Integrating STPA with DEMATEL for identification of critical risk factors and transmission path in chemical process system Authors: Hongli Wang, Xiangyao Li, Bosi Zhang Published: 2025 ABSTRACT The chemical industry occupies a critical position in the national economy, but its complex production process leads to many risks in the chemical process system. From the system theory perspective, this study proposes a risk analysis method that integrates Systematic-Theoretical Process Analysis (STPA) and improved decision-making test and evaluation laboratory interpretation structural model (DEMATEL-ISM) for chemical process systems. First, the potential risk factors of the system are comprehensively identified by STPA, including the risks of equipment failure, operation error and element interaction anomaly; Then, the interactions of the risk factors are quantitatively analyzed by the improved DEMATEL-ISM method. By introducing the Maximum Mean Entropy Decrease (MMDE) algorithm, which constructs the hierarchical structure of the risk factors, revealing the critical risk points and the transmission paths. The results show that the temperature sensor failure and control algorithm error are the critical cause factors. Moreover, the actuator failure is the critical result factor, and the risk transfer path matches the actual operation logic of the system. This methodology is applied and validated through a case study of a typical ammonia synthesis process system. The study provides a scientific risk identification and analysis tool for safety management in high-hazard chemical process industries, such as petrochemicals and fertilizer production. Its application can assist enterprises and regulators in prioritizing safety investments, preventing major accidents, and minimizing associated production losses, environmental damage, and societal costs, thereby promoting sustainable economic development. Introduction As a pillar industry of the national economy, the chemical industry plays a central role in promoting economic development and social progress, and its production process covers complex chemical reactions, high-temperature and high-pressure environments, and large-scale application of hazardous chemicals, which collectively pose significant safety risks to chemical process systems. According to statistics, chemical accidents have occurred frequently at home and abroad in recent years, such as the 2020 Rizhao Chemical Plant in Shandong Province, resulted from uncontrolled reactor temperature and caused seven fatalities. Similarly, the 2019 “3·21” accident in Jiangsu involving improper management of nitrification wastes led to 78 deaths, underscoring the complexity and urgency of risk management in chemical production. Furthermore, statistical analyses of accident root causes underscore the criticality of the risk factors identified in this study. For instance, a review of incident reports by the U.S. Chemical Safety Board (CSB) highlights that instrumentation failures (e.g., sensors) and control system deficiencies are frequent contributors to major chemical process incidents1. Similarly, a statistical analysis of accidents in the Chinese chemical industry indicates that equipment failure, particularly of instrumentation and safety devices, constitutes the most significant category of initiating events, often leading to operator errors based on inaccurate data2. These real-world observations provide a strong justification for focusing on sensor failures and control algorithm errors, confirming that our model addresses empirically significant risks rather than being solely theoretically derived. Given the severe consequences of such accidents, there is an urgent need to develop more systematic and comprehensive risk identification methods. Traditional risk analysis techniques, such as Fault Tree Analysis (FTA) and Hazard and Operability Study (HAZOP), which focus primarily on single-component failures, have proven effective in identifying certain potential risks. However, as chemical production systems grow increasingly complex, these methods are gradually exposing their limitations3. They often fail to adequately capture interactive risks among system components, such as cascading failures caused by equipment coupling, human-machine interactions, and environmental disturbances. Similarly, Daas and Innal highlighted the inadequacy of traditional methods in cost-benefit-safety analysis for firewater systems and introduced a fuzzy DEMATEL approach to enhance systematic decision-making, further underscoring the limitations of conventional techniques in handling complex system interactions4. For instance, Perrow’s “Normal Accident Theory”5 points out that structural coupling and interaction complexity in complex systems can lead to unpredictable accident chains. Traditional methods are often overstretched when dealing with the nonlinear risks triggered by equipment coupling, human-machine interactions, and environmental disturbances in chemical process systems. Their lack of a holistic system perspective makes it easy to overlook hidden risk points and difficult to comprehensively analyze the root causes and propagation mechanisms of risks. Wang6 and Ji et al.7 emphasize multidimensional risk factors (human errors, equipment failures, process instability) and advocate for data-driven models (e.g., DQRA + Bi-LSTM) to quantify risks and predict failures in real time, leveraging time-series data for accuracy. Wang et al.8,9 and Sun et al.10 advance dynamic resilience frameworks using IoT, big data, and hybrid models (MFM-HAZOP-FDBN) to simulate cascading failures and evaluate protective layers (IPLs), addressing the need for multi-disaster coupling analysis and real-time monitoring. Chen et al.11 further supplement this by systematically classifying approaches for modeling domino effects in process industries, highlighting the limitations of traditional methods in capturing human-organizational interactions and proposing integrated physical-systemic management strategies. Yuan et al.12 and Hui et al.13 propose integrated safety-security barrier models and three-dimensional risk early warning systems that merge physical, cyber, human, and environmental factors, using fuzzy evaluation and Bayesian updating to enhance holistic risk quantification and adaptive control. Ahammed et al.14 and Mithun et al.15 utilize covariate-based fault tree analysis and dynamic risk assessment (DRA) with Bayesian updating to incorporate environmental/operational variables and time-varying risks, enabling data-driven, context-specific risk control in flexible manufacturing. Collectively, these studies underscore the shift toward systemic, data-intensive, and adaptive risk management but acknowledge persistent challenges in capturing nonlinear interactions and emerging risks (e.g., cyber-physical threats). Since current risk analysis techniques still fall short in addressing the complexity and interactivity of modern chemical production systems, they remain prone to overlooking potential risks. Systematic-Theoretical Process Analysis (STPA), which is based in systems theory and