficient lighting, no confined space warning signs, bad weather conditions, loud noise, dirty environment, etc.
	Technological environment (X42)	Failure of gas detectors, mechanical ventilation devices and working tools. Lack of necessary safety protection facilities in confined space or failure of facilities, etc.
	Operator states (X43)	The work requirements exceed the scope of personal physical conditions and intelligence. Illness, fatigue, lack of concentration, lack of vigilance, empiricism, risk-taking, flukes, etc.
	Operation team states (X44)	Lack of teamwork, lack of information exchange, and lack of confirmation of safety conditions before operation.
	Personal readiness (X45)	Failure to receive safety education and training before confined space operation. Inadequate work ability. Inadequate work preparation.
Level 5: Unsafe acts (X5)	Skill-based errors (X51)	Errors in operational skills, self-help skills, etc.
	Decision errors (X52)	The decision made during research does not match the actual situation. For example, after an emergency occurs, personnel choose to blindly rescue.
	Perceptual errors (X53)	Errors caused by cognitive biases in objective information within the confined space.
	Violations (X54)	Behaviors that violate regulations or operational requirements, including routine violations and exceptional violations.

.

2.3. Chi-Square Test and Odds Ratio Analysis

The chi-square test is effective for analyzing categorical data and identifying significant associations between various risk factors in confined spaces, such as personnel readiness, supervision quality, and environmental hazards. Since many risk factors in confined spaces are discrete (e.g., presence or absence of safety equipment), the chi-square test provides a statistical method to determine which factors are most likely contributing to accidents, supporting more targeted risk management [27].

The reason for selecting the chi-square test and grey correlation analysis in this study is that both methods effectively reveal the relationships among human factors. The chi-square test assesses the correlation between categorical variables, helping to identify significant risk factors, while grey correlation analysis can handle small samples and uncertainty, quantifying the degree of association among different factors. The combined use of these two methods provides a more comprehensive and detailed causal relationship analysis. Compared to other methods, such as single regression analysis or qualitative studies, this combination better captures complex multiple causal relationships, enhancing the reliability and practicality of the analysis, particularly in the complex environment of confined space accidents, allowing for a more accurate reflection of potential risk factors and their interrelations.

In statistics, the chi-square test is mainly used to test the goodness of fit between the measured value and the inferred value of the sample [28]. When the Pearson (p) value of the chi-square test is less than 0.05, it is considered that there is a significant influence relationship between the measured value and the inferred value of the sample. The odds ratio indicates the degree of correlation between the occurrence of event A and the occurrence of event B [29]. When the odds ratio (OR) value exceeds 1, it indicates that the occurrence of event A can improve the likelihood of occurrence of event B. On the contrary, when the OR value is lower than 1, it indicates that the probability of occurrence of event A and event B is not directly related.

When the p value of the 2 × 2 contingency table chi-square test is less than 0.05 and the OR value is greater than 1, the causal relationship between the upper and lower levels of the HFACS model is considered significant. At the same time, it also shows that the upper-level factors have a greater influence on the lower-level factors, and the occurrence of the upper-level factors can increase the possibility of the lower-level factors [27].

2.4. Grey Correlation Analysis

This method is suitable for analyzing complex, uncertain environments with limited data, like confined spaces, where certain variables may be difficult to quantify or where sample sizes may be smaller. Grey correlation analysis is robust in handling incomplete or ambiguous data, allowing for an effective assessment of the relationship strength between key risk factors [30,31]. This analysis complements HFACS and chi-square tests by quantifying the degree of association among human factors, even when data are less complete.

In the objective world, the relationship between various factors is very complicated. In particular, the randomness of the surface phenomenon of things and the process of change tend to confuse people’s intuition, cover up the nature of things, and make it difficult for people to obtain complete and reliable information in cognition. Therefore, it is difficult to identify primary and secondary factors and conduct a systematic analysis [32,33]. Grey system theory provides us with a new multivariable analysis method—grey correlation analysis (GRA). Grey correlation refers to the non-specific relationship between the factors of a system. GRA has no requirement for sample size and is mainly used to identify the relationship between factors that affect the system [34].

