A Hybrid DEMATEL–ISM–CN–BN Framework for Explosive Risk Analysis in Hydrogenation Processes

Abstract

This paper proposes an integrated DEMATEL–ISM–CN–BN framework for modeling the coupling and evolution of explosion risks in hydrogenation processes. The novelty of this approach lies in its systematic integration of four distinct methodologies, combining static structural analysis with dynamic probabilistic reasoning to comprehensively capture risk interactions across multiple system levels. Specifically, the framework identifies risk factors, reveals causal and system-level hierarchical propagation pathways, and quantifies risk evolution. Applied to a diesel hydrogenation unit, the model captures cross-level interactions involving reactor pressure, compressor anomalies, and flow reversals. A composite weighting system—based on centrality (α), hierarchy (β), and clustering (θ)—facilitates risk prioritization. Sensitivity analysis and the sensitivity–importance (SI) matrix highlight critical components such as the hydrofinishing reactor and hydrogen compressors. The results demonstrate the framework's capability to quantify complex interdependencies and support targeted risk mitigation in high-risk chemical operations.

Keywords

coupling mechanism evolutionary paths DEMATEL-ISM complex networks BN

1. Introduction

With the increasing global demand for clean energy and the ongoing transformation of the energy industry, hydrogenation processes have been increasingly applied to the petrochemical sector (Cui et al., 2024; Mateus-Rubiano et al., 2024). These processes convert various feedstocks into high-quality fuels via catalytic hydrogenation and play a pivotal role in national energy strategies (Qian et al., 2022; Sahu et al., 2024). However, due to the inherent high-pressure and high-temperature operating conditions, as well as the flammable and explosive properties of hydrogen, hydrogenation systems exhibit complex accident evolution mechanisms, including multi-factor coupling and nonlinear risk propagation (Rahim et al., 2024). Incomplete risk identification has become a major contributor to severe fire and explosion accidents in hydrogenation units. Additionally, insufficient understanding of coupling mechanisms further exacerbates the complexity and severity of such incidents (Nazerifard et al., 2023).

Existing studies have explored risk factor identification in hydrogenation processes from multiple perspectives (Ishola et al., 2020). For instance, fuzzy linguistic approaches have been applied to develop hierarchical risk models for refinery units, and process hazard analysis (PHA) templates have been proposed for hydrogenation stages (Wolbers, 2024). However, fuzzy models generally suffer from subjectivity in expert judgment, limiting their objectivity and reproducibility. PHA templates, although systematic, often fail to capture dynamic changes or interactions between risk factors. Machine learning techniques have also been used to screen high-risk components (Elsherir et al., 2015; Rachman & Ratnayake, 2019), and physical simulation tools like ALOHA and PHAST have been employed for consequence analysis (Bariha et al., 2023). Yet, machine learning models depend heavily on data quality and quantity, making them challenging to apply in scenarios with limited historical incident records. Similarly, simulation tools such as ALOHA and PHAST primarily focus on single-event scenarios and rarely consider systemic interactions or cascading risks. Overall, current approaches often exhibit redundancy and fragmentation, failing to address complex multidimensional interactions among risk factors comprehensively. Most rely on static, fragmented analyses and lack dynamic modeling capabilities to elucidate cross-level risk transmission pathways, limiting their applicability in inherently hazardous systems (Yang et al., 2023a; 2023b).

To overcome these methodological gaps and address the limitations identified, this study proposes a comprehensive analytical framework by integrating four distinct methodologies: Decision-Making Trial and Evaluation Laboratory (DEMATEL), Interpretive Structural Modeling (ISM), Complex Networks (CN), and Bayesian Networks (BN) (Zhong & Zhang, 2025). DEMATEL is effective in quantifying causal relationships between factors (Bashan & Ust, 2019; Karasan & Kahraman, 2019; Tan & Zhang, 2020), while ISM clarifies hierarchical structures within systems (Chen et al., 2023; Liu et al., 2020) CN models contribute by revealing global topological structures and local clustering patterns, essential for identifying implicit critical nodes and characterizing time-varying risk migration (Li et al., 2025; Lu et al., 2025). BN enables probabilistic reasoning under uncertainty, allowing for dynamic risk evolution analysis (Feng et al., 2019; Li et al., 2023).

By synergistically combining DEMATEL, ISM, CN, and BN, this study develops a novel four-stage coupling model (DEMATEL-ISM-CN-BN), establishing a full-chain risk analysis methodology that integrates risk identification, relationship analysis, structural hierarchy, and probabilistic evolution. Specifically, DEMATEL quantifies causal interactions among risk factors (Lv et al., 2024); ISM reveals system-level hierarchical dependencies (Li et al., 2024a); CN captures global network topology and node clustering properties; BN models probabilistic risk evolution under uncertainty.

Furthermore, we introduce a composite risk prioritization index combining centrality (α), hierarchical importance (β), and clustering coefficient (θ), along with a sensitivity–importance (SI) matrix to facilitate the identification of critical nodes and interactions. This integrated analytical framework is particularly suited to hydrogenation processes, as these systems frequently involve complex hierarchical fault propagation, nonlinear feedback loops, and intricate cross-subsystem interactions—features difficult to comprehensively analyze using any single approach alone.

The remainder of this paper is organized as follows: Section 2 describes the methodological foundation and computational procedures of the proposed framework. Section 3 presents the case study and analyzes the results. Section 4 summarizes the key findings and outlines directions for future research.

2. Research on the Risk Factor Analysis Method of Hydrogenation Process Explosion Accident Risk Factors

This section details the integrated DEMATEL-ISM-CN-BN framework used to model the coupling mechanisms and dynamic evolution of fire and explosion risks in hydrogenation units. The sequence DEMATEL-ISM-CN-BN is intentionally designed to reflect the logical flow of risk propagation analysis. DEMATEL first quantifies the pairwise causal influences among risk factors, which are then fed into the ISM to construct a hierarchical structural model. This hierarchy serves as the foundation for complex network (CN) analysis to reveal topological patterns and key nodes. Subsequently, the same structural framework is applied to the Bayesian Network (BN) to model the probabilistic evolution of risk (Figure 1).

Figure 1.

Research Idea of This Paper.

2.1 DEMATEL for Causal Relationship Quantification

Considering the high degree of correlation and coupling between the diesel hydrogenation process and its equipment, leading to significant interactions among risk factors, DEMATEL provides a comprehensive framework for analyzing the coupling of these risk factors (Du & Shen, 2024). The process involves the following steps:

Step 1: Determine the Direct Impact Matrix.

Step 1: Construct the initial direct-relation matrix A based on expert judgment using a predefined scale (e.g., 0 to 3 indicating no to strong influence).

\begin{aligned} A = [\begin{matrix} \dots & a_{1 n} \\ ⋮ & ⋱ & ⋮ \\ a_{n 1} & \dots & 0 \end{matrix}] = [a_{i j}]_{n \times n} \end{aligned}

(1)

Step 2: Normalize the direct-relation matrix to obtain the standardized matrix B ( $B = [b_{i j}]_{n \times n}$ ).

\begin{aligned} B = [b_{i j}]_{n \times n} = \frac{a_{i j}}{m a x (\sum_{j = 1}^{n} a_{i j})} \end{aligned}

(2)

Step 3: Compute the total influence matrix T by aggregating direct and indirect effects through matrix operations.

\begin{aligned} T = (B + B^{2} + \dots + B^{n}) = \sum_{i = 1}^{n} B^{i} = B (I - B)^{- 1} \end{aligned}

(3)

Where, I is the denotes matrix.