cybernetics, effectively overcomes the limitations of traditional reliability analysis. By focusing on the system as a whole and treating safety as a control problem, it addresses both component failures and interaction anomalies. STPA has achieved remarkable results in risk identification for complex systems such as aerospace and transportation16–19. To overcome these limitations, this study takes a typical chemical process system, ammonia synthesis process system, as an object, and proposes a risk analysis method that integrates STPA with the Improved Decision-Making Test and Evaluation Laboratory-Interpretation Structural Modelling (DEMATEL-ISM). First, potential risk factors within the system are comprehensively identified using STPA, including equipment failures, operational errors, and interaction risks. Subsequently, the Maximum Mean Entropy Decrease (MMDE) algorithm is introduced to scientifically determine the threshold for DEMATEL-ISM, enabling the construction of a risk propagation path model. This study aims to address the shortcomings of traditional methods in terms of systematic and quantitative analysis, as well as the objectivity of threshold determination, thereby providing an end-to-end solution from risk identification to path analysis for chemical process systems. Such an approach is expected to enhance the accuracy and effectiveness of safety management in chemical enterprises. The rest of this paper is structured as follows: Sect. Process of critical risk factor identification and propagation path modeling for chemical process systems describes the process of critical risk factor identification and propagation path modeling for chemical process systems; Sect. Case study: identification of systematic risk factors based on STPA focuses on the identification of systematic risk factors based on Systematic-Theoretical Process Analysis; Sect. Systematic risk analysis based on improved DEMATEL-ISM presents systematic risk analysis using the improved DEMATEL-ISM method; Sect. Results analysis analyzes the results obtained from the above methodologies; finally, Sect. Conclusion concludes the paper with key findings and future research directions. Process of critical risk factor identification and propagation path modeling for chemical process systems A method for identifying critical risk factors and modeling risk propagation paths in chemical process systems is proposed, which comprises two main stages: 1) In the first stage, potential risk factors in the chemical process system are analyzed using Systematic-Theoretic Process Analysis (STPA). This step aims to accurately identify the root causes of unsafe control actions (UCAs), thereby providing a solid theoretical foundation and data support for the subsequent development of practical risk prevention and control strategies; 2) In the second stage, critical risk factors and risk propagation paths are further analyzed using the improved DEMATEL-ISM approach. Based on the list of risk factors identified by STPA, these factors are input into the DEMATEL-ISM model for further risk analysis. Through DEMATEL analysis, the centrality and causality of factors are obtained, which aids in identifying the critical risk factors. The Maximum Mean Entropy Decrease (MMDE) algorithm is then applied to determine the threshold and compute the reachability matrix. Finally, based on the correlations among system-level hazards, unsafe control actions, and risk factors, the risk propagation paths that may trigger system-level hazards are derived using Interpretive Structural Modeling (ISM). The improved DEMATEL-ISM approach was selected for this study due to its unique capability to not only identify critical factors but also to reveal the complex causal relationships and hierarchical structure among them, which is essential for understanding risk propagation in a complex system. The relationship between the two integrated tools is synergistic: DEMATEL quantitatively analyzes the inter-influences between factors, classifying them into cause-and-effect groups and identifying those with high centrality. Subsequently, ISM utilizes the comprehensive influence matrix derived from DEMATEL to construct a clear, multi-level hierarchical structure, visually mapping the transmission paths of risks. This combination overcomes the limitation of DEMATEL, which identifies key factors but does not explicitly show the layered propagation paths, and the limitation of ISM, which typically relies on subjective binary judgments for building the reachability matrix. The specific the analysis process is shown in Fig. 1. Fig. 1Chemical Process System Critical Risk Factor Identification and Transfer Path Modelling. Chemical Process System Critical Risk Factor Identification and Transfer Path Modelling. Case study: identification of systematic risk factors based on STPA In this study, the ammonia synthesis process—a typical chemical process system—serves as the case for the application and validation of the proposed method. System description The ammonia synthesis process, a cornerstone of the chemical industry, operates under high-pressure and high-temperature conditions with hazardous materials, including hydrogen, nitrogen, and ammonia. This study examines a typical synthesis loop where nitrogen (N2) and hydrogen (H2) are catalytically combined to form ammonia (NH3). The integrated system consists of five key subsystems: Feed Gas Compression: High-purity N2 and H2 are supplied at a controlled stoichiometric ratio (approx. 1:3) and compressed to 150–300 bar for the synthesis reaction.Catalytic Reactor: The compressed gases react exothermically (N2 + 3H2 → 2NH3) in a multi-bed converter using an iron-based catalyst. Temperature control is critical to prevent catalyst damage and ensure reaction efficiency.Cooling & Heat Recovery: Hot reactor effluent is cooled through heat exchangers, recovering energy to preheat incoming feed gas while condensing ammonia.Product Separation: Liquid ammonia is separated from unreacted gases in a high-pressure separator. Unconverted N2 and H2 are recycled to the reactor to maximize yield.Control & Safety System: A centralized network of sensors, programmable controllers (e.g., DCS), and final control elements maintains process variables within safe limits and triggers emergency shutdowns if deviations occur. Feed Gas Compression: High-purity N2 and H2 are supplied at a controlled stoichiometric ratio (approx. 1:3) and compressed to 150–300 bar for the synthesis reaction. Feed Gas Compression: High-purity N2 and H2 are supplied at a controlled stoichiometric ratio (approx. 