The two main indexes of GRA are correlation degree and correlation order [35]. The correlation degree reflects the influence degree among various factors in the system, that is, the contribution of a certain factor to the system behavior. The order of the correlation degree is the correlation order, which indicates the intensity, magnitude, and order of the relationship between the factors. In this paper, the correlation order is used to measure the influence degree of different human factors on confined space accidents. The main steps of GRA are as follows [36,37]:

(1) Determine the reference sequence and the comparison sequence. The reference sequence is the main object of interest in this study, usually representing an ideal or desired state. Comparison sequences are multiple factors that are compared to the reference sequence, aiming to assess their degree of association with the reference sequence. By analyzing the relationships between these sequences, grey correlation analysis can reveal the impact and correlation of each factor on the reference sequence [31,38,39].

The reference sequence Y(k) and the comparison sequence X_i(k) are shown in Equation (1).

$$\begin{array}{ll}Y\left(k\right),& \hspace{1em}k=1,2,\cdots m\\ X_i\left(k\right),& \hspace{1em}i=1,2,\cdots n\end{array}$$

(1)

(2) Initiate the reference sequence and comparison sequence.

In this paper, the averaging method is adopted, and the new series obtained are shown in Equation (2).

$$\begin{array}{ll}{Y}^{\prime }\left(k\right),& \hspace{1em}k=1,2,\cdots m\\ X_i^{\prime }\left(k\right),& \hspace{1em}i=1,2,\cdots n\end{array}$$

(2)

(3) Calculate the difference sequence Δ_i(k), as shown in Equation (3).

$${\Delta }_i\left(k\right)=\left|X_i^{\prime }\left(k\right)-Y^{\prime }\left(k\right)\right|,\begin{array}{cc} & {\Delta }_i=\left({\Delta }_i\left(1\right),{\Delta }_i\left(2\right),\cdots {\Delta }_i\left(n\right)\right)\end{array}$$

(3)

Maximum difference M and minimum difference m can be determined by Equation (4).

$$\begin{array}{l}M=\underset{i}{\max } \underset{k}{\max } {\Delta }_i\left(k\right)\\ m=\underset{i}{\min } \underset{k}{\min } {\Delta }_i\left(k\right)\end{array}$$

(4)

(4) Calculate correlation coefficient ξ_i(k).

$${\xi }_i\left(k\right)={\displaystyle \frac{m+\rho M}{\rho M+{\Delta }_i\left(k\right)}}$$

(5)

where ξ ∈ (0,1), k = 1,2,…m, ρ is the identification coefficient, ρ = 0.5.

(5) Calculate the correlation degree.

$${\xi }_i={\displaystyle \frac{1}{n}}{\displaystyle {\sum }_{k=1}^m{\xi }_i\left(k\right)}$$

(6)

The larger ξ_i is, the higher the correlation degree between comparison sequence X_i and reference sequence Y is. According to the order of ξ_i, the influence degree of different human factors can be analyzed.

3. Results and Discussion

3.1. Chi-Square Test and Odds Ratio Analysis

3.1.1. Classification and Statistics of Human Factors of Confined Space Accidents

In this paper, 76 major accidents involving the limited space operation of industrial and trade enterprises in China from 2013 to 2022 are selected as samples. The data were sourced from the official accident report database of the Emergency Management Bureau as well as relevant records from the manufacturing and trade industries. Specifically, the collected cases focus on more severe accidents occurring in confined spaces, including those resulting in serious injury or death. To ensure data completeness and applicability, only cases with complete records were selected, and any records with incomplete data were excluded. Regarding inclusion and exclusion criteria, we mainly included safety incidents in confined spaces caused by human or environmental factors and excluded cases not directly related to confined space operations. Additionally, as the data rely primarily on official records, there may be some reporting biases (e.g., certain accidents may not be counted due to lack of formal reporting).

According to the improved HFACS model of confined space accidents, the classification analysis of the human factors of accidents can be carried out. Statistics and analysis were conducted on the 76 accidents selected from the sample according to the accident types and confined space types. The results showed that poisoning and asphyxia accounted for 86.9% (poisoning 72.4% and anoxic asphyxia 14.5%) of the accidents in confined space accidents. Other types of accidents (drowning, explosion, collapse and burial, etc.) accounted for 13.1%. Underground confined space accidents accounted for 59.2% of the samples, accounting for the highest proportion. Ground confined space accidents and enclosed (semi-enclosed) equipment accidents account for 22.4% and 18.4%, respectively. Confined space accidents often occur in underground confined space, and the most common accident type is poisoning and suffocation. The selection of samples conforms to the characteristics of confined space operation accidents.