Step 4: Calculate the prominence (R + C) and relation (R - C) of each factor, where R is the sum of row values and C is the sum of column values in the total influence matrix.

\begin{aligned} R_{i} & = \sum_{j = 1}^{n} t_{i j} \end{aligned}

(4)

\begin{aligned} C_{i} & = \sum_{i = 1}^{n} t_{i j} \end{aligned}

(5)

Factors with exhibiting prominence and positive relation values are identified as key drivers of systemic risk propagation (Wang et al., 2024).

2.2 ISM for Hierarchical Structure Modeling

The Interpretive Structural Modelling (ISM) approach is one of the most widely used analytical methods in systems engineering theory. Considering the cascading nature of risk propagation in the hydrogenation process, this study employs ISM to stratify risk factors into different levels and establish a hierarchical model of risk factors within the hydrogenation process (Huo et al., 2023; Li et al., 2024b). The procedure involves the following steps:

Step 1: Calculate the Overall Impact Matrix.

The overall impact matrix H ( $H = [h_{i j}]_{n \times n}$ ) is obtained by summing the total impact matrix T with the identity matrix I:

\begin{aligned} H = T + I \end{aligned}

(6)

Step 2: Determine the Reachability Matrix (K).

The reachability matrix K is derived from the total influence matrix H. Determine a threshold to filter unimportant relations in the reachable matrix K. The threshold λ is usually the mean of the overall impact matrix H added to the standard deviation.

\begin{aligned} k_{i j} = {\begin{array}{l} k_{i j} = 1, h_{i j} \geq λ (i, j = 1, 2, \dots, n) \\ k_{i j} = 0, h_{i j} < λ (i, j = 1, 2, \dots, n) \end{array} \end{aligned}

(7)

Step 3: Construct the Hierarchical Structure.

Remove the corresponding rows and columns from K to identify the highest-level factor set. Repeat this process iteratively until all factors have been assigned levels. The sequence in which rows and columns are removed determines the final hierarchical network structure.

2.3 Complex Network Analysis for Topological Insights

Although the classical clustering coefficient is defined for undirected graphs, it is adapted here to evaluate the local interconnectivity of risk factors in a directed network. Following methods used by Hao et al. (2023), the neighborhood of each node is treated as undirected for the purpose of computing clustering coefficients, which provides a practical measure of local density and coupling strength in risk propagation analysis. Considering the strong chain coupling among the risk factors in the hydrogenation process. This study introduces complex network theory to calculate the clustering coefficients of the risk factors, as well as the degrees of the network nodes, including out-degree, in-degree, and total degree (Kang et al., 2024).

The out-degree number of edges extending from this node to other nodes ( $d_{i}^{o u t} = \sum_{j} x_{i \to j}$ ). The in-degree signifies the quantity of edges sent towards this node from other nodes ( $d_{i}^{i n} = \sum_{j} x_{j \to i}$ ). The total degree is determined by equation (8) (Shi et al., 2024).

\begin{aligned} d_{i} = d_{i}^{o u t} + d_{i}^{i n} \end{aligned}

(8)

The clustering coefficient reflects the local clustering of risk factors in a complex network. The agglomeration coefficient of risk factors is calculated as follows:

\begin{aligned} θ_{i} = \frac{2 l_{i}}{n (n - 1)} \end{aligned}

(9)

Where n denotes the number of neighboring nodes to node X_i, and l_i represents the number of connected edges between the neighboring nodes.

After determining the centrality (α), hierarchical weight values (β), and clustering coefficients (θ) of the risk factors, Eq. (11) is used to calculate the composite weight values (Hao et al., 2023).

\begin{aligned} ω_{i} = \frac{1}{3} \cdot (\frac{α_{i}}{\sum_{i}^{n} α_{i}} + \frac{β_{i}}{\sum_{i}^{n} β_{i}} + \frac{θ_{i}}{\sum_{i}^{n} θ_{i}}) \end{aligned}

(10)

2.4 Bayesian Network for Probabilistic Inference

BN (Bayesian Network) is employed to quantify the probabilistic influence of risk factors and simulate accident evolution. Based on the ISM structure (Hong & Sheng, 2023; Lixia & Si, 2024):

1.Root nodes represent fundamental causes;

2.Intermediate nodes correspond to transmission factors;

3.Leaf nodes represent direct accident scenarios.

In hydrogenation processes, the available datasets are often insufficient to populate a comprehensive training library. Additionally, expert-based methods are prone to subjective bias. To address these limitations and ensure reliable results, fuzzy set theory is adopted as the methodological basis for deriving the probability table. The CPT is subsequently determined by integrating expert knowledge with fuzzy set theory.

The sensitivity (Rov) of risk factors to a flaring accident is calculated based on Eq. (11), and the identification of key risk factors for the flaring accident scenario of a hydrogenation unit is systematically carried out (Zhang et al., 2024).

\begin{aligned} R o v (X_{i}) = \frac{P (U | E) - P (U)}{P (U)} \end{aligned}

(11)

By integrating the two attributes of importance and sensitivity of risk factors, the SI (sensitivity and importance) index is defined as the product of the degree of importance and the degree of sensitivity, as shown in equation (12).

\begin{aligned} S I_{i} = S_{i \times} I_{i} \end{aligned}

(12)

Where S_i and I_i represent the importance weight and sensitivity weight of the ith risk factor, respectively.

3 Case Study: Risk Analysis of a Diesel Hydrogenation Process Flaring Accident in a Petrochemical Company

3.1 Case Introduction

In this study, a diesel hydrofinishing unit with an annual capacity of 4 million tonnes from a petrochemical company in China is used as an example. A portion of the diesel hydrotreating process flow is illustrated in Figure 2. After the mixed feedstock oil passes through the feedstock oil buffer tank SR-1, it is heated in the heating furnace E-1, and then enters the hydrogenation reactor R-1 together with the mixed hydrogen. The reaction products are separated by the hot and cold high-pressure separator, and the resulting products then proceed to the refining stage. The final product is high-quality diesel fuel. The key equipment involved in this process includes the hydrogenation reactor, compressor, separator, and other associated components.

Figure 2.

Diesel Hydrogenation Process Flow Diagram.

3.2 Identification of Risk Factors for Hydrogenation Process Explosion Accidents

Risk factor identification was performed through a combination of historical documentation and expert-based hazard analysis. Data sources included piping and instrumentation diagrams (P&IDs), process flow diagrams (PFDs), accident investigation reports, and plant safety evaluations. Hazard and Operability Analysis (HAZOP) and expert interviews were utilized to systematically identify process deviations and assess their potential consequences within the diesel hydrotreating system.

As summarized in Table 1, nine primary equipment-level risk factors (denoted as C1–C9) were identified, including the feedstock buffer tank V101, hydrogen compressors, and the hydrofinishing reactor. Additionally, 28 process deviation events (F1–F28) were extracted as secondary risk factors. Together, these two groups form a hierarchical two-tier risk factor system employed in the subsequent modeling stages.

Table 1.
Hydrogenation Unit Fire or Explosion Risk Assessment System (A).