1:3) and compressed to 150–300 bar for the synthesis reaction. Catalytic Reactor: The compressed gases react exothermically (N2 + 3H2 → 2NH3) in a multi-bed converter using an iron-based catalyst. Temperature control is critical to prevent catalyst damage and ensure reaction efficiency. Catalytic Reactor: The compressed gases react exothermically (N2 + 3H2 → 2NH3) in a multi-bed converter using an iron-based catalyst. Temperature control is critical to prevent catalyst damage and ensure reaction efficiency. Cooling & Heat Recovery: Hot reactor effluent is cooled through heat exchangers, recovering energy to preheat incoming feed gas while condensing ammonia. Cooling & Heat Recovery: Hot reactor effluent is cooled through heat exchangers, recovering energy to preheat incoming feed gas while condensing ammonia. Product Separation: Liquid ammonia is separated from unreacted gases in a high-pressure separator. Unconverted N2 and H2 are recycled to the reactor to maximize yield. Product Separation: Liquid ammonia is separated from unreacted gases in a high-pressure separator. Unconverted N2 and H2 are recycled to the reactor to maximize yield. Control & Safety System: A centralized network of sensors, programmable controllers (e.g., DCS), and final control elements maintains process variables within safe limits and triggers emergency shutdowns if deviations occur. Control & Safety System: A centralized network of sensors, programmable controllers (e.g., DCS), and final control elements maintains process variables within safe limits and triggers emergency shutdowns if deviations occur. This high level of integration and automation makes the system vulnerable to cascading risks, where a localized fault can propagate rapidly through control loops and physical linkages. Define the purpose of the analysis To perform risk identification using STPA, it is essential to first define accidents and system-level hazards. Based on an in-depth analysis of the overall operational process of an ammonia synthesis system, along with the collection and organization of extensive historical accident data combined with process-specific characteristics—such as chemical reaction features, equipment operating parameters, and material properties—the primary accident types are summarized as follows: (1) Personnel poisoning or casualties (A1): Caused by ammonia leakage leading to poisoning, posing a serious threat to human life. (2) Equipment damage or failure (A2): Structural or performance degradation of equipment due to factors such as overheating, overpressure, or corrosion. (3) Environmental pollution (A3): Release of substances such as ammonia or hydrogen, contaminating air, soil, or water bodies. Further analysis yields a list of system-level hazards: (1) H1: Temperature control failure leading to overheating or explosion - Intense exothermic synthesis reactions can trigger explosions if temperature control is lost, threatening personnel, equipment, and environmental safety. (2) H2: Abnormal pressure control causing equipment rupture or leakage - Excessive pressure can breach equipment seals and structures, leading to leaks or even explosions. (3) H3: Failure of ammonia leakage detection and handling mechanisms - Inability to promptly detect and address ammonia leaks can result in toxic gas diffusion, increasing risks of personnel poisoning and environmental pollution. (4) H4: Raw material supply interruption or proportion imbalance - Disruption of reaction equilibrium due to supply issues reduces production efficiency and may damage equipment. Among these, H1 and H2 can potentially trigger accidents of types A1, A2, and A3; H3 is primarily linked to A1 and A3; and H4 typically leads to A2 and A3 incidents. Building a safety control structure The ammonia synthesis process system is designed to synthesize ammonia from nitrogen and hydrogen under specific catalytic conditions. The system comprises multiple tightly integrated subsystems-including raw material supply, reaction, cooling, separation, and control-which work in concert to ensure stable process operation. The raw material supply subsystem delivers nitrogen and hydrogen to the reactor via pumps and pipelines. Precise regulation of flow rate, pressure, and mixture ratio is essential for the smooth progression of the reaction. Within the reactor, nitrogen and hydrogen undergo complex chemical reactions in the presence of a catalyst to form ammonia. These reactions require stringent operating conditions, where temperature, pressure, and catalyst activity critically influence the process. The cooling subsystem serves to remove reaction heat in a timely manner, thereby controlling the reaction temperature and preventing catalyst deactivation and equipment damage. The separation subsystem employs techniques such as condensation and adsorption to effectively separate ammonia from unreacted gases, thereby improving raw material utilization. The control system utilizes temperature sensors, pressure sensors, flow meters, and other instrumentation to monitor system status in real time, with feedback signals sent to the controller to enable automated control. Using a typical ammonia synthesis process as an example, a safety control structure diagram (Fig. 2) has been developed based on its operational mechanism. Fig. 2Safety control structure. Safety control structure. Identification of unsafe control actions and causal analysis Based on the established safety control structure, unsafe control actions (UCAs) in the ammonia synthesis process system are comprehensively identified across four dimensions: not providing a required control action, providing a hazardous control action, incorrect timing of control action, and incorrect duration of control action. Using the reactor temperature controller, pressure controller, and flow controller as examples, the detailed identification results are summarized in Table 1. Table 1Unsafe control actions Identification.ControllersControlling behaviorControl behavior not providedProviding control behavior that leads to dangerImproperly timed provision of control behaviorInadequate duration of providing control behaviorReactor temperatureRegulating temperatureUnregulated temperature U1-1Misregulation of temperature U1-2Adjustment delay U1-3Inadequate duration of regulation U1-4Pressure ControllersRegulating pressurePressure U2-1 not regulatedFaulty regulation of pressure U2-2Adjustment delay U2-3Insufficient regulation time U2-4Flow ControllerRegulating flow rateUnregulated flow U3-1Faulty regulation of flow rate U3-2Adjustment delay U3-3Insufficient regulation time U3-4 Table 1Unsafe control actions Identification.ControllersControlling behaviorControl behavior not providedProviding control behavior that leads to dangerImproperly timed provision of control behaviorInadequate duration of providing control behaviorReactor temperatureRegulating temperatureUnregulated temperature U1-1Misregulation of temperature U1-2Adjustment delay U1-3Inadequate duration of regulation U1-4Pressure ControllersRegulating pressurePressure U2-1 not regulatedFaulty regulation of pressure U2-2Adjustment delay U2-3Insufficient regulation time U2-4Flow ControllerRegulating flow rateUnregulated flow U3-1Faulty regulation of flow rate U3-2Adjustment delay U3-3Insufficient regulation time U3-4 Unsafe control actions Identification. Referring to the STPA manual20, the general causal analysis framework of STPA has been refined and summarized, as illustrated in Fig. 3. The causes of UCAs are analyzed in depth across multiple levels, including the controller, control pathways, controlled process, and feedback information. By integrating