A method and system for human factor analysis of aircraft runway excursion events

Abstract

The application discloses an aircraft runway excursion event human factor analysis method and system, the method comprises the following steps: combining the fault tree analysis method and the improved HFACS model, constructing a human factor investigation decision support model HFIDSM, identifying and acquiring the explicit factors and the implicit factors leading to the aircraft runway excursion event; mining the cause factor of the aircraft runway excursion event; constructing a human factor cause Bayesian network model of the aircraft runway excursion event, constructing a key cause chain leading to the aircraft runway excursion event; analyzing the key cause chain, and optimizing the operation procedure of the aircraft according to the analysis result. The application can accurately identify and quantitatively analyze the key cause chain behind the event by deeply analyzing the complex causes of the runway excursion event, and can enhance the understanding of the interaction relationship among the multi-dimensional cause factors such as human error, environmental factors and equipment failure, and promote the systematization and intelligentization of the investigation process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of civil aviation safety management and accident investigation, and particularly relates to a human factor analysis method and system for an aircraft runway excursion event. BACKGROUND

[0002] The aircraft runway excursion event is a major accident type that seriously threatens flight safety, and there are various analysis methods for the aircraft runway excursion event in the prior art, such as: a comprehensive study on the runway excursion event of a special airport, risk coupling assessment and prediction and early warning, which can reduce the probability of the runway excursion event and mitigate the impact of the accident consequences to a certain extent; there is also a method for real-time prediction of landing runway excursion risk, which can reduce the probability of runway excursion during landing to a certain extent. However, the above accident investigation methods focus on technical failures and operational errors, ignore the deep analysis of human factors, and are difficult to systematically reveal the complex causal relationship behind the accident; and lack of systematicness and precision, and are difficult to effectively support accident prevention and decision improvement, especially in the analysis of human errors and organizational factors intertwined with multiple factors. SUMMARY

[0003] The present application provides a human factor analysis method and system for an aircraft runway excursion event to solve the technical problems of ignoring the hierarchical analysis of human factors and the low accuracy and efficiency of event investigation in the prior art aircraft runway excursion event investigation and analysis method.

[0004] In view of the above technical problems, the present application provides a human factor analysis method for an aircraft runway excursion event, comprising:

[0005] Combining the fault tree analysis method and the improved HFACS model, a human factor investigation decision support model HFIDSM is constructed to identify and obtain the explicit factors and implicit factors leading to the aircraft runway excursion event;

[0006] The human factor investigation decision support model HFIDSM is used to mine the cause factors of the aircraft runway excursion event;

[0007] A human factor cause Bayesian network model for the aircraft runway excursion event is constructed to quantify the contribution of the cause factors to the event occurrence and construct a key cause chain leading to the aircraft runway excursion event;

[0008] The key cause chain is analyzed, and the operation procedures of the aircraft are optimized according to the analysis results.

[0009] The present application also provides a human factor analysis system for an aircraft runway excursion event, comprising:

[0010] A constructing module is used for combining the fault tree analysis method and the improved HFACS model to construct a human factor investigation decision support model HFIDSM, and identify and acquire explicit factors and implicit factors causing the aircraft runway excursion event;

[0011] A mining causation factor module is used for mining the causation factors of the aircraft runway excursion event through the human factor investigation decision support model HFIDSM;

[0012] A key causation chain constructing module is used for constructing the human factor causation Bayesian network model of the aircraft runway excursion event, quantifying the contribution degree of the causation factors to the event occurrence, and constructing the key causation chain causing the aircraft runway excursion event.

[0013] An optimization module is used for analyzing the key causation chain, and optimizing the operation procedure of the aircraft according to the analysis result.

[0014] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0015] In the present application, the human factor investigation decision support model HFIDSM realizes comprehensive analysis of the causation chain of the aircraft runway excursion unsafe event by combining the fault tree and the improved HFACS model, covers each link from the direct operation mistake to the deep management defect, and the airline company and the regulatory agency can acquire the scientific basis required for formulating the targeted improvement measures according to the model, thereby effectively improving the aviation safety.

[0016] In the present application, the HFIDSM model not only deeply mines the causation of the runway excursion event, but also comprehensively reveals the complex relationship between the explicit and implicit factors, provides a solid theoretical basis and decision support for the accident prevention, investigation and subsequent improvement measures. In addition, the model also introduces the Bayesian network BN, which has strong expansibility and flexibility, can efficiently identify the key causation chain in the accident, and reveal the dynamic interaction process of the complex accident cause; not only improves the comprehensiveness and depth of the accident investigation, but also ensures the accuracy and efficiency of the investigation result, provides a strong scientific basis for the formulation of the aviation safety management and prevention measures. The present application provides a new systematic and scientific tool for the investigation and analysis of the aircraft runway excursion event, has important application value and popularization prospect. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 Fig. 1 is a flowchart of the human factor analysis method of the aircraft runway excursion event in an embodiment of the present application;

[0018] Figure 2 Fig. 2 is a structural diagram of the improved HFACS model in an embodiment of the present application;