Primary Risk Factors Secondary Risk Factors Risk Type

Overpressure flash explosion in raw material buffer tank V101 C1 Low diesel flow F1 Flow-related

Raw material oil filter SR101 high differential pressure F2 Pressure-related

Raw Oil Filter SR101 Outlet Component Abnormal F3 Composition/Other

Raw material buffer tank V101 level is low F4 Level-related

Reaction feed pump P101 outlet flow countercurrent F5 Flow-related

Circulating hydrogen compressor K102 overpressure leakage C2 Cold and high pressure separator V1605 high level F6 Pressure-related

Circulating hydrogen desulphurization tower C101 liquid level high F7 Level-related

High pressure of circulating hydrogen compressor K102 F8 Pressure-related

Separation unit overpressure leakage C3 Cold and High Pressure Separator V1605 Low Liquid Level F9 Pressure-related

Cold and low pressure separator V1606 high pressure F10 Pressure-related

Hot High Pressure Separator V1604 Low Level F11 Pressure-related

Hydrofinishing reactor R-101 overpressure leakage C4 Hydro-finishing Reactor R-101 High Pressure F12 Pressure-related

Hydro-finishing reactor R-101 differential pressure high F13 Pressure-related

Flash explosion in reactant injection tank V105 C5 Reactor injection tank V105 low level F14 Level-related

Reactor injection tank V105 pressure high F15 Pressure-related

Reactor injection pump P102 outlet flow countercurrent F16 Flow-related

Flash explosion in liquid-rich flash tank V108 C6 Circulating Hydrogen Desulfurization Tower C101 Depleted Amine Liquid Flow Low F17 Flow-related

Amine Enrichment Flash Tank V108 high pressure F18 Pressure-related

Low level in circulating hydrogen compressor inlet separator tank V-111 F19 Level-related

Flash explosion C7 in lean buffer tank V107 C7 Lean Amine buffer tank V107 low level F20 Level-related

Lean Amine buffer tank V107 high pressure F21 Pressure-related

Circulating hydrogen desulphurization tower lean solvent pump P103 outlet flow countercurrent F22 Flow-related

Overpressure leakage in fresh hydrogen compressor K101 C8 New hydrogen compressor K101 interstage pressure high F23 Pressure-related

New hydrogen compressor secondary inlet distributor tank V110A high pressure F24 Pressure-related

New hydrogen compressor K101 outlet pressure high F25 Pressure-related

Hydrogen desulfurization stripper tower C201 overpressure leak C9 H₂S removal stripper C201 liquid level high F26 Level-related

H₂S removal stripper C201 pressure high F27 Pressure-related

H₂S removal stripper C201 liquid level low F28 Level-related

Primary Risk Factors	Secondary Risk Factors	Risk Type
Overpressure flash explosion in raw material buffer tank V101 C1	Low diesel flow F1	Flow-related
Raw material oil filter SR101 high differential pressure F2	Pressure-related
Raw Oil Filter SR101 Outlet Component Abnormal F3	Composition/Other
Raw material buffer tank V101 level is low F4	Level-related
Reaction feed pump P101 outlet flow countercurrent F5	Flow-related
Circulating hydrogen compressor K102 overpressure leakage C2	Cold and high pressure separator V1605 high level F6	Pressure-related
Circulating hydrogen desulphurization tower C101 liquid level high F7	Level-related
High pressure of circulating hydrogen compressor K102 F8	Pressure-related
Separation unit overpressure leakage C3	Cold and High Pressure Separator V1605 Low Liquid Level F9	Pressure-related
Cold and low pressure separator V1606 high pressure F10	Pressure-related
Hot High Pressure Separator V1604 Low Level F11	Pressure-related
Hydrofinishing reactor R-101 overpressure leakage C4	Hydro-finishing Reactor R-101 High Pressure F12	Pressure-related
Hydro-finishing reactor R-101 differential pressure high F13	Pressure-related
Flash explosion in reactant injection tank V105 C5	Reactor injection tank V105 low level F14	Level-related
Reactor injection tank V105 pressure high F15	Pressure-related
Reactor injection pump P102 outlet flow countercurrent F16	Flow-related
Flash explosion in liquid-rich flash tank V108 C6	Circulating Hydrogen Desulfurization Tower C101 Depleted Amine Liquid Flow Low F17	Flow-related
Amine Enrichment Flash Tank V108 high pressure F18	Pressure-related
Low level in circulating hydrogen compressor inlet separator tank V-111 F19	Level-related
Flash explosion C7 in lean buffer tank V107 C7	Lean Amine buffer tank V107 low level F20	Level-related
Lean Amine buffer tank V107 high pressure F21	Pressure-related
Circulating hydrogen desulphurization tower lean solvent pump P103 outlet flow countercurrent F22	Flow-related
Overpressure leakage in fresh hydrogen compressor K101 C8	New hydrogen compressor K101 interstage pressure high F23	Pressure-related
New hydrogen compressor secondary inlet distributor tank V110A high pressure F24	Pressure-related
New hydrogen compressor K101 outlet pressure high F25	Pressure-related
Hydrogen desulfurization stripper tower C201 overpressure leak C9	H₂S removal stripper C201 liquid level high F26	Level-related
H₂S removal stripper C201 pressure high F27	Pressure-related
H₂S removal stripper C201 liquid level low F28	Level-related

3.3 Research on Risk Factor Coupling Mechanism Based on DEMATEL -ISM-CN

3.2.1 Based on DEMATEL Risk Factor Centrality Calculations

To evaluate interdependencies among risk factors, the DEMATEL method was applied to the 28 sary risk events identified earlier. A structured questionnaire using a 4-point scale (0–3) was developed for the expert panel to assess the direct impact of factors on each other. The expert panel comprised two senior engineers with extensive professional experience, one corporate safety manager, two academic researchers specializing in process safety, and one on-site technician (Schedules 1). A consensus was reached through a structured Delphi procedure, whereby initial ratings were independently provided, followed by collaborative discussions to resolve significant discrepancies, culminating in the aggregation of the final assessment matrix.

Schedules 1.
Expert Information.

Expert	Position	Work Experience	Education	Age
Expert 1	Researcher	26	Bachelor	55
Expert 2	Senior Engineer	22	PhD	48
Expert 3	Senior Engineer	20	Master	42
Expert 4	Security Supervisor	18	Bachelor	54
Expert 5	Researcher	12	Master	40
Expert 6	Technician	10	Bachelor	34

Schedules 2.

Table of Data on the Weights of the First-Level Factors.

	Judgement Matrix	Eigenvalues	Consistency Test
	F₈	F₁₂	F₁₃	F₁₆	W’	$λ$
F₈	0.5	0.25	0.25	0.5	0.2083	0.0833
F₁₂	0.75	0.5	0.5	0.75	0.2917
F₁₃	0.75	0.5	0.5	0.75	0.2917
F₁₆	0.5	0.25	0.25	0.5	0.2083

Schedules 3.

Data Table of Secondary Factor Weights.

	Judgement Matrix	Eigenvalues	Consistency Test
a	F₈	F₆	F₇	F₁₁	F₉	F₁₀	W’	$λ$
	F₆	0.5	0.65	0.65	0.6	0.6	0.225	0.073
	F₇	0.35	0.5	0.3	0.25	0.25	0.1575
	F₁₁	0.35	0.7	0.5	0.45	0.45	0.1975
	F₉	0.4	0.75	0.55	0.5	0.5	0.21
	F₁₀	0.4	0.75	0.55	0.5	0.5	0.21
b	F₁₂	F₆	F₇	F₁₁	F₉	F₁₀	W’	$λ$
	F₆	0.5	0.65	0.65	0.6	0.6	0.225	0.073
	F₇	0.35	0.5	0.3	0.25	0.25	0.1575
	F₁₁	0.35	0.7	0.5	0.45	0.45	0.1975
	F₉	0.4	0.75	0.55	0.5	0.5	0.21
	F₁₀	0.4	0.75	0.55	0.5	0.5	0.21
c	F₁₃	F₆	F₇	F₁₁	F₉	F₁₀	W’	$λ$
	F₆	0.5	0.65	0.65	0.6	0.6	0.225	0.073
	F₇	0.35	0.5	0.3	0.25	0.25	0.1575
	F₁₁	0.35	0.7	0.5	0.45	0.45	0.1975
	F₉	0.4	0.75	0.55	0.5	0.5	0.21
	F₁₀	0.4	0.75	0.55	0.5	0.5	0.21
d	F₁₆	F₁₄	F₁₅	W’	$λ$
	F₁₄	0.5	0.3	0.4	0.5
	F₁₅	0.7	0.5	0.6

Schedules 4.