the relationships between UCAs and system-level hazards, potential risk factors are consolidated. Using “temperature control failure leading to overheating or explosion (H1)” as an example, the corresponding list of risk factors is presented in Table 2. Fig. 3STPA generic causal analysis framework. STPA generic causal analysis framework. Table 2List of risk factors associated with H1.Risk factorsSpecific descriptionSTPA generic causal categoriesAssociated UCATemperature sensor failure f1Sensor does not accurately measure temperatureInappropriate feedback receivedU1–2、U1–3、 U1–4Control algorithm error f2Flawed control algorithmInappropriate control algorithmU1–2;U2–2༛ U3–2Faulty actuator f3Faulty actuatorImproper execution of control behaviorU1–1、U1–2、U1–3、U1–4;Faulty cooling medium supply f4Insufficient or interrupted cooling medium flowImproper execution of control behaviorU1–1、U1–2、U1–3、U1–4Damage to reactor cooling structure f5Impact on reactor cooling efficiencyImproper execution of control interactionsU1–1、U1–2、U1–3、U1–4Faulty pressure sensor f6Inaccurate pressure measurement by sensorInappropriate feedback receivedU2–1、U2–2、U2–3、U2–4Flow sensor failure f7Incorrect flow rate data measured by the sensorReceiving inappropriate feedbackU3–1、U3–2、U3–3、U3–4Controller hardware failure f8Controller hardware failure, affecting the command outputController-related faultsU1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Control wiring deterioration f9Signal transmission is unstable or interrupted due to the aging of the wiringImproper execution of control behaviorU1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Catalyst poisoning f10Poisoning and deactivation of catalyst due to exposure to harmful impuritiesControl interaction improperly executed (affects response process)U1–1、U1–2、U1–3、U1–4Unstable raw gas quality f11High fluctuation of impurity content in raw gasControl input related risk factors (affecting the response process)U1–1、U1–2、U1–3、U1–4Cooling equipment fouling f12Scaling inside the cooling equipment reduces the cooling effectControl interaction improperly executedU1–1、U1–2、U1–3、U1–4Separation device failure f13Deterioration of separation effect affecting raw material circulation and product qualityImproper execution of control interactionsU2–3、U3–3Safety valve failure f14Safety valves cannot open properly to relieve pressure in case of overpressure.Improper execution of control behavior (failure of safety protection link)U2–1、U2–2、U2–3、U2–4Emergency stop system failuref15In case of emergency, the stopping program cannot be activated normallyImproper execution of control behaviorU1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Operator misuse f16Operators incorrectly set the control parameters or operate in violation of the lawInappropriate control algorithms (decision level)U1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Inaccurate calibration of instrumentation f17Various types of instruments are not calibrated regularly, and the measurement data deviation is large.Inappropriate feedback receivedU1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Ambient temperature too high or too low f18Extreme ambient temperatures affecting equipment performance and reaction processesOther inputs or disturbances (external environmental factors)U1–1、U1–2、U1–3、U1–4 Table 2List of risk factors associated with H1.Risk factorsSpecific descriptionSTPA generic causal categoriesAssociated UCATemperature sensor failure f1Sensor does not accurately measure temperatureInappropriate feedback receivedU1–2、U1–3、 U1–4Control algorithm error f2Flawed control algorithmInappropriate control algorithmU1–2;U2–2༛ U3–2Faulty actuator f3Faulty actuatorImproper execution of control behaviorU1–1、U1–2、U1–3、U1–4;Faulty cooling medium supply f4Insufficient or interrupted cooling medium flowImproper execution of control behaviorU1–1、U1–2、U1–3、U1–4Damage to reactor cooling structure f5Impact on reactor cooling efficiencyImproper execution of control interactionsU1–1、U1–2、U1–3、U1–4Faulty pressure sensor f6Inaccurate pressure measurement by sensorInappropriate feedback receivedU2–1、U2–2、U2–3、U2–4Flow sensor failure f7Incorrect flow rate data measured by the sensorReceiving inappropriate feedbackU3–1、U3–2、U3–3、U3–4Controller hardware failure f8Controller hardware failure, affecting the command outputController-related faultsU1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Control wiring deterioration f9Signal transmission is unstable or interrupted due to the aging of the wiringImproper execution of control behaviorU1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Catalyst poisoning f10Poisoning and deactivation of catalyst due to exposure to harmful impuritiesControl interaction improperly executed (affects response process)U1–1、U1–2、U1–3、U1–4Unstable raw gas quality f11High fluctuation of impurity content in raw gasControl input related risk factors (affecting the response process)U1–1、U1–2、U1–3、U1–4Cooling equipment fouling f12Scaling inside the cooling equipment reduces the cooling effectControl interaction improperly executedU1–1、U1–2、U1–3、U1–4Separation device failure f13Deterioration of separation effect affecting raw material circulation and product qualityImproper execution of control interactionsU2–3、U3–3Safety valve failure f14Safety valves cannot open properly to relieve pressure in case of overpressure.Improper execution of control behavior (failure of safety protection link)U2–1、U2–2、U2–3、U2–4Emergency stop system failuref15In case of emergency, the stopping program cannot be activated normallyImproper execution of control behaviorU1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Operator misuse f16Operators incorrectly set the control parameters or operate in violation of the lawInappropriate control algorithms (decision level)U1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Inaccurate calibration of instrumentation f17Various types of instruments are not calibrated regularly, and the measurement data deviation is large.Inappropriate feedback receivedU1–1、U1–2、U1–3、U1–4;U2–1、U2–2、U2–3、U2–4༛U3–1、U3–2、U3–3、U3–4Ambient temperature too high or too low f18Extreme ambient temperatures affecting equipment performance and reaction processesOther inputs or disturbances (external environmental factors)U1–1、U1–2、U1–3、U1–4 List of risk factors associated with H1. Systematic risk analysis based on improved DEMATEL-ISM DEMATEL-based analysis of critical factors The Decision-Making Trial and Evaluation Laboratory (DEMATEL) method21,22, which is based on graph theory and matrix theory, provides an effective approach for analyzing interactions among complex factors23,24. In this study, it is applied to analyze risk factors within the ammonia synthesis process system. The specific steps involved are as follows: Construct direct impact matrix H. Construct direct impact matrix H. Construct direct impact matrix H. The direct influence matrix H was constructed through a structured expert elicitation process. A panel of five experts, each with over ten years of experience in chemical process safety, was consulted. To ensure a consistent