[0019] Figure 3 An analysis diagram of explicit factors in human factors of an aircraft runway excursion event in an embodiment of the present application;

[0020] Figure 4 An analysis diagram of implicit factors in human factors of an aircraft runway excursion event in an embodiment of the present application;

[0021] Figure 5 A structure diagram of a Bayesian network model in an embodiment of the present application;

[0022] Figure 6 A structure diagram of a human factor analysis system of an aircraft runway excursion event in an embodiment of the present application;

[0023] Figure 7 A causal reasoning diagram of a runway excursion event Bayesian model in an embodiment of the present application;

[0024] Figure 8 A reverse diagnosis reasoning diagram of a runway excursion event in an embodiment of the present application;

[0025] Figure 9 A Bayesian network sensitivity analysis result diagram in an embodiment of the present application;

[0026] Figure 10 A diagram of the most critical causal chain in human factors of a runway excursion in an embodiment of the present application;

[0027] Figure 11 A diagram of effectiveness verification of a human factor analysis model of a Bayesian network in an embodiment of the present application. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0029] In an embodiment, as shown in FIG. 1, the present application provides a human factor analysis method of an aircraft runway excursion event, including the following steps: Figure 1

[0030] ​S10, in combination with the fault tree analysis method and the improved HFACS model, a human factor investigation decision support model HFIDSM is constructed to identify and obtain the explicit factors and implicit factors leading to the aircraft runway excursion event; Understandably, the fault tree analysis method is a tool for system safety analysis, which uses logical gates (such as AND gate, OR gate) to refine the failure causes of the system layer by layer, forming an inverted tree diagram, and using the fault tree analysis method to structurally decompose the aircraft runway excursion event to identify each explicit factor (direct cause) leading to the event. The improved HFACS model uses its improved HFACS framework to analyze the implicit factors (root causes) that trigger the explicit factors.

[0031] The improved HFACS model is an improvement based on the existing HFACS model. Specifically, each sub-index of each level is modified according to the aircraft runway excursion event, and there is no longer a complex connection between levels. The accident causation chain tends to be simple.

[0032] The improved HFACS model divides human factors into multiple levels, which are unsafe behavior, prerequisite for unsafe behavior, unsafe supervision, and poor organizational management. Among them, unsafe behavior is usually an explicit factor, and prerequisite for safe behavior, unsafe supervision, and poor organizational management are usually implicit factors. Implicit factors are often the deep-seated causes of unsafe behavior. Through the analysis of implicit factors, the improved HFACS model can systematically reveal the problems between the management level and the operation level, and provide deeper insights for accident prevention.

[0033] In an embodiment, as shown in Figure 2 In order to improve the phenomenon that the content of each level index in the existing HFACS model is disconnected from the factual information and needs to be optimized, the HFACS model is improved from four aspects of poor organizational management, unsafe supervision, prerequisite for unsafe behavior, and unsafe behavior, in order to analyze the deep-seated causes of human factors and dig the causation chain in unsafe events, thereby avoiding similar events from happening again. Specifically:

[0034] Poor organizational management includes improper leadership decision-making, improper resource management, and improper policy-making. Among them, improper leadership decision-making includes weak risk classification control awareness, untimely risk hidden danger checking, inaccurate and imperfect decision-making, etc.; improper resource management includes incomplete organizational structure, imperfect rules and regulations, and insufficient human resource investment; and improper policy-making includes imperfect standard operating procedures and insufficient safety risk control.

[0035] Unsafe supervision includes improper shift management, insufficient supervision, untimely risk prompt, and inadequate safety training. Among them, improper shift management includes unreasonable team resource allocation, weak load management capability, fatigue / overwork, etc.; insufficient supervision includes lack of real-time supervision, failure to perform supervision duties, lack of feedback and reward mechanism, and safety inspection in name only; untimely risk prompt includes untimely control intelligence suggestion, weak risk control capability, etc.; inadequate safety training includes untimely policy publicity, and incomplete skill training, etc.

[0036] The prerequisites for unsafe behavior include poor personnel state, complex operating environment, and mechanical / equipment failure. Among them, the poor personnel state includes personnel qualifications, physical and mental health, business ability, and work style. Personnel qualifications include expired license and expired physical examination certificate; physical and mental health mainly includes physiological abnormalities and psychological abnormalities; business ability includes poor theoretical / practical ability, poor coordination ability, and poor situational awareness; work style includes poor initiative and poor sense of responsibility. Complex operating environment includes marginal weather and complex terrain, marginal weather includes low visibility / night operation, wind shear / inverted wind, thunderstorm weather; complex terrain includes highland / high-highland airport and special airport; mechanical / equipment failure includes hardware design / manufacturing defects, hardware maintenance not according to procedures, software system early warning function is not perfect, and personnel misoperation leads to equipment failure.

[0037] According to the definition of situational awareness, unsafe behavior can be divided into improper information preprocessing, improper problem solving / decision making, and improper action execution; to facilitate investigators to find the problems of unsafe behavior from different aspects based on factual information, so as to make targeted improvement.