Data Table for Three Levels of Factor Weights.

	Judgement Matrix	Eigenvalues	Consistency Test
a	F₆	F₂₆	F₂₇	F₂₈	W’	$λ$
F₂₆	0.5	0.35	0.5	0.3083	0.0426
F₂₇	0.65	0.5	0.65	0.3833
F₂₈	0.5	0.35	0.5	0.3083
b	F₉	F₂₆	F₂₇	F₂₈	W’	$λ$
F₂₆	0.5	0.35	0.5	0.3083	0.0426
F₂₇	0.65	0.5	0.65	0.3833
F₂₈	0.5	0.35	0.5	0.3083

Schedules 5.

Four-Level Factor Weighting Data Table.

	Judgement Matrix	Eigenvalues	Consistency Test
a	F₁₂	F₄	F₅	W’	$λ$
F₄	0.5	0.35	0.425	0.0375
F₅	0.65	0.5	0.575
b	F₁₃	F₄	F₅	W’	$λ$
F₄	0.5	0.35	0.425	0.0375
F₅	0.65	0.5	0.575
c	F₂₆	F₁₉	F₂₂	W’	$λ$
F₁₉	0.5	0.75	0.625	0.0625
F₂₂	0.25	0.5	0.375
d	F₂₇	F₁₈	F₁₉	W’	$λ$
F₁₈	0.5	0.55	0.525	0.0125
F₁₉	0.45	0.5	0.475
e	F₂₈	F₁₉	F₂₂	W’	$λ$
F₁₉	0.5	0.75	0.625	0.0625
F₂₂	0.25	0.5	0.375

Schedules 6.

Table of Data on the Weights of the Five Levels of Factors.

	Judgement Matrix	Eigenvalues	Consistency Test
a	F₄	F₂	F₃	W’		$λ$
	F₂	0.5	0.55	0.525	0.0125
	F₃	0.45	0.5	0.475
b	F₅	F₂	F₃	W’		$λ$
	F₂	0.5	0.55	0.525	0.0125
	F₃	0.45	0.5	0.475
c	F₁₈	F₁₇	F₂₀	F₂₁	W’		$λ$
	F₁₇	0.5	0.7	0.65	0.3917	0.0563
	F₂₀	0.3	0.5	0.45	0.2917
	F₂₁	0.35	0.55	0.5	0.3167
d	F₁₉	F₁₇	F₂₀	F₂₁	W’		$λ$
	F₁₇	0.5	0.7	0.65	0.3917	0.0563
	F₂₀	0.3	0.5	0.45	0.2917
	F₂₁	0.35	0.55	0.5	0.3167
e	F₂₂	F₂₀	F₂₁	W’		$λ$
	F₂₀	0.5	0.45	0.475	0.0125
	F₂₁	0.55	0.5	0.525

Schedules 7.

Data Table for six Levels of Factor Weights.

	Judgement Matrix	Eigenvalues	Consistency Test
a	F₂	F₁	F₂₃	F₂₄	F₂₅	W’	$λ$
F₁	0.5	0.4	0.4	0.4	0.225	0.0638
F₂₃	0.6	0.5	0.65	0.65	0.2833
F₂₄	0.6	0.35	0.5	0.6	0.2542
F₂₅	0.6	0.35	0.4	0.5	0.2375
b	F₃	F₁	F₂₃	F₂₄	F₂₅	W’	$λ$
F₁	0.5	0.4	0.4	0.4	0.225	0.0638
F₂₃	0.6	0.5	0.65	0.65	0.2833
F₂₄	0.6	0.35	0.5	0.6	0.2542
F₂₅	0.6	0.35	0.4	0.5	0.2375

3.2.2 Hierarchical Structuring via ISM

Following DEMATEL analysis, Interpretive Structural Modeling (ISM) was applied to identify the hierarchical relationships among risk factors. Using the total influence matrix T, a reachability matrix was derived with a predefined threshold λ, determined by the mean and standard deviation of T. Iterative filtering of the matrix yielded a six-level hierarchical structure.

To quantify the relative importance of each factor in the hierarchy, a Fuzzy Analytic Hierarchy Process (FAHP) was employed. Pairwise comparison matrices were constructed within and across levels using expert judgment, and consistency checks were applied. The resulting hierarchical weights (β) are presented in Schedules 2 through 7. Factors F12, F13, and F5 emerged with high hierarchical importance, further validating their central roles (Yan et al., 2016).

3.2.3 CN-Based Calculation of Risk Factor Clustering Coefficients

The complex network model, developed based on DEMATEL–ISM analysis consists of 28 directed nodes and 126 weighted edges, where each node represents a risk factor associated with fire and explosion hazards in the hydrogenation process. The directed edges capture the magnitude and direction of influence between pairs of factors. The network exhibits a high degree of interconnectivity and feedback, with several nodes exhibiting multiple outgoing and incoming links, highlighting their central role in the propagation of risk. Each node represents a secondary risk factor, and directed edges denote causal links derived from matrix T. For each node, in-degree, out-degree, and total degree were computed, indicating the intensity and direction of interconnections. The local clustering coefficient (θ) was then calculated using Equation (9), offering insight into the extent of neighborhood interconnectivity.

3.2.4 Bayesian Network Based Risk Evolution Path Analysis for Hydrogenation Process

Based on the hierarchical structure derived in Section 3.3, a Bayesian Network (BN) model was constructed to evaluate the probabilistic evolution of combustion and explosion risks in the diesel hydrogenation process (Figure 3). The nodes are arranged in a three-tier structure derived from ISM analysis: root nodes (e.g., F1, F23–F25) represent initiating causes, intermediate nodes indicate propagation pathways, and leaf nodes correspond to direct accident consequences. The top event (T = 1) denotes the occurrence of a fire or explosion. This diagram reflects the logical and probabilistic dependencies among risk factors. The conditional probability tables (CPTs) were populated through a combination of sensor data from field operations and expert elicitation, utilizing fuzzy set theory to mitigate subjectivity in qualitative estimates.

Figure 3.

BN Modelling of Fire and Explosion Incidents in the Hydrogenation Process.

To assess the practical utility of the developed BN model, a historical incident was selected for validation. On March 12, 2018, a fire and explosion accident occurred in the feedstock buffer tank of a diesel hydrogenation unit at a petrochemical facility in Jiujiang, Jiangxi Province. According to the official accident investigation report, the event was triggered initiated by abnormal pressure fluctuations in the circulating hydrogen compressor. This initiated an interlock shutdown of the hydrogen feed pump, allowing high-pressure hydrogen to backflow into the feedstock buffer tank. The resulting overpressure breach led to ignition and explosion. This scenario corresponds directly to risk factors F5 (flow reversal at the feed pump) and F8 (compressor overpressure).

By updating the relevant nodes in the BN model to reflect this accident scenario, the posterior probability of the top event (T = 1) was recomputed using reverse inference. The resulting value of 90.74% indicates a high probability of accident occurrence under these conditions, demonstrating the model's effectiveness in replicating real-world accident dynamics. The posterior update is illustrated in Figure 5, which highlights the increased probabilities of upstream causal nodes following evidence injection.

Figure 4.

Causality Diagram of Risk Factors.

Figure 5.

Multi-Level Hierarchical Structure Model of Risk Factors for Hydrogenation Plants Fire and Explosion Accidents.

3.4 Results

Based on expert input, the direct influence matrix A was constructed, normalized, and extended into a total influence matrix T using matrix algebra (Table 2).

Table 2.
Combined Impact Matrix of Risk Factors T.