understanding, all experts were provided with detailed descriptions of the 18 risk factors identified in Sect. 3, along with a unified scoring guideline. Each expert performed pairwise comparisons to assess the direct influence between factors using a predefined five-level scale (as shown in Table 3): “Very High Influence” (4), “High Influence” (3), “Medium Influence” (2), “Low Influence” (1), and “No Influence” (0). For instance, the influence of temperature sensor failure on control algorithm error was rated as “High” to “Very High,” reflecting that inaccurate temperature data can lead to erroneous control commands. The individual scores from all experts were aggregated using the arithmetic mean to form the final group direct influence matrix, thereby synthesizing collective judgment while mitigating the effect of individual biases. According to the degree of influence assigned to the construction of the direct impact matrix:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:H={{[h}_{ij}\:]}_{n\times\:n}$$\end{document} Table 3Linguistic terms and corresponding scores for expert assessment in DEMATEL.Linguistic TermAbbreviationScoreDescription of InfluenceNo InfluenceN0Factor i has no predictable influence on factor j.Low InfluenceL1Factor i has a weak influence on factor j.Medium InfluenceM2Factor i has a moderate influence on factor j.High InfluenceH3Factor i has a strong influence on factor j.Very High InfluenceVH4Factor i has a very strong influence on factor j.No InfluenceN0Factor i has no predictable influence on factor j.Low InfluenceL1Factor i has a weak influence on factor j. Table 3Linguistic terms and corresponding scores for expert assessment in DEMATEL.Linguistic TermAbbreviationScoreDescription of InfluenceNo InfluenceN0Factor i has no predictable influence on factor j.Low InfluenceL1Factor i has a weak influence on factor j.Medium InfluenceM2Factor i has a moderate influence on factor j.High InfluenceH3Factor i has a strong influence on factor j.Very High InfluenceVH4Factor i has a very strong influence on factor j.No InfluenceN0Factor i has no predictable influence on factor j.Low InfluenceL1Factor i has a weak influence on factor j. Linguistic terms and corresponding scores for expert assessment in DEMATEL. 2)Calculate the regular matrix N and the comprehensive impact matrix T. 2)Calculate the regular matrix N and the comprehensive impact matrix T. Calculate the regular matrix N and the comprehensive impact matrix T. Calculate the regular matrix N according to the formula (2) eliminating the differences in the scale and value of the original matrix, so that the data are comparable.2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:N=\frac{H}{\underset{1\le\:i\le\:n}{{max}}\left({\sum\:}_{j=1}^{n}{h}_{ij}\right)}$$\end{document} Then calculate the comprehensive impact matrix T according to the formula (3) to reflect the direct and indirect impacts between factors.3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:T\:=\:\underset{n\to\:\infty\:}{{lim}}{\left(N+{N}^{2}+\cdots\:+{N}^{n}\right)=N(I-N)}^{-1}$$\end{document} 3)Calculate the degree of influence, degree of being influenced, degree of centrality and degree of cause of each factor. 3)Calculate the degree of influence, degree of being influenced, degree of centrality and degree of cause of each factor. Calculate the degree of influence, degree of being influenced, degree of centrality and degree of cause of each factor. Calculate the sum of the elements of each row of the matrix T to get the degree of influence \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Ci$$\end{document}, reflecting the ability of the factor to influence the combination of the other factors; Calculate the sum of the elements of each column to get the degree of being influenced \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Di$$\end{document}, reflecting the degree of the factor affected by the other factors; degree of centrality \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{i} = C_{i} + D_{i}$$\end{document}, measuring the degree of importance of the factor in the system; degree of cause \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{i} = C_{i} + D_{i}$$\end{document}, judging the factor as a cause factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{i} > 0$$\end{document}or an effect factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{i} < 0$$\end{document}. 4)Drawing cause and effect diagrams and identifying critical factors. 4)Drawing cause and effect diagrams and identifying critical factors. Drawing cause and effect diagrams and identifying critical factors. Drawing cause and effect diagrams with centrality as the horizontal coordinate and causality as the vertical coordinate. Factors with a high degree of centrality have a critical position in the system, and positive and negative causality distinguish the types of factors. Accordingly, critical factors that have a significant impact on systemic risk are identified to provide a focus for risk prevention and control. Risk propagation path analysis based on MMDE-ISM Threshold calculation. Threshold calculation. Threshold calculation. Traditional threshold determination methods, such as the expert determination method, are highly subjective and time-consuming, and the statistically based methods may not match the actual distribution. In this study, the maximum mean entropy reduction (MMDE) algorithm25 is introduced to determine the unique threshold value based on the scientific calculation of information entropy. This approach significantly enhances the accuracy and efficiency of threshold identification. The MMDE algorithm calculates the screening threshold based on the entropy reduction value to assess the impact of adding new nodes on the existing influence relationships. The entropy reduction value is calculated as: 4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^{D} = H\left( {\frac{1}{n},\frac{1}{n}, \cdots ,\frac{1}{n}} \right) - H(p1,p2, \cdots ,pn)$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p_i}$$\end{document}is the frequency of the occurrence of the i th element in the set, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H\left( {\frac{1}{n},\frac{1}{n}, \cdots ,\frac{1}{n}} \right)$$\end{document} is the uniform distribution entropy, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H\left( {{p_1},{p_2}, \cdots {p_n}} \right)$$\end{document}. is the actual distribution entropy. The average entropy reduction value is:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H^{MDE}}=\frac{{{H^D}}}{{N\left( S \right)}}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N\left( S \right)$$\end{document}is the number of unique elements in the set S, the specific calculation steps are as follows: Extract the elements of the n order-integrated impact matrix T, sort them in descending order, and combine the element rank values to generate the ordered triple\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( {{t_{ij}},i,j} \right)$$\end{document}; Extract the values of i from the triad to form the set of sending nodes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T^{Di}}$$\end{document} and