The HFACS model of confined space accidents is used to classify the human factors in the accident causation. If the direct cause of an accident is the illegal operation of the operator, it is considered as “Violations (X₅₄)”. According to the main body of accident responsibility in the indirect causes, the possible human factors of the accidents are selected according to the accident description, such as “Inadequate supervision of operation process (X₃₁)” and “Personal readiness (X₄₅)”. Finally, from the root cause of the accidents, management factors are selected according to in-depth analysis of the operation process and whether there are “Improper management processes (X₂₁)”, “Poor safety culture (X₂₂)”, and so on. Finally, the frequency and proportion of the human factors of 76 accidents involving confined space in industry and trade enterprises were obtained, as shown in

Table 2. Frequency and proportion of 76 major accidents in confined space of industry and trade enterprises.

Levels of HFACS Model	Human Factors	Frequency	Proportion
Level 1: X1	X11	58	76.31%
Level 2: X2	X21	69	90.79%
	X22	62	81.58%
	X23	61	80.26%
Level 3: X3	X31	48	63.16%
	X32	27	35.53%
	X33	50	65.79%
	X34	54	71.05%
Level 4: X4	X41	65	85.53%
	X42	24	31.58%
	X43	33	43.42%
	X44	35	46.05%
	X45	55	72.37%
Level 5: X5	X51	23	30.26%
	X52	33	43.42%
	X53	15	19.74%
	X54	60	78.95%

.

Figure 3 describes the frequency and proportion of human factors in 76 confined space accidents. As can be seen from the figure, human factors such as “Improper management processes (X₂₁)”, “Physical environment (X₄₁)”, “Poor safety culture (X₂₂)”, “Improper resource management (X₂₃)”, and “Violations (X₅₄)” result in more than 60 confined space accidents. They are considered to be the main cause of confined space accidents. More than 80% of the 76 accidents are related to “Organizational influence (X₂)”, which is considered to have the greatest impact on confined space accidents among the four levels. In the level of “Unsafe supervision (X₃)”, the proportion of “Supervision violations of safety management personnel (X₃₄)” and “Failed to correct problem (X₃₃)” is relatively high. The important human factors in the level of “Preconditions for unsafe acts (X₄)” are “Physical environment (X₄₁)” and “Personal readiness (X₄₅)”. In the level of “Unsafe acts (X₅)”, “Violations (X₅₄)” is the causal factor of most accidents, and 60 accidents are caused by “Violations (X₅₄)”.

3.1.2. Chi-Square Test and Odds Ratio Analysis of Confined Space Accidents

The chi-square test and odds ratio calculation of human factors in X₁ and X₂, X₂ and X₃, X₃ and X₄, and X₄ and X₅ are carried out. The statistical calculation results are shown in

Table 3. Chi-square test and odds ratio analysis of human factors in X₁ and X₂.

	Human Factors	Chi-Square Test		OR	Confidence Interval of 95%
		${\mathit{\chi }}^2$	p		Limit Inferior	Limit Superior
X1 and X2	X11 and X21	7.032	0.008	10.769	1.876	61.807
	X11 and X22	0.048	0.826	0.855	0.210	3.475
	X11 and X23	11.248	<0.001	8.667	2.478	30.313

,

Table 4. Chi-square test and odds ratio analysis of human factors in X₂ and X₃.

	Human Factors	Chi-Square Test		OR	Confidence Interval of 95%
		${\mathit{\chi }}^2$	p		Limit Inferior	Limit Superior
X2 and X3	X21 and X31	3.964	0.046	5.000	0.900	27.770
	X21 and X32	0.163	0.687	1.420	0.256	7.867
	X21 and X33	4.745	0.029	5.714	1.025	31.852
	X21 and X34	4.681	0.030	7.647	1.357	43.084
	X22 and X31	0.258	1.277	1.952	0.605	6.305
	X22 and X32	0.222	1.489	2.316	0.585	9.160
	X22 and X33	5.356	0.021	4.765	1.396	16.258
	X22 and X34	5.059	0.024	4.571	1.357	15.398
	X23 and X31	0.08	0.777	1.182	0.371	3.761
	X23 and X32	4.019	0.045	4.514	0.936	21.778
	X23 and X33	0.473	0.492	0.645	0.183	2.268
	X23 and X34	0.727	0.394	0.553	0.140	2.189

,

Table 5. Chi-square test and odds ratio analysis of human factors in X3 and X4.