F1 F2 F3 … F26 F27 F28

F1 0.1005 0.0917 0.1839 … 0.0073 0.0085 0.0079

F2 0.0148 0.0846 0.0928 … 0.0133 0.0153 0.0121

F3 0.0923 0.0077 0.0923 … 0.0142 0.0164 0.0131

$⋮$ $⋮$ $⋮$ $⋮$ $⋮$ $⋮$ $⋮$ $⋮$

F26 0.0189 0.0016 0.2237 … 0.0241 0.1875 0.1012

F27 0.0112 0.0109 0.1297 … 0.0937 0.0312 0.0161

F28 0.0102 0.0048 0.1205 … 0.1018 0.0327 0.1891

	F1	F2	F3	…	F26	F27	F28
F1	0.1005	0.0917	0.1839	…	0.0073	0.0085	0.0079
F2	0.0148	0.0846	0.0928	…	0.0133	0.0153	0.0121
F3	0.0923	0.0077	0.0923	…	0.0142	0.0164	0.0131
$⋮$	$⋮$	$⋮$	$⋮$	$⋮$	$⋮$	$⋮$	$⋮$
F26	0.0189	0.0016	0.2237	…	0.0241	0.1875	0.1012
F27	0.0112	0.0109	0.1297	…	0.0937	0.0312	0.0161
F28	0.0102	0.0048	0.1205	…	0.1018	0.0327	0.1891

The results, summarized in Table 3, reveal that factors such as F12 (high pressure in R-101) and F13 (high differential pressure in R-101) exhibit the highest centrality, indicating their key role in propagating systemic risk. In contrast, factors like F7 and F11 possess low centrality, suggesting limited systemic influence.

Table 3.

Influenced Degree, Influencing Degree,Centrality and Causality Degree of Risk Factors.

Factors	Influenced Degree(Ci)	Influencing Degree(Ri)	Centrality(R + C)	Causality(R-C)
F1	0.0167	0.862	0.8787	0.8453
F2	0.3013	0.7078	1.0091	0.4065
F3	0.6428	0.8071	1.4499	0.1643
F4	0.8572	0.5874	1.4446	−0.2698
F5	0.876	0.6938	1.5698	−0.1822
F6	0.4537	0.4406	0.8943	−0.0131
F7	0.0203	0.3445	0.3648	0.3242
F8	0.7602	0.3334	1.0936	−0.4268
F9	0.0778	0.5406	0.6184	0.4628
F10	0.1688	0.4406	0.6094	0.2718
F11	0.0109	0.4445	0.4554	0.4336
F12	2.3206	0.95	3.2706	−1.3706
F13	2.0478	0.98	3.0278	−1.0678
F14	0.0929	0.4323	0.5252	0.3394
F15	0.0859	0.3383	0.4242	0.2524
F16	0.4384	0.3584	0.7968	−0.08
F17	0.1717	0.8384	1.0101	0.6667
F18	1.2763	0.7463	2.0226	−0.53
F19	0.6311	0.7704	1.2015	0.1407
F20	0.4291	0.7342	1.1633	0.3051
F21	0.5291	0.7342	1.2633	0.2051
F22	0.871	0.3079	1.1789	−0.5631
F23	0.1711	1.6734	1.8445	1.5023
F24	0.2443	1.2172	1.4615	0.9729
F25	0.322	1.0867	1.4087	0.7647
F26	0.4541	0.5631	1.0172	0.109
F27	0.523	0.4711	0.9941	−0.0519
F28	0.3541	0.5631	0.9172	0.209

The causal structure of the identified risk factors is visualized in Figure 4, which illustrates the direct and indirect influence pathways throughout the system. As shown, F12 (high pressure in hydrofinishing reactor R-101) and F13 (high differential pressure in R-101) exhibit the highest centrality values, identifying them as core nodes within the causal network. Their deviation has the potential to initiate multi-level cascading failures, particularly impacting downstream separation and compression units.

In contrast, factors such as F7 (low air feed flow) and F11 (low level in the hot high-pressure separator V1604) possess low centrality scores, indicating a limited impact on system-wide dynamics. The causality metric (R−C) further underscores the systemic risk posed by F23 (high pressure in new hydrogen compressor K101), which serves as a dominant source node capable of triggering chain reactions. Meanwhile, factors like F4 (low level in feedstock buffer tank) and F5 (reverse flow at feed pump outlet) exhibit high susceptibility (i.e., elevated C_i values), suggesting that they are easily perturbed by upstream anomalies and should be prioritized for real-time monitoring.

As shown in Figure 5, the hierarchical structure derived from the ISM model organizes the 28 risk factors into six distinct levels (L1–L6) according to their causal depth and propagation role. The lowest tier (L6) represents the root causes of hazardous events, such as upstream flow deficiencies (e.g., F1: low diesel flow) and compressor-related disturbances (e.g., F23–F25). These factors are latent in nature and exert indirect but long-range influence over downstream systems. In contrast, the highest tier (L1) includes direct triggering events, such as overpressure in the hydrofinishing reactor (F12) and flow countercurrent anomalies (F16), which are typically detected by alarms and serve as immediate precursors to accidents.

The intermediate levels (L2–L5) act as transitional causes, capturing the cascading interactions that transmit risk from root-level disturbances to surface-level failures. Notably, some factors such as F5 (feed pump flow reversal) and F22 (lean solvent pump backflow) appear in the middle layers and have both upstream and downstream connections, indicating their key roles in cross-level risk mediation. This layered structure reveals not only the vertical propagation logic but also the network centrality of certain nodes, offering a systemic perspective for prioritizing control measures.

As shown in Figure 6, F12 (Reactor High Pressure), F13 (Reactor Differential Pressure), and F5 (Flow Countercurrent at Feed Pump) are situated near the core of the network, suggesting their potential criticality. To quantify the structural importance of each node and identify key risk contributors, the clustering coefficient for each node is calculated using Equation (10). This provides a basis for understanding local density and identifying tightly coupled risk clusters within the system (Table 4).

Figure 6.

Cause-Effect Diagram of Risk Factors for Fire and Explosion Accidents at Hydrogenation Plants.

Table 4.

Centrality,Hierarchical Weights,Clustering Coefficient,Comprehensive Weights of Risk Factors.

Factors	Centrality	Hierarchical Weights	Clustering Coefficient	Comprehensive Weights
F1	0.8787	0.1313	0.5476	0.0299
F2	1.0091	0.3063	0.5238	0.0428
F3	1.4499	0.2771	0.5179	0.0450
F4	1.4446	0.2479	0.4444	0.0413
F5	1.5698	0.3355	0.4273	0.0482
F6	0.8943	0.1781	0.4503	0.0311
F7	0.3648	0.1247	0.6667	0.0270
F8	0.4554	0.1564	0.6659	0.0301
F9	0.6184	0.1663	0.5833	0.0305
F10	0.6094	0.1663	0.4997	0.0286
F11	1.0936	0.2083	0.3571	0.0331
F12	3.2706	0.2917	0.1601	0.0560
F13	3.0278	0.2917	0.1522	0.0534
F14	0.5252	0.0833	0.6667	0.0257
F15	0.4242	0.125	0.6637	0.0275
F16	0.7968	0.2083	0.4893	0.0332
F17	1.0101	0.1037	0.6579	0.0317
F18	2.0226	0.0693	0.2821	0.0309
F19	1.2015	0.1954	0.2463	0.0309
F20	1.1633	0.115	0.9167	0.0397
F21	1.2633	0.1256	0.9058	0.0412
F22	1.1789	0.0796	0.7492	0.0337
F23	1.8445	0.1653	0.5697	0.0422
F24	1.4615	0.1483	0.5803	0.0375
F25	1.4087	0.1386	0.5714	0.0361
F26	1.0172	0.1062	0.6004	0.0307
F27	0.9941	0.132	0.5979	0.0322
F28	0.9172	0.1062	0.6012	0.0297

The comprehensive weights of each risk factor—incorporating centrality (α), hierarchical weight (β), and clustering coefficient (θ)—was calculated using Equation (10), and the results are summarized in Table 3. To enable a more intuitive comparison of the risk factor weights, the results were visualized, as shown in Figure 7. The analysis identifies F12 and F13 as the most influential risk sources, with integrated weights of 0.0560 and 0.0534, respectively. These values affirm the dominant role of reactor pressure control in ensuring system integrity.