calculate their average entropy reduction \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H^{MDE}}$$\end{document}. m Taking values from 1 to n, we get n \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^{MDE}$$\end{document}values; Extract the first m elements sequentially from\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T^{Di}}$$\end{document}to form the set\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{m}^{{Di}}$$\end{document}, and find the set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^{MDE}$$\end{document} corresponding to the maximum value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^{MDE}$$\end{document}; Extract the values of j from the triples in\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{{\hbox{max} }}^{{Di}}$$\end{document}to form the set of receiving nodes\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T^{\operatorname{Re} }}$$\end{document}and repeat the above steps to get\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{{\hbox{max} }}^{{\operatorname{Re} }}$$\end{document}; Take the union of the triples\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{{\hbox{max} }}^{{Di}}$$\end{document}and\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{{\hbox{max} }}^{{\operatorname{Re} }}$$\end{document}and find the smallest value of\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t_{ij}}$$\end{document}in the union set, i.e. the threshold value. 2)Construction of Hierarchical Structure for Risk Factors. 2)Construction of Hierarchical Structure for Risk Factors. Construction of Hierarchical Structure for Risk Factors. On the basis of the comprehensive impact matrix T obtained by the DEMATEL method, the overall impact matrix G is calculated according to the formula\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G=I+T$$\end{document}by taking into account the effects of each factor itself. On this basis, the threshold value\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document} obtained by the MMDE algorithm is combined with the matrix G, which is simplified according to formula (6) to obtain the reachability matrix P. ⎩6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p_{ij}}=\left\{ {\begin{array}{*{20}{c}} {1,{\text{ }}{g_{ij}} \geqslant \lambda } \\ {0,{\text{ }}{g_{ij}}<\lambda } \end{array}} \right.$$\end{document} According to the reachability matrix P, the current reachable set\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R\left( {{u_i}} \right)$$\end{document}, the prior set\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S\left( {{u_i}} \right)$$\end{document}and the common set Y are calculated by the formula (7).7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{ {\begin{array}{*{20}{c}} {R\left( {{u_i}} \right)=\left\{ {{u_j}|{u_j} \in U,{p_{ij}}=1} \right\}} \\ {S\left( {{u_i}} \right)=\left\{ {{u_j}|{u_j} \in U,{p_{jij}}=1} \right\}} \\ {Y=\left\{ {{u_j}|R\left( {{u_j}} \right) \cap S\left( {{u_j}} \right)=R\left( {{u_j}} \right)} \right\}} \end{array}} \right.$$\end{document} Y The elements in are in the same level, and the corresponding rows and columns are crossed out in turn, and the calculation is repeated until all rows and columns have been processed. The elements in Y obtained first are the top factors and the ones obtained later are the bottom factors, so as to construct a hierarchical structure of risk factors and clearly present the hierarchical relationship between factors and their influence paths. 3)Methodological application and analysis of results. 3)Methodological application and analysis of results. Methodological application and analysis of results. The direct influence matrix of risk factors for the ammonia synthesis process system was constructed based on the DEMATEL method (Fig. 4). The MMDE algorithm program was written and run in Matlab environment, and the maximum value of was calculated to be 0.0197, and the maximum value of column was 0.04297, and the threshold value was determined to be 0.15, which led to the reachability matrix (Fig. 5). The calculated values for each row\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^{MDE}$$\end{document} and column\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H^{MDE}$$\end{document} are shown in Fig. 6. Fig. 4Direct Impact Matrix. Direct Impact Matrix. Fig. 5Accessibility Matrix. Accessibility Matrix. Fig. 6Calculated values. Calculated values. Following Eq. (3), the comprehensive influence matrix T was calculated. The prominence (D + R) and relation (D-R) for each factor were then derived from matrix T. As an illustrative excerpt, the calculated influence (D), influenced (R), centrality (D + R), and causality (D-R) values for the top five critical factors are presented in Table 4. These numerical outputs form the direct basis for identifying the critical factors and constructing the causal diagram. Based on the DEMATEL analysis process, the values of the attributes of Influence, Influenced, Centrality and Causality were calculated for each risk factor, and the results are shown below Table 4. Table 4Factor attribute value calculation results.Risk factorsDegree of influence(D)Influenced degree(R)Centrality(D + R)Causality(D-R)Sort by Center DegreeTemperature sensor error f13.22.15.92.32Control algorithm error f23.51.96.63.41Actuator error f32.43.06.3-0.93Faulty cooling medium supply f42.02.53.3-1.55Damage to reactor cooling structure f51.82.73.6-2.16Pressure sensor fault f62.22.33.81.34Faulty flow sensor f71.92.43.8-1.17Controller hardware failure f82.11.83.1-1.28Control wiring deterioration f91.72.02.8-1.29Catalyst poisoning f101.61.92.6-1.310Unstable raw gas quality f111.51.82.3-1.311Cooling equipment fouling f121.41.72.1-1.312Separation device failure f131.31.61.9-1.313Safety valve failure f141.21.51.7-1.314Emergency stop system failuref151.11.41.5-1.315Operator misuse f161.01.31.30.316Inaccurate calibration of instrumentation f170.91.20.80.917Ambient temperature too high or too low f180.81.10.3-1.918 Table 4Factor attribute value calculation results.Risk factorsDegree of influence(D)Influenced degree(R)Centrality(D + R)Causality(D-R)Sort by Center DegreeTemperature sensor error f13.22.15.92.32Control algorithm error f23.51.96.63.41Actuator error f32.43.06.3-0.93Faulty cooling medium supply f42.02.53.3-1.55Damage to reactor cooling structure f51.82.73.6-2.16Pressure sensor fault f62.22.33.81.34Faulty flow sensor f71.92.43.8-1.17Controller hardware failure f82.11.83.1-1.28Control wiring deterioration f91.72.02.8-1.29Catalyst poisoning f101.61.92.6-1.310Unstable raw gas quality f111.51.82.3-1.311Cooling equipment fouling f121.41.72.1-1.312Separation device failure f131.31.61.9-1.313Safety valve failure f141.21.51.7-1.314Emergency stop system failuref151.11.41.5-1.315Operator misuse f161.01.31.30.316Inaccurate calibration of instrumentation f170.91.20.80.917Ambient temperature too high or too low f180.81.10.3-1.918 Factor attribute value calculation results. Based on the data in Table 4, a causal relationship diagram of risk factors is drawn (Fig. 7). Combine the reachability matrix and the correlation