	Human Factors	Chi-Square Test		OR	Confidence Interval of 95%
		${\mathit{\chi }}^2$	p		Limit Inferior	Limit Superior
X3 and X4	X31 and X41	1.732	0.188	2.345	0.644	8.548
	X31 and X42	3.863	0.049	3.014	0.977	9.301
	X31 and X4	0.309	0.579	1.308	0.507	3.373
	X31 and X44	0.182	0.669	1.227	0.480	3.136
	X31 and X45	0.154	0.695	0.810	0.281	2.332
	X32 and X41	2.691	0.101	6.667	0.804	55.246
	X32 and X42	0.060	0.807	1.133	0.415	3.095
	X32 and X43	1.212	0.271	1.700	0.658	4.391
	X32 and X44	7.163	0.007	3.765	1.395	10.163
	X32 and X45	0.084	0.772	0.858	0.303	2.430
	X33 and X41	4.948	0.026	4.237	1.110	16.176
	X33 and X42	2.789	0.095	2.574	0.831	7.970
	X33 and X43	1.748	0.186	0.525	0.201	1.371
	X33 and X44	0.223	0.637	1.259	0.484	3.273
	X33 and X45	4.257	0.039	2.933	1.035	8.315
	X34 and X41	0.018	0.895	0.908	0.217	3.796
	X34 and X42	2.572	0.109	2.647	0.785	8.931
	X34 and X43	0.546	0.460	0.688	0.254	1.862
	X34 and X44	0.330	0.566	1.341	0.492	3.659
	X34 and X45	20.073	<0.001	11.75	3.622	38.12

and

Table 6. Chi-square test and odds ratio analysis of human factors in X₄ and X₅.

	Human Factors	Chi-Square Test		OR	Confidence Interval of 95%
		${\mathit{\chi }}^2$	p		Limit Inferior	Limit Superior
X4 and X5	X41 and X51	0.504	0.815	1.185	0.284	4.941
	X41 and X52	2.139	0.144	0.381	0.101	1.433
	X41 and X53	0.920	0.337	2.745	0.323	23.308
	X41 and X54	4.608	0.032	4.091	1.058	15.818
	X42 and X51	0.460	0.497	0.686	0.231	2.042
	X42 and X52	0.044	0.834	0.901	0.338	2.397
	X42 and X53	0.027	0.870	1.105	0.332	3.679
	X42 and X54	0.001	0.975	1.020	0.311	3.347
	X43 and X51	0.000	0.995	1.003	0.374	2.693
	X43 and X52	4.757	0.029	2.811	1.098	7.196
	X43 and X53	2.091	0.148	2.313	0.730	7.330
	X43 and X54	0.357	0.550	0.714	0.236	2.159
	X44 and X51	2.914	0.088	2.370	0.870	6.457
	X44 and X52	0.008	0.927	0.958	0.386	2.381
	X44 and X53	0.399	0.528	1.439	0.463	4.470
	X44 and X54	0.043	0.835	1.125	0.371	3.415
	X45 and X51	0.039	0.843	1.118	0.370	3.380
	X45 and X52	4.543	0.033	3.319	1.067	10.323
	X45 and X53	0.009	0.926	1.063	0.297	3.798
	X45 and X54	6.587	0.010	5.143	1.592	16.619

.

After the screening and sorting of the results in Table 3, Table 4, Table 5 and Table 6, the results with p < 0.05 and OR > 1, that is, the results with significant causal relationship, are statistically analyzed, as shown in

Table 7. Human factors with significant causal relationship between the upper and lower levels of the HFACS model (p < 0.05, OR > 1).