[0038] In an embodiment, the step S10 further comprises the following steps:

[0039] S101, using the fault tree analysis method, taking the runway excursion event as the top event for reverse deduction, analyzing layer by layer downward, identifying and obtaining all explicit factors of each level of the aircraft runway excursion event; it can be understood that through layer-by-layer analysis, the direct causes (explicit factors) leading to the runway excursion event can be clearly identified and obtained, such as Figure 3 As shown in the figure, the fault tree analysis reveals the logical relationship between each explicit factor, which helps to understand the specific path and cause of the event.

[0040] S102, using the improved HFACS model to analyze and obtain the corresponding implicit factors of all the explicit factors; it can be understood that the combination of explicit factors and implicit factors can form a complete causal relationship chain, which helps to comprehensively understand the deep reasons for the event. Through the analysis of implicit factors, potential systemic defects and weak links in safety management can be revealed, and targeted suggestions for improvement measures can be provided. For example Figure 4 As shown in the analysis results of the implicit factors, the specific content of these implicit factors can be clearly displayed, and the logical relationship between them can also be clearly expressed.

[0041] S103, combining the explicit factors and the implicit factors to construct the human factor investigation decision support model HFIDSM. It can be understood that the human factor investigation decision support model HFIDSM combines the explicit factors and the implicit factors to form a comprehensive evaluation method, which can systematically and scientifically analyze the human factors of the aircraft runway excursion event. Specifically, the construction steps of the human factor investigation decision support model HFIDSM include: receiving and preprocessing the collected data, identifying and analyzing the explicit factors, identifying and analyzing the implicit factors, and constructing the HFIDSM model by combining the correlation between the explicit factors and the implicit factors. The specific construction procedure of the HFIDSM model can be set according to the needs.

[0042] S20, using the human factor investigation decision support model HFIDSM to mine the causation factors of the aircraft runway excursion event; it can be understood that by deeply analyzing the human factors in the unsafe event from the macro and micro perspectives, as well as the logical relationship between the causes, the causation factors are finally obtained.

[0043] In an embodiment, the step S20 includes the following steps:

[0044] S201, collecting various types of related data of the aircraft runway excursion event based on the human factor investigation decision support model HFIDSM, summarizing the various types of related data into associated data, preprocessing the associated data, eliminating invalid data and irrelevant data, and obtaining the associated data; it can be understood that the associated data of the runway excursion event includes accident investigation reports, pilot operation records, flight data records, cockpit voice records, weather data, and airport runway conditions.

[0045] S202, using an accident analysis method, determining the top-level event as the aircraft off the runway event, starting from the top-level event, layer by layer, decomposing the aircraft off the runway event, analyzing the direct causes of the aircraft off the runway event, and constructing an accident tree; Understandably, this accident analysis method can show in detail how unsafe behavior gradually leads to the occurrence of the event through a series of logical steps, and reveal the position and role of explicit behavior (such as pilot's decision-making error, controller's improper instruction, etc.) in the cause chain.

[0046] In an embodiment, the step S202 of constructing the accident tree further comprises the following steps:

[0047] Starting from the top-level event, layer by layer, decomposing the aircraft off the runway event;

[0048] Identify and obtain the direct causes of the top-level event, i.e. intermediate events;

[0049] Continue to decompose each intermediate event until the basic event is found;

[0050] Connect the top-level event, intermediate event and basic event using logical gates to form a graphical accident tree.

[0051] Understandably, the accident tree starts from the top-level event (i.e. the aircraft off the runway event) and is structured and decomposed layer by layer, gradually refining the complex event into manageable components. This method helps to clarify the logical relationship of the event and avoid missing important influencing factors. Through the construction of the accident tree, the direct causes (i.e. intermediate events) leading to the top-level event can be systematically identified. These intermediate events provide clear direction and clues for further analysis. Continue to decompose the intermediate events until the basic events (i.e. indivisible events or failures) are found. The basic events are the smallest units of the accident tree, which directly reflect the specific causes of the event. Using logical gates to connect the top-level event, intermediate event and basic event forms a graphical accident tree. This graphical expression intuitively shows the causal relationship and logical relationship between events, making it easy to understand and analyze. Accident tree analysis not only considers the role of a single factor, but also considers the interaction and influence between multiple factors, which helps to reveal the cause factors of the event comprehensively and systematically.

[0052] S203, using the accident tree for analysis, according to the results of the accident tree analysis, taking the direct causes of the aircraft off the runway event as the starting point for improving the HFACS model, and using the multi-level classification framework of the improved HFACS model to analyze and obtain the implicit factors behind the unsafe behavior; Understandably, through the analysis of these implicit factors, potential systemic defects can be revealed and weak links in safety management can be identified.

[0053] The multi-level classification framework of the improved HFACS model includes organizational influences, unsafe supervision, preconditions of unsafe acts, and unsafe acts.