between UCA and risk factors to determine the risk transfer paths that may cause system-level hazards, and take “temperature control failure leading to over-temperature or explosion (H1)” as an example to draw a risk transfer path diagram (Fig. 8). Fig. 7Risk factor causality diagram. Risk factor causality diagram. Fig. 8Risk transmission pathways. Risk transmission pathways. Results analysis Analysis of core risk transmission paths To empirically validate the risk transmission paths identified by our model, we compared the results with a documented incident. The 2020 explosion at the Rizhao Chemical Plant in Shandong Province11, mentioned in the introduction, serves as a pertinent case. The official investigation report concluded that the direct cause was a failure to control the reactor temperature. This aligns precisely with our identified system-level hazard H1 (“Temperature control failure leading to overheating or explosion”). The report further detailed that a faulty temperature sensor provided inaccurate data, which led to erroneous outputs from the control system, ultimately resulting in insufficient cooling-a causal scenario that mirrors the path “Temperature sensor failure (f1) → Control algorithm error (f2) → Faulty cooling medium supply (f4) → H1” in our model (Fig. 8). This concordance between our model’s output and the findings of a real accident investigation significantly strengthens the credibility and practical relevance of our proposed methodology. Based on the causality diagram (Fig. 7), temperature sensor failure (f1) and control algorithm error (f2) were identified as critical cause factors, while actuator failure (f3) was determined to be the critical result factor. In the context of the ammonia synthesis process system, the failure of a temperature sensor leads to inaccurate temperature readings, causing the control algorithm to receive erroneous signals and output incorrect commands. This, in turn, affects actuator performance and compromises temperature control stability, thereby increasing the risk of overheating or explosion. The relatively low centrality of “operator misuse (f16)” aligns with the highly automated nature of modern ammonia synthesis processes. In such systems, core control functions—such as temperature and pressure regulation—are managed by automated components (f1, f2, f3), which diminishes the direct and frequent impact of individual operator actions compared to failures within automated control loops. This outcome underscores the model’s capacity to prioritize risks contextually, indicating that the most critical threats in this specific system lie within the automated control infrastructure, whereas human error remains more relevant during non-routine operations. Control algorithm error directly leads to deviations in control commands, resulting in system malfunction and elevated safety risks. Actuator failure, influenced by upstream elements such as sensors and control algorithms, may fail to execute commands correctly, directly causing temperature, pressure, or flow control failures and serving as a critical node for risk propagation. From the risk propagation path diagram (Fig. 8), critical factors are closely associated with multiple UCAs and involve numerous potential hazard pathways. Taking temperature sensor failure as an example, it is linked to UCA1-UCA4, which may lead to issues such as incorrect control algorithm adjustments and delayed actuator response, triggering overheating or explosion risks. This validates the reasonableness of identifying such factors as critical. Further causal scenario analysis illustrates that temperature sensor failure may cause the control algorithm to incorrectly adjust cooling medium flow based on wrong temperature data, leading to insufficient cooling and reactor overheating. Similarly, actuator failure may prevent control commands from being executed effectively, resulting in loss of temperature control and potential explosion. These causal scenarios align logically with the actual system operation and confirm the reliability of the identified risk propagation paths. A key point of discussion involves the uncertainty in DEMATEL analysis due to its reliance on expert judgment. Although measures were taken in Sect. 4.1 to enhance robustness, variability in input scores remains possible. The critical question is whether such variations would alter the identification of critical risk factors. The core findings are likely robust for two reasons. First, the identified critical cause factors—temperature sensor failure (f1) and control algorithm error (f2)—exhibit significantly higher centrality and cause values than most other factors (as shown in Table 4; Fig. 7). This substantial margin suggests that their top ranking is stable and would likely withstand minor input perturbations. Second, the logical consistency of these factors as primary drivers, along with their alignment with historical incident data, provides independent, non-expert validation of their criticality. Therefore, while the exact centrality and cause values may vary with different expert panels, the overarching conclusion regarding the critical importance of f1 and f2 remains well-supported. Based on the above analysis, to ensure the safe operation of the ammonia synthesis process system, the following measures are recommended: 1)During the design stage, enhance the reliability of temperature sensors and control algorithms through redundant design and algorithm optimization. 2)During the operation stage, strengthen equipment maintenance, perform regular sensor calibration and algorithm updates, and mitigate risk propagation from critical cause factors. 3)For critical result factors such as actuators, intensify equipment monitoring, maintenance, and operator training to ensure correct command execution. 4)For complex risk propagation paths, implement chain-breaking measures such as adding redundant control loops and optimizing control logic to block risk transmission. 5)For simpler paths, focus on preventing and controlling underlying critical factors-for example, conducting regular temperature sensor tests to ensure normal operation. Comparative analysis with traditional methods To substantiate the claim that the proposed method enhances risk identification accuracy, a comparative analysis with traditional methods is presented. This paper focus on the risks identified by this model that are likely to be missed or underemphasized by a conventional Fault Tree Analysis (FTA) or HAZOP study. Identification of Interactive and Emergent Risks: This paper analysis identified several critical risk factors stemming from component interactions, which are not single-component failures. For example: Catalyst poisoning (f10) affecting temperature control: In an FTA, catalyst poisoning might be modeled as a basic event affecting reactor efficiency. However, STPA-based approach explicitly links it to the failure of temperature control (UCA1-4 and H1), revealing how a chemical state change (catalyst activity) directly compromises a control function, a causal path that is an emergent property of the system’s