	Human Factors	Chi-Square Test		OR	Confidence Interval of 95%
		${\mathit{\chi }}^2$	p		Limit Inferior	Limit Superior
X1 and X2	X11 and X21	7.032	0.008	10.769	1.876	61.807
	X11 and X23	11.248	<0.001	8.667	2.478	30.313
X2 and X3	X21 and X31	3.964	0.046	5.000	0.900	27.770
	X21 and X33	4.745	0.029	5.714	1.025	31.852
	X21 and X34	4.681	0.030	7.647	1.357	43.084
	X22 and X33	5.356	0.021	4.765	1.396	16.258
	X22 and X34	5.059	0.024	4.571	1.357	15.398
	X23 and X32	4.019	0.045	4.514	0.936	21.778
X3 and X4	X31 and X42	3.863	0.049	3.014	0.977	9.301
	X32 and X44	7.163	0.007	3.765	1.395	10.163
	X33 and X41	4.948	0.026	4.237	1.110	16.176
	X33 and X45	4.257	0.039	2.933	1.035	8.315
	X34 and X45	20.073	<0.001	11.75	3.622	38.12
X4 and X5	X41 and X54	4.608	0.032	4.091	1.058	15.818
	X43 and X52	4.757	0.029	2.811	1.098	7.196
	X45 and X52	4.543	0.033	3.319	1.067	10.323
	X45 and X54	6.587	0.010	5.143	1.592	16.619

.

As can be seen from Table 7, there are six, five, and five significant causal relationships between level 1 and level 2, level 2 and level 3, and level 3 and level 4, respectively. Among them, the causal relationship between “Supervision violations of safety management personnel (X₃₄)” and “Personal readiness (X₄₅)” in level 3 and level 4 is the most significant, and the OR value reaches 11.75. The higher OR value indicates that the occurrence of supervision violations of safety management personnel will increase the possibility of insufficient personnel readiness by more than 11 times. The OR value between “Improper management processes (X₂₁)” and “Supervision violations of safety management personnel (X₃₄)” in level 2 and level 3 is 7.647, indicating that the occurrence of improper management process will increase the possibility of supervision violations of safety management personnel by about seven times. In addition, X₁₁ and X₂₁ (OR = 10.769), X₁₁ and X₂₃ (OR = 8.667), X₂₁ and X₃₃ (OR = 5.714), X₄₅ and X₅₄ (OR = 5.143), X₂₁ and X₃₁ (OR = 5.000), X₂₂ and X₃₃ (OR = 4.765), X₂₂ and X₃₄ (OR = 4.571), X₂₃ and X₃₂ (OR = 4.514), X₃₃ and X₄₁ (OR = 4.237), and X₄₁ and X₅₄ (OR = 4.091) all have significant influence relationships.

3.1.3. Causal Relationship Analysis

According to the chi-square test and odds ratio analysis results, the causal relationship between the upper and lower levels of the HFACS model of confined space accidents is obtained, as shown in Figure 4. It can be seen that there is a clear hierarchical relationship between the human factors in the model.

According to the causal relationship between human factors in Figure 4, five paths leading to confined space accidents can be obtained, as shown in Figure 5. The causal pathway diagram illustrating the factors leading to accidents is designed to visually depict the relationships and influence pathways between different human factors. By visualizing these factors and their interactions, researchers can gain a clearer understanding of the mechanisms behind accident occurrences, identify key risk factors, and provide a basis for developing targeted safety measures. Figures in brackets in the figure represent the proportion of corresponding human factors in the causes of the 76 accidents studied in this paper. The probability of the confined space accident on the path is expressed by the product of the proportion of human factors involved in the path. This product value only represents the probability of the confined space accident developing along this path, rather than the true accident probability [40]. The five accident paths include the following: Political and economic factors (X₁₁) → Improper management processes (X₂₁) → Failed to correct problem (X₃₃) → Physical environment (X₄₁) → Violations (X₅₄) → accidents, Political and economic factors (X₁₁) → Improper management processes (X₂₁) → Failed to correct problem (X₃₃) → Personal readiness (X₄₅) → Violations (X₅₄) → accidents, Political and economic factors (X₁₁) → Improper management processes (X₂₁) → Failed to correct problem (X₃₃) → Personal readiness (X₄₅) → Decision errors (X₅₂) → accidents, Political and economic factors (X₁₁) → Improper management processes (X₂₁) → Supervision violations of safety management personnel (X₃₄) → Personal readiness (X₄₅) → Violations (X₅₄) → accidents, and Political and economic factors (X₁₁) → Improper management processes (X₂₁) → Supervision violations of safety management personnel (X₃₄) → Personal readiness (X₄₅) → Decision errors (X₅₂) → accidents. The probabilities of these five paths leading to the confined space accident are 30.78%, 26.04%, 14.32%, 28.12%, and 15.47%, respectively. There are three paths that have a relatively high probability of causing confined space accidents, exceeding 25%. It can be found that the human factors such as “Improper management processes (X₂₁)”, “Failed to correct problem (X₃₃)”, “Supervision violations of safety management personnel (X₃₄)”, “Physical environment (X₄₁)”, “Personal readiness (X₄₅)”, and “Violations (X₅₄)” are critical to preventing confined space incidents.