[0054] S204, combine the explicit factors of the fault tree analysis and the implicit factors of the improved HFACS model analysis to obtain the causation factors of the aircraft runway excursion event. The causation factors can provide a basis for the prevention and management improvement of similar events in the future.

[0055] In an embodiment, the explicit factors include pilot operation errors, mechanical equipment failures, operating environments, human management omissions, improper scheduling management, poor personnel states, and operating environments.

[0056] The implicit factors include the psychological state of the pilot, flight task allocation, training management, execution of operation procedures, organizational culture, and resource allocation.

[0057] S30, construct a human factor causation Bayesian network model of the aircraft runway excursion event, quantify the contribution of the causation factors to the occurrence of the event, and construct a key causation chain leading to the aircraft runway excursion event. Understandably, based on the explicit factors and the implicit factors, a systematic and scientific model is constructed through the Bayesian network for analyzing the human factor causation chain in the aircraft runway excursion event. The model can help investigators more accurately identify and analyze the key factors in the event and provide data support for preventing similar events.

[0058] In an embodiment, the step S30 includes the following steps:

[0059] S301, define variable nodes based on the explicit factors and the implicit factors, represent the causal relationship between the variable nodes through directed edges, and establish a human factor causation Bayesian network structure according to the variable nodes, the directed edges, and the conditional probability table, as shown in Figure 5 The human factor causation Bayesian network model is constructed, and the aircraft runway excursion event.

[0060] In an embodiment, the step S301 further includes the following steps:

[0061] S3011, identify all relevant human factors of the aircraft runway excursion event, including the explicit factors and the implicit factors. Understandably, it is ensured that any human factor that may affect the aircraft runway excursion event is not missed, including the explicit factors and the implicit factors.

[0062] S3012, define variable nodes based on human factors, represent the causal relationship between variable nodes through directed edges, connect variable nodes through directed edges, and obtain the logical order and dependency relationship reflected by the directed edges; Understandably, by defining variable nodes and directed edges, the causal relationship and logical order between each human factor are clearly represented. The directed edges not only represent the connection between factors, but also reflect the logical order and dependency relationship of event occurrence.

[0063] S3013, equip each variable node and its parent node with a conditional probability table, quantify the specific value of the correlation strength between variable nodes, and set the conditional probability value, and construct the conditional probability table according to the conditional probability value; Understandably, each variable node represents a human factor, such as pilot operation error, fatigue work, etc.; The directed edge represents the influence of one variable node on another variable node, such as fatigue work leading to pilot operation error.

[0064] S3014, establish the structure of the human factor causation Bayesian network model according to the variable nodes, directed edges and conditional probability table, verify the human factor causation Bayesian network model using actual cases or simulation data, and adjust the model structure and parameters according to the verification result. Understandably, by quantifying the correlation strength between each variable node through the conditional probability table, the model analysis is more scientific and accurate. When there is not enough data, expert knowledge is used to directly assign probability values to improve the practicality of the model.

[0065] S302, use Bayesian inference method to infer the conditional probability table, identify the key factors in the causation factors, determine the key factors as key nodes, and evaluate the weight of the key nodes; Understandably, the conditional probability table lists all possible state combinations and their corresponding probability values for each non-root node.

[0066] In an embodiment, the step S302 further comprises the following steps:

[0067] S3021, after the human factor causation Bayesian network model is constructed, the probability change of other unobserved factors is inferred in reverse according to the human factors using the Bayesian inference method; Understandably, through Bayesian inference, the interaction between each potential factor and its comprehensive influence on event occurrence can be considered comprehensively and systematically; Reverse inference can help identify which factors have the greatest impact on event occurrence under unobserved conditions, providing support for subsequent decision-making.

[0068] S3022, analyze the influence of each node in the causation chain through the results of Bayesian inference, and determine the weight of the key nodes; Understandably, determining the weight of the key nodes helps to clarify the priority in prevention and management, and to solve those factors that have the greatest impact on the event first; Through weight analysis, targeted improvement measures can be proposed to maximize the probability of event occurrence.

[0069] S3023, verify the inference result using actual accident data. If it is found that the inference result does not match the actual situation, recheck the data input, Bayesian network model structure and conditional probability table, and make corresponding adjustments and optimizations. Understandably, through the verification of the actual data, the accuracy and reliability of the inference result can be ensured; by continuously adjusting and optimizing the model, the adaptability and prediction ability of the model can be improved, so that it is more in line with the actual situation.

[0070] S303, apply the human factor causation Bayesian network model to analyze the influence of different combinations of causation factors on the aircraft runway excursion event, perform diagnostic reasoning, and identify the key causation chain of the aircraft runway excursion event according to the posterior probability. The aircraft runway excursion event provides improvement direction for the management part.

[0071] In an embodiment, the step S303 further comprises the following steps:

[0072] S3031, apply forward reasoning, starting from the known node, use the topology structure and conditional probability table of the Bayesian network to predict the probability of occurrence of the aircraft runway excursion event; Understandably, by using the constructed Bayesian network model and the known node state (such as equipment state, personnel qualification, operating environment, etc.), the probability of occurrence of the aircraft runway excursion event is calculated through forward reasoning (predictive reasoning). This step can help the safety management department to predict the risk according to the current state and condition before the event actually occurs, and to take preventive measures in advance.