Figure 7.

Combined Importance of Risk Factors.

Other significant contributors include F5 (reverse flow at the reaction feed pump) and F23 (high pressure in K101 compressor), with respective scores of 0.0482 and 0.0422. These findings emphasize that both pressure-related and flow-related deviations represent critical failure modes that must be addressed through interlock logic and predictive maintenance.

As shown in Figure 8, based on the BN structure developed in Section 3.2.4, a top event (T = 1) representing a fire/explosion is assumed to have occurred, and evidence is injected into the network. The posterior probabilities of related risk factors (e.g., F5, F8, F12) are recalculated through reverse inference. The color-coded bars indicate the variation in probability before and after the evidence update, thereby highlighting the most causally influential factors. Notably, F12 and F13 exhibited the largest increases in posterior probability, reinforcing their causal dominance in triggering top-level failures. This probabilistic insight complements the structural centrality results and confirms the need for continuous monitoring and control of reactor operating parameters, especially under transient or upset conditions.

Figure 8.

A Priori and a Posteriori Probabilities of Risk Factors.

The Rate of Variation (RoV) was computed for each risk factor using Equation (11), quantifying the relative sensitivity of individual variables to changes in the target event's probability. As illustrated in Figure 9, several factors exhibit RoV values above the threshold of 2.00, including F6, F8, and F12. These high-RoV factors exert disproportionate influence on system behavior, meaning that minor deviations in their state can lead to significant shifts in the likelihood of fire or explosion scenarios.

Figure 9.

Sensitivity of Risk Factors.

Such findings underscore the importance of prioritizing these variables for enhanced monitoring strategies. From an engineering control perspective, they represent optimal nodes for the deployment of redundant sensors, dynamic alarm thresholds, and predictive diagnostics. Moreover, integrating high-RoV nodes into emergency response logic and real-time Bayesian updating could improve the system's resilience to evolving failure conditions.

3.5 Discussion

As shown in Table 5, the Sensitivity-Importance (SI) values of the risk factors were calculated using Equation (12). The SI values of the primary-level risk factors, representing equipment-level risks, were obtained by summing the SI values of the corresponding secondary risk factors.The results indicate that secondary risk factors with higher SI values are primarily concentrated in high-risk equipment, such as the hydrofinishing reactor R-101, the recirculating hydrogen compressor K102, and the reaction feed pump P101. These components are associated with key risks, including abnormal reactor pressures, flow reversal in pumping systems, and compressor parameter fluctuations. Furthermore, the high SI values of F12 and F13 reinforce previous findings on the sensitivity of the hydrofinishing reactor R-101, confirming that loss of pressure control constitutes a major accident risk in hydrogenation units. This conclusion aligns with existing literature, further validating the critical role of precise pressure regulation in maintaining system stability and mitigating accident risks (Dang et al., 2024).

Table 5.
SI index of Risk Factors.

Primary Risk Factors SI Secondary Risk Factors SI

Overpressure flash explosion in raw material buffer tank V101 C1 0.0189 Low diesel flow F1 0.0013

Raw material oil filter SR101 high differential pressure F2 0.0023

Raw Oil Filter SR101 Outlet Component Abnormal F3 0.0029

Raw material buffer tank V101 level is low F4 0.0044

Reaction feed pump P101 outlet flow countercurrent F5 0.008

Circulating hydrogen compressor K102 overpressure leakage C2 0.0189 Cold and high pressure separator V1605 high level F6 0.0076

Circulating hydrogen desulphurization tower C101 liquid level high F7 0.0024

High pressure of circulating hydrogen compressor K102 F8 0.0089

Separation unit overpressure leakage C3 0.0129 Cold and High Pressure Separator V1605 Low Liquid Level F9 0.0049

Cold and low pressure separator V1606 high pressure F10 0.0035

Hot High Pressure Separator V1604 Low Level F11 0.0045

Hydrofinishing reactor R-101 overpressure leakage C4 0.0226 Hydro-finishing Reactor R-101 High Pressure F12 0.0141

Hydro-finishing reactor R-101 differential pressure high F13 0.0085

Flash explosion C5 in reactant injection tank V105 0.0098 Reactor injection tank V105 low level F14 0.0019

Reactor injection tank V105 pressure high F15 0.0017

Reactor injection pump P102 outlet flow countercurrent F16 0.0062

Flash explosion C6 in liquid-rich flash tank V108 0.0018 Circulating Hydrogen Desulfurization Tower C101 Depleted Amine Liquid Flow Low F17 0.0003

Amine Enrichment Flash Tank V108 high pressure F18 0.0005

Low level in circulating hydrogen compressor inlet separator tank V-111 F19 0.001

Flash explosion C7 in lean buffer tank V107 0.0024 Lean Amine buffer tank V107 low level F20 0.0008

Lean Amine buffer tank V107 high pressure F21 0.0006

Circulating hydrogen desulphurization tower lean solvent pump P103 outlet flow countercurrent F22 0.001

Overpressure leakage C8 in fresh hydrogen compressor K101 0.002 New hydrogen compressor K101 interstage pressure high F23 0.0007

New hydrogen compressor secondary inlet distributor tank V110A high pressure F24 0.0006

New hydrogen compressor K101 outlet pressure high F25 0.0007

Hydrogen desulfurization stripper tower C201 overpressure leak C9overpressure leak C9 0.0061 H₂S removal stripper C201 liquid level high F26 0.002

H₂S removal stripper C201 pressure high F27 0.0021

H₂S removal stripper C201 liquid level low F28 0.002

Primary Risk Factors	SI	Secondary Risk Factors	SI
Overpressure flash explosion in raw material buffer tank V101 C1	0.0189	Low diesel flow F1	0.0013
		Raw material oil filter SR101 high differential pressure F2	0.0023
		Raw Oil Filter SR101 Outlet Component Abnormal F3	0.0029
		Raw material buffer tank V101 level is low F4	0.0044
		Reaction feed pump P101 outlet flow countercurrent F5	0.008
Circulating hydrogen compressor K102 overpressure leakage C2	0.0189	Cold and high pressure separator V1605 high level F6	0.0076
		Circulating hydrogen desulphurization tower C101 liquid level high F7	0.0024
		High pressure of circulating hydrogen compressor K102 F8	0.0089
Separation unit overpressure leakage C3	0.0129	Cold and High Pressure Separator V1605 Low Liquid Level F9	0.0049
		Cold and low pressure separator V1606 high pressure F10	0.0035
		Hot High Pressure Separator V1604 Low Level F11	0.0045
Hydrofinishing reactor R-101 overpressure leakage C4	0.0226	Hydro-finishing Reactor R-101 High Pressure F12	0.0141
		Hydro-finishing reactor R-101 differential pressure high F13	0.0085
Flash explosion C5 in reactant injection tank V105	0.0098	Reactor injection tank V105 low level F14	0.0019
		Reactor injection tank V105 pressure high F15	0.0017
		Reactor injection pump P102 outlet flow countercurrent F16	0.0062
Flash explosion C6 in liquid-rich flash tank V108	0.0018	Circulating Hydrogen Desulfurization Tower C101 Depleted Amine Liquid Flow Low F17	0.0003
		Amine Enrichment Flash Tank V108 high pressure F18	0.0005
		Low level in circulating hydrogen compressor inlet separator tank V-111 F19	0.001
Flash explosion C7 in lean buffer tank V107	0.0024	Lean Amine buffer tank V107 low level F20	0.0008
		Lean Amine buffer tank V107 high pressure F21	0.0006
		Circulating hydrogen desulphurization tower lean solvent pump P103 outlet flow countercurrent F22	0.001
Overpressure leakage C8 in fresh hydrogen compressor K101	0.002	New hydrogen compressor K101 interstage pressure high F23	0.0007
		New hydrogen compressor secondary inlet distributor tank V110A high pressure F24	0.0006
		New hydrogen compressor K101 outlet pressure high F25	0.0007
Hydrogen desulfurization stripper tower C201 overpressure leak C9overpressure leak C9	0.0061	H₂S removal stripper C201 liquid level high F26	0.002
		H₂S removal stripper C201 pressure high F27	0.0021
		H₂S removal stripper C201 liquid level low F28	0.002