interactions and not obvious in a component-centric FTA. Additionally, Unstable raw gas quality (f11) as a control input disturbance: This factor is identified as a “control input related risk factor” that disrupts the expected process response. A HAZOP might identify “deviation in feed composition,” but this method systematically traces its impact through the control loop (e.g., leading to improper control actions U1-1 to U1-4), providing a clearer propagation path to a system-level hazard.Explicit Modeling of Software and Algorithmic Flaws: A significant advantage of STPA is its treatment of control algorithm errors (f2) as a first-class risk factor. Traditional methods like FTA often lack the granularity to model flawed software logic effectively, typically subsuming it under a generic “controller failure” event. Our method explicitly identifies and highlights it as a critical cause factor with high centrality, directing management attention to a potentially overlooked area. Identification of Interactive and Emergent Risks: This paper analysis identified several critical risk factors stemming from component interactions, which are not single-component failures. For example: Catalyst poisoning (f10) affecting temperature control: In an FTA, catalyst poisoning might be modeled as a basic event affecting reactor efficiency. However, STPA-based approach explicitly links it to the failure of temperature control (UCA1-4 and H1), revealing how a chemical state change (catalyst activity) directly compromises a control function, a causal path that is an emergent property of the system’s interactions and not obvious in a component-centric FTA. Additionally, Unstable raw gas quality (f11) as a control input disturbance: This factor is identified as a “control input related risk factor” that disrupts the expected process response. A HAZOP might identify “deviation in feed composition,” but this method systematically traces its impact through the control loop (e.g., leading to improper control actions U1-1 to U1-4), providing a clearer propagation path to a system-level hazard. Identification of Interactive and Emergent Risks: This paper analysis identified several critical risk factors stemming from component interactions, which are not single-component failures. For example: Catalyst poisoning (f10) affecting temperature control: In an FTA, catalyst poisoning might be modeled as a basic event affecting reactor efficiency. However, STPA-based approach explicitly links it to the failure of temperature control (UCA1-4 and H1), revealing how a chemical state change (catalyst activity) directly compromises a control function, a causal path that is an emergent property of the system’s interactions and not obvious in a component-centric FTA. Additionally, Unstable raw gas quality (f11) as a control input disturbance: This factor is identified as a “control input related risk factor” that disrupts the expected process response. A HAZOP might identify “deviation in feed composition,” but this method systematically traces its impact through the control loop (e.g., leading to improper control actions U1-1 to U1-4), providing a clearer propagation path to a system-level hazard. Explicit Modeling of Software and Algorithmic Flaws: A significant advantage of STPA is its treatment of control algorithm errors (f2) as a first-class risk factor. Traditional methods like FTA often lack the granularity to model flawed software logic effectively, typically subsuming it under a generic “controller failure” event. Our method explicitly identifies and highlights it as a critical cause factor with high centrality, directing management attention to a potentially overlooked area. Explicit Modeling of Software and Algorithmic Flaws: A significant advantage of STPA is its treatment of control algorithm errors (f2) as a first-class risk factor. Traditional methods like FTA often lack the granularity to model flawed software logic effectively, typically subsuming it under a generic “controller failure” event. Our method explicitly identifies and highlights it as a critical cause factor with high centrality, directing management attention to a potentially overlooked area. This comparative discussion demonstrates that the STPA-DEMATEL-ISM framework provides a more comprehensive and systemic risk profile by explicitly capturing interaction-based and control-centric risks that traditional, linear methods are not designed to systematically uncover. Conclusion This study integrates STPA with the improved DEMATEL-ISM method to analyze risks in the ammonia synthesis process system, and the following results were obtained: 1) Constructed a safety control structure for the ammonia synthesis process system using STPA, systematically and comprehensively identifying risk factors while fully accounting for the system’s emergent characteristics, thus providing a reliable basis for subsequent quantitative analysis. 2) Introduced the MMDE algorithm to determine the threshold for the DEMATEL-ISM model, overcoming the limitations of traditional methods, enhancing the scientific rigor and efficiency of threshold determination, and ensuring the accuracy of risk analysis results. 3) Utilized the improved DEMATEL-ISM model to identify critical risk factors, construct risk propagation paths, uncover causal scenarios, and propose targeted risk prevention recommendations. Crucially, the identified critical factors (temperature sensor and control algorithm errors) and their propagation paths were corroborated by historical incident data, moving beyond logical consistency to demonstrate empirical validity. This offers scientific and empirically grounded guidance for the safe design, operation, and maintenance of ammonia synthesis process systems. Despite these contributions, this research primarily establishes a static causal network of risks. A recognized limitation is its inability to model the dynamic evolution of risks over time. Building upon the foundational causal structure identified here, future research will focus on incorporating a dynamic risk analysis perspective. Promising directions include: Integrating Dynamic Bayesian Networks (DBN): The static causal relationships and hierarchy identified by ISM can serve as the initial network structure for a DBN. By incorporating time-series data on equipment degradation (e.g., sensor drift) and process parameters (e.g., temperature fluctuations), the DBN can model the time-dependent failure probabilities and update risk levels in real-time. Developing a Dynamic Simulation: The risk transmission paths revealed in this study (e.g., Fig. 8) can be formalized into a system dynamics or agent-based model. This model could simulate how a small perturbation, like a sensor fault, dynamically propagates through the system under different control strategies and time delays, allowing for the evaluation of cumulative effects and the optimization of intervention points. Such dynamic extensions would transform the current static risk picture into a predictive and adaptive risk management tool, further enhancing the safety and resilience of chemical process systems.