[0073] S3032, apply backward reasoning, set the aircraft runway excursion event as a known state, and inversely deduce the state probability changes of other nodes to identify the factors that have the greatest impact on the aircraft runway excursion event, i.e. the key causation factors; Understandably, by setting the aircraft runway excursion event as a known state (i.e. setting its occurrence probability to 100%), the factors that have the greatest impact on the aircraft runway excursion event, i.e. the key causation factors, are calculated and identified through reverse reasoning (diagnostic reasoning). This step can help investigators quickly locate the cause of the accident and find out the links that need to be focused on.

[0074] S3033, combine the results of forward and backward reasoning, and use sensitivity analysis to identify key parameters to obtain the key causation chain leading to the aircraft runway excursion event; Understandably, by comprehensively analyzing the results of forward and backward reasoning, the causation chain of the aircraft runway excursion event can be more comprehensively understood. Further, by using sensitivity analysis to evaluate the influence of different parameter changes on the final event occurrence probability, the key parameters and causation chain that have the greatest impact on the result are identified. This helps to determine the key direction of prevention and improvement measures.

[0075] S3034, display the key causal chain in the form of a chart or list, show the causal relationship and influencing factors in each chain, and provide intuitive analysis results for the aviation safety management department. Understandably, the key causal chain can be displayed in the form of a chart or list, which can clearly show the causal relationship and influencing factors in each chain. This not only facilitates the understanding of complex accident causes by safety management departments and investigators, but also provides intuitive reference for the development of prevention measures and improvement programs.

[0076] S40, analyze the key causal chain and optimize the operation procedures of the aircraft according to the analysis results. Understandably, the model analysis results can also be used to optimize existing operation procedures, training plans and risk warning systems, and improve the safety of overall aviation operation. By constructing and analyzing the key causal chain, the deep-seated causes of the aircraft runway excursion event can be systematically revealed, including organizational management, supervision, personnel state, operation behavior and other potential defects. These defects are often difficult to find through individual analysis, and the construction of the key causal chain makes these potential problems exposed.

[0077] Based on the analysis results of the key causal chain, the operation procedures of the aircraft can be optimized. For example, if it is found that imperfect regulations are an important cause of the accident, the relevant regulations can be optimized to ensure their integrity and effectiveness. If it is found that improper scheduling management affects the fatigue level of pilots, the scheduling system can be adjusted to ensure that pilots get enough rest.

[0078] The analysis results can also be used to improve the training plan of the pilots. By identifying the inappropriate behavior of the pilots in a specific situation, the relevant training can be targeted to improve the skill level and ability of the pilots to deal with complex situations. For example, if it is found that the pilots have poor situational awareness in complex environments, situational awareness training can be strengthened to improve the response ability of the pilots.

[0079] By optimizing the operation procedures, improving the training plan and the risk warning system, the safety of the overall aviation operation can be significantly improved. This not only helps to reduce the occurrence of aircraft runway excursion and other accidents, but also improves the operational efficiency and market competitiveness of the airline.

[0080] It should be understood that the size of the serial number of each step in the above embodiments does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0081] In an embodiment, an aircraft runway excursion event human factor analysis system is provided, which corresponds to the aircraft runway excursion event human factor analysis method in the above embodiment. As shown in FIG. 6, the aircraft runway excursion event human factor analysis system comprises:

[0082] The construction module 100 is configured to construct a human factor investigation decision support model HFIDSM by combining the fault tree analysis method and the improved HFACS model, identify and obtain explicit factors and implicit factors leading to the aircraft runway excursion event;

[0083] The cause factor mining module 200 is configured to mine the cause factors of the aircraft runway excursion event through the human factor investigation decision support model HFIDSM;

[0084] The key cause chain construction module 300 is configured to construct a human factor cause Bayesian network model of the aircraft runway excursion event, quantify the contribution of the cause factors to the occurrence of the event, and construct a key cause chain leading to the aircraft runway excursion event;

[0085] The optimization module 400 is configured to analyze the key cause chain and optimize the operation procedures of the aircraft according to the analysis result.