3.2. Grey Correlation Analysis

3.2.1. Correlation Analysis Between Primary Risk Factors and Total Number of Human Factors

The purpose of GCA is to assess and quantify the degree of association among different factors, particularly in situations with insufficient data or high uncertainty. Its core principles include processing incomplete information based on grey system theory, measuring the similarity between factors through correlation coefficients, standardizing data to eliminate dimensional influences, and considering the dynamic changes in factors. Through these principles, GCA effectively identifies and quantifies the complex relationships among influencing factors, making it particularly suitable for studying complex issues like confined space accidents.

GCA is a quantitative method to analyze the key human factors that affect confined space accidents. It can deeply explore the interaction between HFACS model levels, so as to take preventive measures for confined space accidents and reduce the occurrence of accidents. In this paper, based on the established HFACS model, an index system for GCA of human factors in confined space accidents is established. And the causes of 76 accidents are classified and statistically analyzed according to the established HFACS model. Taking the total number of human factors in the confined space accident as the reference series and five primary risk factors as the comparison series, each sequence is performed by dimensionless processing. According to the dimensionless sequence obtained after initializing, the absolute difference between the corresponding elements of the comparison sequence and the reference sequence is further calculated, and the difference sequence is obtained as shown in

Table 8. The calculation results of difference sequence.

Difference Sequence	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022
Δ1	0.041	0.026	0.063	0.176	0.003	0.162	0.132	0.008	0.478	0.148
Δ2	0.149	0.034	0.015	0.061	0.043	0.029	0.109	0.082	0.055	0.067
Δ3	0.029	0.109	0.200	0.040	0.071	0.030	0.067	0.162	0.060	0.012
Δ4	0.004	0.065	0.043	0.022	0.069	0.055	0.003	0.018	0.051	0.041
Δ5	0.171	0.092	0.362	0.001	0.075	0.089	0.075	0.073	0.243	0.180

.

Therefore, $M=\underset{i}{\max } \underset{k}{\max } {{\displaystyle \Delta }}_i\left(k\right)$ = 0.478 and $m=\underset{i}{\min } \underset{k}{\min } {{\displaystyle \Delta }}_i\left(k\right)$ = 0.001. According to Equation (5), the correlation coefficients of the corresponding elements of each comparison sequence and reference sequence can be calculated, respectively, and the results are shown in

Table 9. The calculation results of correlation coefficients.

	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022
1	0.8570	0.9054	0.7953	0.5786	0.9902	0.5988	0.6463	0.9708	0.3347	0.6205
2	0.5515	0.8478	0.9286	0.7521	0.8119	0.8660	0.6276	0.6920	0.7712	0.7331
3	0.8667	0.6269	0.4777	0.8226	0.7222	0.8626	0.7325	0.5306	0.7552	0.9418
4	0.9838	0.7398	0.8125	0.8986	0.7288	0.7721	0.9891	0.9146	0.7845	0.8209
5	0.5170	0.6667	0.3349	0.9999	0.7109	0.6741	0.7118	0.7165	0.4295	0.5042

.

According to the correlation coefficient in Table 9, the grey correlation degree can be calculated by Equation (6). The correlation degree between the five primary risk factors and total number of human factors is 0.7298, 0.7582, 0.7339, 0.8445, and 0.6266, respectively. The grey correlation coefficient and average grey correlation degree of the five primary risk factors are drawn in a line chart, as shown in Figure 6. According to the value of the average grey correlation degree, the final grey correlation order is determined as follows: X₄ > X₂ > X₃ > X₁ > X₅, indicating that the “Preconditions for unsafe acts (X₄)” has the strongest correlation with the confined space accident risk.

3.2.2. GCA Between the Secondary Risk Factors

Table 10. GCA between X₁ and X₂.

	X21	X22	X23
X11	0.7929	0.7612	0.8035

,

Table 11. GCA between X₂ and X₃.