As shown in Figure 10, the Sensitivity-Importance matrix is spilt into four groups which are corresponding to four classes by three contour lines of SI (Ma et al., 2022). The yellow contour at SI = 0.0034 represents the average SI value of all risk factors. By further dividing the factors into two subgroups and averaging their SI values, the red contour at SI = 0.007 and the blue contour at SI = 0.0014 were obtained. Three contours divide the matrix into risk priority zones: SI > 0.007 (high-risk areas), 0.0034 < SI ≤ 0.007 (medium-high risk area), 0.0014 < SI ≤ 0.0034 (medium-risk area), and SI ≤ 0.0014 (low-risk areas). Key factors (e.g., F12, F13) appear in the high-risk areas. This matrix offers a dual-metric basis for prioritizing control interventions.

Figure 10.

Importance-Sensitivity Matrix.

The results indicate that F12 (high pressure in hydrofinishing reactor R-101) has the highest SI value, indicating its dominant contribution to systemic risk. Factors like F17, F18, and F21 demonstrated the lowest SI values, implying limited influence on risk propagation. The SI matrix provides a quantitative foundation for prioritizing risk control measures. High-SI nodes should be equipped with real-time monitoring and fault-tolerant design elements, including redundant sensors, dynamic alarm thresholds, and periodic simulation testing. This dual-metric approach offers a more robust decision-making basis than relying solely on structural centrality or expert intuition.

4. Conclusion

This study developed an integrated DEMATEL–ISM–CN–BN framework to analyze the coupling mechanisms and dynamic evolution of fire and explosion risks in hydrogenation processes. By combining causal structure modeling, hierarchical decomposition, network topology analysis, and probabilistic inference, the framework effectively identified key risk factors such as reactor pressure anomalies and feed pump flow reversals. A composite weighting system and sensitivity–importance (SI) analysis enabled the prioritization of critical nodes, providing a quantitative basis for targeted risk control. The Bayesian model accurately reproduced a historical accident scenario, demonstrating strong predictive validity. The Sensitivity–Importance (SI) matrix provided a quantitative basis for prioritizing control strategies. Factors with Hydro-finishing Reactor R-101 High Pressure (F12), Hydro-finishing reactor R-101 differential pressure high (F13)—were identified as critical control targets requiring enhanced monitoring and fail-safe designs. This approach addresses limitations in conventional static assessments and offers practical guidance for optimizing safety strategies in complex chemical systems.

The proposed DEMATEL–ISM–CN–BN framework offers a structured approach to risk governance in hydrogenation systems. By identifying causally critical, hierarchically dominant, and topologically central nodes, the model enables targeted management of high-risk components such as reactor pressure and compressor failures. The SI matrix supports the prioritization of safety interventions, guiding the optimization of alarm thresholds and maintenance strategies. The Bayesian inference mechanism further enhances scenario simulation and emergency planning through probabilistic reasoning, aiding both pre-incident decision-making and post-incident analysis. Future work will focus on improving computational efficiency and automation to enhance the framework's industrial scalability.

Footnotes

Abbreviations

Acknowledgements

This work was supported by the TaiShan Scholar Youth Expert (tsqn202306132), Ministry of Education China University of Petroleum (East China) Independent Innovation Research Program Platform Project (24CX02022A).

ORCID iDs

Zihan Gao

Dongfeng Zhao

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Independent Innovation Research Program Platform Project, TaiShan Scholar Youth Expert, (grant number 24CX02022A, tsqn202306132).

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

Bariha

Ojha

D. K.

Srivastava

V. C.

, et al. (2023). Fire and risk analysis during loading and unloading operation in liquefied petroleum gas (LPG) bottling plant[J]. Journal of Loss Prevention in the Process Industries, 81, 5–7. https://doi.org/10.1016/j.jlp.2022.104928

Bashan

Ust

(2019). Application of fuzzy dematel method to analyse s-CO2 brayton power systems[J]. Journal Of Intelligent & Fuzzy Systems, 37(6), 8483–8498. https://doi.org/10.3233/JIFS-191133

Chen

Xin

Zhang

, et al. (2023). Decision model of mine shaft development mode based on ISM-GAHP-FCA[J]. Journal Of Intelligent & Fuzzy Systems, 44(1), 1323–1336. https://doi.org/10.101610.3233/JIFS-212119

Cui

Dikhtiarenko

Kulkarni

S. R.

, et al. (2024). Regulating the crude oil–to–chemical process in a multizone fluidized bed reactor using unconventional catalyst formulations[J]. Powder Technology, 438, 119573. https://doi.org/10.1016/j.powtec.2024.119573

Dang

Sun

Yang

, et al. (2024). A comprehensive risk assessment method of hydrogenation units integrating DEMATEL-TOPSIS-CM[J]. International Journal Of Hydrogen Energy, 79, 411–422. https://doi.org/10.1016/j.ijhydene.2024.01.094

Shen

(2024). Large-scale group hierarchical DEMATEL method with automatic consensus reaching[J]. Information Fusion, 108, 102411. https://doi.org/10.1016/j.inffus.2024.102411

Elsherir

Manan

Z. A.

Kamsah

M. Z.

(2015). State-of-the-art of hydrogen management in refinery and industrial process plants[J]. Journal Of Natural Gas Science And Engineering, 24, 346–356. https://doi.org/10.1016/j.jngse.2015.03.046

Feng

Jiang

J. C.

Liu

J. Q.

, et al. (2019). Airline safety assessment based on fuzzy mathematics and Bayesian networks[J]. Journal Of Intelligent & Fuzzy Systems, 37(6), 8577–8587. https://doi.org/10.3233/JIFS-190273

Hao

Zheng

Wang

, et al. (2023). Development of a method for weight determination of disaster-causing factors and quantitative risk assessment for tailings dams based on causal coupling relationships[J]. Stochastic Environmental Research and Risk Assessment, 37(2), 749–775. https://doi.org/10.1007/s00477-022-02316-w

10.

Hong

Sheng

(2023). A DEMATEL-ISM-BN model of mine water inrush accidents[J]. Mine Water And The Environment, 42(1), 178–186. https://doi.org/10.1007/s10230-022-00907-1

11.

Huo

Cong

Cheng

, et al. (2023). What is the driving mechanism for the carbon emissions in the building sector? An integrated DEMATEL-ISM model[J]. Energy, 274, 127399. https://doi.org/10.1016/j.energy.2023.127399

12.

Ishola

Matellini

D. B.

Wang

(2020). A proactive approach to quantitative assessment of disruption risks of petroleum refinery operation[J]. Safety Science, 127, 1107–1112. https://doi.org/10.1016/j.ssci.2020.104666

13.