[0086] In an embodiment, the runway excursion events with a large proportion of serious incidents in the past 10 years are verified, and the cause chain of the runway excursion event is obtained as shown in Table 1:

[0087] Table 1 Human factor cause chain of runway excursion event

[0088]

[0089]

[0090] Based on the cause chain analyzed by the HFIDSM, a Bayesian network model of the runway excursion event is constructed, which contains four continuous causal paths affecting the probability of the runway excursion event, i.e. poor organizational management, unsafe supervision, prerequisite conditions for unsafe behavior, and unsafe behavior, and each causal path also contains multiple human factor sub-nodes. The runway excursion accident investigation reports in recent years are analyzed to extract the keywords related to the cause factors, calculate the frequency and convert it into the prior probability of the factor as shown in Table 2:

[0091] Table 2 Prior probability analysis of runway excursion human factors

[0092]

[0093] Direct assignment is an effective method for defining the conditional probabilities of non-root nodes in Bayesian networks, especially when data is scarce or domain knowledge is complex. This method is achieved through four core steps: first, the topological structure of the Bayesian network, i.e., the causal relationships between nodes, is determined, and the parent nodes of each non-root node are identified; second, all possible state combinations of the non-root nodes and their parent nodes are listed, ensuring that all possible cases are covered; third, according to expert knowledge, each non-root node is directly assigned a probability value under different state combinations of its parent nodes, and this process relies on the rich experience and professional knowledge of experts; finally, a conditional probability table (CPT) is constructed, and the probability values obtained based on expert knowledge are filled into the table for subsequent analysis. This direct assignment method not only improves the accuracy of the model in the case of data scarcity, but also enhances the application ability of the model in complex fields, providing reliable data support for the optimization of aircraft operation procedures.

[0094] In an embodiment, as shown in Figure 7 , by combining the prior probability obtained based on the statistical frequency of accident investigation reports and the conditional probability obtained based on expert experience, these probability values are input into each node of the constructed Bayesian network model. This process adopts a top-down reasoning approach, i.e., causal reasoning, through the complex structure of the Bayesian network, it can systematically analyze and quantify the causal relationships between human factors, and thus obtain the causal reasoning results of the human factors of runway excursion events. This method provides a scientific basis and comprehensive evaluation for the improvement of aviation safety management measures.

[0095] As shown in Figure 8 , the inference mechanism of the Bayesian network relies on the probability relationships defined in the network, and the posterior probability of the observed variables is calculated through the propagation and update of information. This inference process can be divided into forward reasoning (or predictive reasoning) and backward reasoning (also known as diagnostic reasoning). Forward reasoning focuses on inferring potential results from known causes, while backward reasoning reverses this process, starting from known results to infer possible causes and their impact.

[0096] In Bayesian networks, diagnostic reasoning refers to the process of reverse analysis and updating the posterior probability distribution of all potential influencing factors (i.e., parent nodes) by setting the probability of a certain node (usually the result node) to a specific value (e.g., "runway excursion" set to 100%). This method helps to assess the specific impact of each risk factor on the occurrence of a specific event, revealing the key factors behind the event. By using Bayesian network software such as GeNIe, researchers can easily perform this bottom-up reasoning process. By setting the probability value of the target node to a specific value (e.g., 100%) and triggering the software's probability refresh function, the software will automatically calculate and display the posterior probability values of all related parent nodes, providing intuitive data support for understanding the causal relationships of complex events. This diagnostic reasoning method has wide application value in the fields of aviation safety analysis, risk assessment, and decision-making.

[0097] The higher the posterior probability value of a Bayesian network node, the greater the risk contribution of that factor to the runway excursion event. As can be seen, the risk contribution of the second layer of human factors to the runway excursion event is in descending order: unsafe behavior, prerequisite conditions for unsafe behavior, poor organizational management, and unsafe supervision. By comparing the diagnostic reasoning results with the prior probability and conditional probability results of each node, it can be seen that the probabilities of each root node have not changed, and the posterior probabilities of most non-root nodes have changed to varying degrees, as shown in Table 3:

[0098] Table 3: Probability change results of reverse reasoning for node "runway excursion"

[0099]

[0100] As shown in Figure 9 Sensitivity analysis is a key method for evaluating the sensitivity of a Bayesian network model to changes in input parameters, aiming to explore how small changes in each parameter (especially probability distribution) in the network significantly affect the model output (posterior probability). Through sensitivity analysis, model designers can clearly identify which parameters or network structures play a core role in the model, and their changes will directly cause significant fluctuations in the model output. This process not only deepens the understanding of the internal mechanisms of the model, but also provides important clues for subsequent model optimization, ensuring that the model has sufficient stability and accuracy in practical applications.

[0101] In Bayesian networks, sensitivity analysis is a crucial technique for validating and optimizing probabilistic parameters. By making small adjustments to the model's numerical parameters (including prior and conditional probabilities) and observing how these adjustments affect changes in the output parameters (posterior probabilities), a quantitative assessment of the model's sensitivity can be achieved. Parameters that have the most significant impact on the output, i.e., high-sensitivity parameters, are considered key factors in the model's inference process. Identifying and focusing on these parameters helps model designers estimate their values ​​more accurately, thereby significantly improving the model's accuracy and reliability. Furthermore, sensitivity analysis helps reveal potential weaknesses in the model, providing directional guidance for improving and optimizing the model structure.