	X31	X32	X33	X34
X21	0.7340	0.6687	0.7492	0.7894
X22	0.6769	0.5400	0.6353	0.6710
X23	0.6979	0.6388	0.6378	0.7365

,

Table 12. GCA between X₃ and X₄.

	X41	X42	X43	X44	X45
X31	0.7544	0.6139	0.5654	0.5726	0.6866
X32	0.5829	0.5591	0.6629	0.6147	0.6272
X33	0.7683	0.5622	0.6051	0.7231	0.7565
X34	0.6717	0.5481	0.5938	0.6112	0.8078

and

Table 13. GCA between X₄ and X₅.

	X51	X52	X53	X54
X41	0.6711	0.6682	0.6155	0.8019
X42	0.6776	0.6550	0.5816	0.7217
X43	0.6371	0.7446	0.6738	0.7504
X44	0.6886	0.6803	0.5723	0.7016
X45	0.5985	0.6189	0.5228	0.7651

show the calculation results of the GCA between secondary risk factors.

As can be seen from Table 10, the influence relationship between “Political and economic factors (X₁₁)” and “Improper resource management (X₂₃)” is the most significant, with an average correlation degree of 0.8035. The influence relationship between “Political and economic factors (X₁₁)” and “Poor safety culture (X₂₂)” is the weakest, with an average correlation degree of 0.7612. This shows that the formation of the safety culture is a long-term process. Although it has a certain influence on the “External safety level (X₁)”, compared with resource management and management processes, the influence relationship is not so significant.

At level 2, “Improper management processes (X₂₁)” is a significant factor affecting the four secondary risk factors of “Unsafe supervision (X3)”. It has the greatest influence on “Supervision violations of safety management personnel (X₃₄)”, followed by “Failed to correct problem (X₃₃)”, with a correlation degree of 0.7894 and 0.7492, respectively. The “Poor safety culture (X₂₂) mainly influences the “Inadequate supervision of operation process (X₃₁)” and “Supervision violations of safety management personnel (X₃₄)”. “Improper resource management (X₂₃)” is an important reason for “Supervision violations of safety management personnel (X₃₄)”.

As can be seen from Table 12, in the four human factors of level 3, the average correlation degree (0.683) between “Failed to correct problem (X₃₃)” and the five human factors in the “Preconditions for Unsafe Acts (X₄)” is the highest, followed by the average correlation degree (0.6465) of “Supervision violations of safety management personnel (X₃₄)”. “Personnel readiness (X₄₅)” is most susceptible to the influence of the four human factors in “Unsafe supervision (X₃)”, with an average correlation of 0.7195. “Supervision violations of safety management personnel (X₃₄)” has the highest correlation with “Personnel readiness (X₄₅)”, reaching 0.8078. “Inadequate supervision of operation process (X₃₁)” is the key to influencing whether the “Physical environment (X₄₁)” of the confined space operation meets the operation standard, and it also has a great influence on the “Personnel readiness (X₄₅)”.

As can be seen from Table 13, “Operator states (X₄₃)” has the highest average correlation with the four human factors of “Unsafe acts (X₅)”. The “Violations (X₅₄)” and five human factors of “Preconditions for unsafe acts (X₄)” shows a significant causal relationship, with an average correlation degree of 0.7481. This suggests that “Violations (X₅₄)” is the most susceptible to the human factors in the upper level.

3.3. Discussion

The chi-square test, odds ratio analysis, and GCA are used to analyze the causal relationship between the upper and lower levels of the improved HFACS model of confined space accidents. The results obtained by the two methods show strong consistency, but there are some differences. The main consistencies are as follows:

(1)

In the chi-square test, there is a significant causal relationship between “Improper management processes (X₂₁)” and the human factors of “Unsafe supervision (X₃)”. In the results of the GCA, the average correlation degree between “Improper management processes (X₂₁)” and the human factors of “Unsafe supervision (X₃)” reaches 0.7353. Both suggest that “Improper management processes (X₂₁)” is the most critical factor affecting “Unsafe supervision (X₃)”. This reflects the significant role that management practices play in ensuring safety in confined spaces within industrial and trade enterprises. In such environments, where hazards are prevalent and conditions can change rapidly, effective management processes are essential for establishing safety protocols, providing training, and ensuring compliance with regulations. When management processes are inadequate, it can lead to a lack of proper supervision, insufficient risk assessments, and failure to enforce safety measures, thereby increasing the likelihood of unsafe supervision practices. This highlights the interconnectedness of management quality and supervisory effectiveness in preventing accidents in confined spaces.