Kang

Wang

Jin

, et al. (2024). Risk analysis and management of hydrogen station fire and explosion accidents using DEMATEL-ISM and complex network models[J]. International Journal of Hydrogen Energy, 136, 1230–1241. https://doi.org/10.1016/j.ijhydene.2024.02.141

14.

Karasan

Kahraman

(2019). A novel intuitionistic fuzzy DEMATEL - ANP - TOPSIS integrated methodology for freight village location selection[J]. Journal Of Intelligent & Fuzzy Systems, 36(2), 1335–1352. https://doi.org/10.3233/JIFS-17169

15.

Han

, et al. (2023). Flood risk assessment by using an interpretative structural modeling based Bayesian network approach (ISM-BN): An urban-level analysis of Shenzhen, China[J]. Journal Of Environmental Management, 329, https://doi.org/10.1016/j.jenvman.2022.117040

16.

Yang

Liu

, et al. (2024b). Factors of migrant children's participation in basic medical insurance in China: an empirical study based on DEMATEL-ISM method[J]. Frontiers In Public Health, 12, 4–5. https://doi.org/10.3389/fpubh.2024.1343860

17.

Yao

, et al. (2024a). A method for the core accident chain based on fuzzy-DEMATEL-ISM: An application to aluminium production explosion[J]. Journal of Loss Prevention in the Process Industries, 92, 105414. https://doi.org/10.1016/j.jlp.2024.105414

18.

She

, et al. (2025). Complex state networks based safety analysis of complex engineering systems considering closed-loop feedback[J]. Reliability Engineering & System Safety, 259, 110931. https://doi.org/10.1016/j.ress.2025.110931

19.

Liu

Kang

Yin

(2020). Construction of interpretive structure modeling for the influencing factors of emergency industry development[J]. Journal Of Intelligent & Fuzzy Systems, 38(6), 27–35. https://doi.org/10.3233/JIFS-179771

20.

Lixia

(2024). Human reliability of the intelligent construction site tower crane driver interface based on DEMATEL-ISM -BN[J]. PLOS ONE, 19(10). https://doi.org/10.1371/journal.pone.0303996

21.

, et al. (2025). Research on AI-driven complex network and management system of coal and gas outburst accident[J]. Journal of Safety and Sustainability, 2(1), 32–44. https://doi.org/10.1016/j.jsasus.2025.02.002

22.

Shi

Liu

(2024). Risk Assessment of Highway Infrastructure REITs Projects Based on the DEMATEL-ISM Approach[J]. Sustainability, 16(12), 5129. https://doi.org/10.3390/su16125159

23.

Lan

, et al. (2022). A methodology to assess the interrelationships between contributory factors to maritime transport accidents of dangerous goods in China[J]. Ocean Engineering, 266, 7–8. https://doi.org/10.1016/j.oceaneng.2022.112769

24.

Mateus-Rubiano

Castillo

A. C.

León

, et al. (2024). Effect of hydrotreatment process on the physicochemical properties of a Colombian heavy crude oil post-catalytic aquathermolysis[J]. Energy, 298, 131349. https://doi.org/10.1016/j.energy.2024.131349

25.

Nazerifard

Mohammadpourfard

Zeinali Heris

(2023). Design, thermodynamic and economic evaluation, and optimization of gasoline production from refinery furnaces flue gas[J]. Energy Conversion and Management, 293, 117492. https://doi.org/10.1016/j.enconman.2023.117492

26.

Qian

Song

, et al. (2022). The comparative life-cycle environmental effects of chemical feedstock change driven by energy system transition : A case study from China's ethylene glycol industry[J]. Journal Of Cleaner Production, 355, 131764. https://doi.org/10.1016/j.jclepro.2022.131764

27.

Rachman

Ratnayake

R. M. C.

(2019). Machine learning approach for risk-based inspection screening assessment[J]. Reliability Engineering & System Safety, 185, 518–532. https://doi.org/10.1016/j.ress.2019.02.008

28.

Rahim

M. S. A.

Reniers

Yang

, et al. (2024). Risk assessment methods for process safety, process security and resilience in the chemical process industry : A thorough literature review[J]. Journal Of Loss Prevention In The Process Industries, 88, 105274. https://doi.org/10.1016/j.jlp.2024.105274

29.

Sahu

Udit Kumar

Kumar

Aman

Pradhan

Ashis Kumar

(2024). Analyzing energy intensity of Indian Chemical industry : A panel data approach Energy Nexus, 13, 100279. https://doi.org/10.1016/j.nexus.2024.100279

30.

Shi

Liu

Feng

, et al. (2024). Evolutionary model and risk analysis of ship collision accidents based on complex networks and DEMATEL[J]. Ocean Engineering, 305, 117965. https://doi.org/10.1016/j.oceaneng.2024.117965

31.

Tan

Zhang

(2020). Multiple attribute decision making method based on DEMATEL and fuzzy distance of trapezoidal fuzzy neutrosophic numbers and its application in typhoon disaster evaluation[J]. Journal Of Intelligent & Fuzzy Systems, 39(3), 18–19. https://doi.org/10.3233/JIFS-191758

32.

Wang

(2024). Risk analysis of human evacuation aboard passenger ships based on fuzzy DEMATEL-ISM -BN[J]. Ocean Engineering, 313, 119520. https://doi.org/10.1016/j.oceaneng.2024.119520

33.

Wolbers

(2024). An overview of PHA templates and checklists developed for a refining system and lessons learned[J]. Process Safety Progress, 43(1), 6–8. https://doi.org/10.1002/prs.12467

34.

Yan

, et al. (2016). Hazard assessment for biomass gasification station using general set pair analysis[J]. Bioresources, 11(4), 7–14. https://doi.org/10.15376/biores.11.4.8307-8324

35.

Yang

Cai

, et al. (2023a). Digital twin-driven fault diagnosis method for composite faults by combining virtual and real data[J]. Journal Of Industrial Information Integration, 33. https://doi.org/10.1016/j.jii.2023.100469

36.

Yang

Cai

Zhang

, et al. (2023b). Cross-validation enhanced digital twin driven fault diagnosis methodology for minor faults of subsea production control system[J]. Mechanical Systems And Signal Processing, 204, 110813. https://doi.org/10.1016/j.ymssp.2023.110813

37.

Zhang

Cheng

, et al. (2024). Risk analysis of underground debris flows in mines based on a coupled weighted Bayesian network[J]. International Journal Of Disaster Risk Reduction, 114, 104922. https://doi.org/10.1016/j.ijdrr.2024.104922

38.

Zhong

Zhang

(2025). Schedule risk analysis of prefabricated building projects based on DEMATEL-ISM and bayesian networks[J]. Buildings, 15(3), 508. https://doi.org/10.3390/buildings15030508

A Hybrid DEMATEL–ISM–CN–BN Framework for Explosive Risk Analysis in Hydrogenation Processes

Abstract

Keywords

1. Introduction

2. Research on the Risk Factor Analysis Method of Hydrogenation Process Explosion Accident Risk Factors

2.1 DEMATEL for Causal Relationship Quantification

2.3 Complex Network Analysis for Topological Insights

3.1 Case Introduction

3.2.1 Based on DEMATEL Risk Factor Centrality Calculations

Schedules 1. Expert Information.

3.2.3 CN-Based Calculation of Risk Factor Clustering Coefficients

3.2.4 Bayesian Network Based Risk Evolution Path Analysis for Hydrogenation Process

Footnotes

Abbreviations

Acknowledgements

ORCID iDs

Funding

Declaration of Conflicting Interests

References

Schedules 1.
Expert Information.