[0102] like Figure 10 As shown, assume the target node's "off-track" state is Yes, i.e., p("off-track" state = Yes) = 100%. Then, the posterior probability of network nodes is updated through reverse reasoning. At this point, "unsafe behavior" is the parent node with the highest posterior probability of the "off-track" state, so the key causal chain reasoning is {"unsafe behavior" → "off-track"}. Then, the search continues from the node "unsafe behavior" to find the parent node with the highest posterior probability of the state, until the root node of the network. Ultimately, the most critical causal chain among the human factors causing the "off-track" state can be found to be {"improper action execution / improper information preprocessing" → "unsafe behavior" → "off-track"}, which is crucial for the investigation and prevention of unsafe events. Similarly, three other key human-caused causal chains for deviation from the target path can be identified: {"Poor theoretical and practical skills / poor coordination" → "Work skills" → "Poor personnel condition" → "Prerequisites for unsafe behavior" → "Deviation from the target path"}, {"Insufficient safety risk management" → "Inappropriate policy formulation" → "Poor organizational management" → "Deviation from the target path"}, and {"Incomplete skills training" → "Inadequate safety training" → "Unsafe supervision" → "Deviation from the target path"}. In the diagram, red nodes contain parameters important to the posterior probability distribution of the target node (deviation from the target path), while gray nodes do not contain any parameters used to calculate the target posterior probability distribution. The darker the color, the greater its importance in the network; the thicker the arrow, the larger the weight value.

[0103] In a specific embodiment, the effectiveness of the human factor causal Bayesian network model is verified. On July 7, 2018, Shenzhen Airlines ZH9127 flight deviated from the runway during landing in Hohhot, causing the left outer main wheel and left front wheel of the aircraft to burst, but no one was injured. The event was determined to be a human responsibility reason for serious traffic accidents. After the event, the North Bureau responded quickly, established an investigation team and deployed emergency response and investigation work, and launched a comprehensive investigation through on-site investigation, data decoding, evidence collection and other means. Finally, it was determined that the operation of the pilot's flight skills was insufficient, the crew's resource management was lacking, and the coordination problem was the direct cause, while the crew's safety awareness was weak, the lack of emergency experience and the complex thunderstorm weather also exacerbated the severity of the event.

[0104] To further analyze the human factors of this runway deviation event, we compare the investigation results with the Bayesian network model of runway deviation human factors. As shown in Figure 11 , in the model, we set the states of multiple key nodes such as "thunderstorm weather", "coordination", and "improper information preprocessing" to "Yes" according to the event situation to simulate the actual environment and human factors. Then, through top-down causal reasoning, we refresh the network node state and calculate the probability of runway deviation event occurrence as 78%. This result is highly consistent with the actual situation, not only revealing the complex human-machine interaction process of the event, but also verifying the effectiveness and practicality of the human factor analysis model based on Bayesian network in aviation safety event investigation.

[0105] The successful analysis of this runway deviation event not only provides a scientific basis for event investigation, but also further verifies the potential of Bayesian network model in identifying and managing aviation human factor risks. Through the application of the model, we can more accurately assess the impact of various human factors on flight safety and provide strong support for the development of targeted preventive measures. In the future, with the continuous emphasis on safety management in the aviation industry and the continuous progress of technical means, the human factor analysis model based on Bayesian network is expected to play a more important role in the field of aviation safety and contribute to improving the global level of aviation safety.

[0106] The above is only an embodiment of the aircraft runway deviation event human factor analysis method and system of the present application, and does not limit the present application. Any modification, equivalent replacement and improvement within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method of human factors analysis of an aircraft runway excursion event, characterized in that, The method comprises the following steps: S10, combining the fault tree analysis method and the improved HFACS model, constructing a human factor investigation decision support model HFIDSM, and identifying and obtaining the explicit factors and implicit factors leading to the runway excursion event of the aircraft; the step S10 further comprises the following steps: S101, using the fault tree analysis method, taking the runway excursion event as the top event for reverse deduction, analyzing layer by layer downward, identifying and obtaining all explicit factors of each level of the runway excursion event of the aircraft; S102, using the improved HFACS model to analyze and obtain the implicit factors corresponding to all the explicit factors; S103, combining the explicit factors and the implicit factors to construct the human factor investigation decision support model HFIDSM; S20, mining the cause factors of the runway excursion event of the aircraft through the human factor investigation decision support model HFIDSM; the step S20 comprises the following steps: S201, collecting various types of relevant data of the runway excursion event of the aircraft based on the human factor investigation decision support model HFIDSM, summarizing the various types of relevant data into associated data, pre-processing the associated data, eliminating invalid data and irrelevant data, and obtaining the associated data; S202, using the accident analysis method, determining the top event as the runway excursion event of the aircraft, starting from the top event, and performing structural decomposition on the runway excursion event of the aircraft layer by layer downward to analyze the direct causes leading to the runway excursion event of the aircraft, so as to construct an accident tree; S203, using the accident tree for analysis, according to the result of the accident tree analysis, taking the direct causes leading to the runway excursion event of the aircraft as the starting point of the analysis of the improved HFACS model, and applying the multi-level classification framework of the improved HFACS model to analyze and obtain the implicit factors behind the unsafe behaviors; The multi-level classification framework of the improved HFACS model comprises organizational influence, unsafe supervision, prerequisite of unsafe behavior and unsafe behavior; S204, combining the explicit factors of the accident tree analysis and the implicit factors of the analysis of the improved HFACS model, obtaining the cause factors of the runway excursion event of the aircraft; S30, constructing a human factor cause Bayesian network model of the runway excursion event of the aircraft, quantifying the contribution degree of the cause factors to the event occurrence, and constructing a key cause chain leading to the runway excursion event of the aircraft; the step S30 further comprises the following steps: S301, based on the explicit factors and the implicit factors, defining variable nodes, representing the causal relationship between the variable nodes through directed edges, establishing a human factor cause Bayesian network structure according to the variable nodes, directed edges and conditional probability table, and constructing a human factor cause Bayesian network model of the runway excursion event of the aircraft; S302, using the Bayesian inference method to infer the conditional probability table, identifying the key factors in the cause factors, determining the key factors as key nodes, and evaluating the weights of the key nodes; S303, the application factor cause Bayesian network model is used to analyze the influence of different combinations of cause factors on the aircraft runway excursion event, to carry out diagnostic reasoning, and to identify the key cause chain of the aircraft runway excursion event according to the posterior probability, so as to provide improvement direction for the management part; S40, analyzing the key cause chain and optimizing the operation procedure of the aircraft according to the analysis result.