(2)

In the chi-square test, “Failed to correct problem (X₃₃)” has a significant causal relationship with “Physical environment (X₄₁)” and “Personnel readiness (X₄₅)”. In the results of the GCA, “Failed to correct problem (X₃₃)” also has the highest correlation with “Physical environment (X₄₁)” and “Personnel readiness (X₄₅)”. In confined space operations within industrial and trade enterprises, when safety hazards or issues arise, a failure to promptly take corrective measures by management can lead to a deterioration of the physical environment (such as equipment malfunctions, poor ventilation, etc.), thus affecting operational safety. At the same time, personnel readiness may also be compromised, as employees may lack adequate training and emergency response capabilities when faced with danger.

In general, the chi-square test and odds ratio analysis can be used to test whether there is a significant causal relationship between the upper level and the lower level of the established HFACS model, while GCA can reflect the mutual influence relationship between the upper and lower levels. For confined space accidents, the human factors involved are complex and diverse, and the correlations between them are also numerous. Combining the two analysis methods can more effectively analyze the key human factors affecting the confined space accidents. There are 17 groups of significant causal relationships in the chi-square test and odds ratio analysis, of which 11 groups have correlation degrees above 0.7 in the GCA. These 11 groups of significant causal relationships are shown in Figure 7.

By combining the results obtained by the two methods, three paths leading to confined space accidents across the five levels of the established HFACS model can be finally obtained, as shown in Figure 8. The three accident paths are the top three paths of accident probability in the chi-square test analysis, indicating that these three paths are the main paths of confined space accidents. Therefore, the prevention of confined space accidents should focus on strengthening the control of the human factors in the three paths, so as to cut off the basic path of accident development.

To improve confined space safety, enterprises and regulatory bodies could take the following concrete measures: First, establish regular safety reviews and training programs to ensure employees are aware of potential risks and response measures. Second, implement stricter management processes to ensure timely correction of safety hazards and regular assessments of the physical environment’s safety. Finally, it is recommended that regulatory bodies develop relevant policies to encourage enterprises to implement standardized safety management systems, particularly in confined space operations, ensuring that all personnel receive necessary training and certification. These interventions will help reduce risks associated with the identified key factors.

4. Conclusions

In this paper, an improved HFACS model is established according to the characteristics of confined space accidents, and human factors identification is carried out in confined space operations. The chi-square test and grey correlation analysis are used to analyze the causal relationship between the human factors. The specific conclusions are as follows:

(1)

Based on the traditional HFACS model, an improved HFACS model with “external safety level” is proposed. Based on this model, the risk identification of confined space operations is carried out, and a total of 5 primary risk factors, including external safety level, organizational influence, unsafe supervision, preconditions for unsafe acts and unsafe acts are obtained, and 17 secondary risk factors such as political and economic factors and improper management processes are obtained.

(2)

Based on mathematical statistical methods such as the chi-square test and grey correlation analysis, the obvious influence and influenced relationship between the levels of the HFACS model are studied. There are 11 groups of significant causal correlation between risk factors. Among them, the causal relationship between “Supervision violations of safety management personnel (X₃₄)” and “Personal readiness (X₄₅)” is the most significant.

The conclusions of this study significantly contribute to confined space safety research by identifying critical human factors and their interrelations that lead to accidents. These findings not only enhance the understanding of the underlying causes of such incidents but also provide a foundation for future studies to explore targeted interventions and preventive measures. By emphasizing the importance of effective management processes and personnel readiness, this research offers valuable insights for policymakers and industry practitioners, potentially influencing safety regulations and operational practices to mitigate risks in confined spaces. Ultimately, the results aim to foster safer working environments, thereby reducing the incidence of accidents and enhancing overall workplace safety.

The limitations of this study mainly include the restricted sample range and insufficient consideration of the dynamic nature of the accident environment. Future research directions could focus on expanding the sample size to validate the model’s generalizability, combining quantitative and qualitative analyses to delve deeper into human factors, and exploring the effectiveness of new evaluation models in reducing the risks associated with confined space accidents. These studies will provide important support for further understanding and mitigating the risks of confined space accidents.