2. The method of claim 1, wherein, The correlation data of the runway excursion event includes accident investigation reports, pilot operation records, flight data records, cockpit voice records, meteorological data and airport runway conditions.

3. The method of claim 1, wherein, The explicit factors include pilot operation errors, mechanical equipment failures, operation environments, human management omissions, improper scheduling management, poor personnel state and operation environment; The implicit factors include pilot psychological state, flight task allocation, training management, operation procedure execution, organizational culture and resource allocation; The step S202 of constructing the fault tree further includes the following steps: Starting from the top event, the aircraft runway excursion event is structurally decomposed layer by layer downward; The direct causes leading to the top event, i.e. intermediate events, are identified and obtained; Each intermediate event is further decomposed until the basic event is found; The top event, intermediate event and basic event are connected by using a logic gate to form a graphical fault tree.

4. The method of claim 1, wherein, The step S301 further includes the following steps: S3011, identifying all relevant human factors of the aircraft runway excursion event, the human factors including explicit factors and implicit factors; S3012, defining variable nodes based on the human factors, representing the causal relationship between the variable nodes by using directed edges, connecting the variable nodes by using the directed edges, and obtaining the logical order and dependency relationship of the event occurrence reflected by the directed edges; S3013, providing a conditional probability table for each variable node and its parent node, quantifying the specific numerical value of the correlation strength between the variable nodes, and setting the conditional probability value, and constructing the conditional probability table according to the conditional probability value; S3014, establishing the structure of the human factor cause Bayesian network model according to the variable nodes, directed edges and conditional probability table, verifying the human factor cause Bayesian network model using actual cases or simulation data, and adjusting the model structure and parameters according to the verification result.

5. The method of claim 4, wherein, The step S302 further includes the following steps: S3021, after the human factor cause Bayesian network model is constructed, the probability change of other unobserved factors is inferred in reverse according to the human factors by using the Bayesian inference method; S3022, through the result of Bayesian inference, the influence of each node in the cause chain is analyzed to determine the weight of the key node; S3023, the inference result is verified using actual accident data, if it is found that the inference result does not match the actual situation, the data input, Bayesian network model structure and conditional probability table are rechecked, and corresponding adjustment and optimization are carried out.

6. The method of claim 4, wherein, The step S303 further includes the following steps: S3031, applying forward reasoning, starting from the known node, using the topological structure of the Bayesian network and the conditional probability table, and predicting the probability of occurrence of the aircraft runway excursion event; S3032, apply backward reasoning, set the aircraft runway excursion event as the known state, deduce the state probability changes of other nodes in reverse, and identify the factors that have the greatest impact on the aircraft runway excursion event, i.e. the key causal factors; S3033, combine the results of forward and backward reasoning, and use sensitivity analysis to identify key parameters to obtain the key causal chain leading to the aircraft runway excursion event; S3034, display the key causal chain in the form of a chart or list, show the causal relationship and influencing factors in each chain, and provide intuitive analysis results for aviation safety management departments.

7. An aircraft runway excursion event human factor analysis system implementing the method of claim 1, wherein, Including: A construction module is used to combine the fault tree analysis method and the improved HFACS model to construct a human factor investigation decision support model HFIDSM, and identify and obtain the explicit and implicit factors leading to the aircraft runway excursion event; The mining causal factor module is used to mine the causal factors of the aircraft runway excursion event through the human factor investigation decision support model HFIDSM; The key causal chain construction module is used to construct a human factor causal Bayesian network model of the aircraft runway excursion event, quantify the contribution of the causal factors to the occurrence of the event, and construct the key causal chain leading to the aircraft runway excursion event; The optimization module is used to analyze the key causal chain and optimize the operation procedures of the aircraft according to the analysis results.

Images