A method for quantitatively evaluating human factors affecting the risk of civil aircraft flight test

Abstract

The present invention discloses a method for quantitatively evaluating the human factors affecting the risk of civil aircraft flight test, which relates to the technical field of human risk assessment for civil aircraft flight test, including: adaptively improving the human factor analysis and classification system in combination with flight test data, identifying the human risk factors affecting the safety of civil aircraft flight test; constructing a fault tree model through the identified human risk factors to obtain conditional probability; constructing a Bayesian network model based on the mapping relationship between the fault tree model and the Bayesian network model; adopting an expert survey method, and comprehensively considering the expert fuzzy opinions of consensus differences based on an improved similarity aggregation method, obtaining the root node prior probability through the aggregation of expert opinions and the defuzzification process of expert opinions; calculating the conditional probability and the root node prior probability through a Bayesian network model to obtain key risk factors. The present invention improves the quality and comprehensiveness of human risk assessment for civil aircraft flight test, carries out targeted prevention and control, and reduces the occurrence of accidents.

Description

Technical Field

[0001] The present invention relates to the technical field of human risk assessment for civil aircraft flight test, and in particular to a method for quantitatively assessing human factors affecting the risk of civil aircraft flight test. Background Art

[0002] Civil aircraft flight test is a complex system engineering, and its fundamental purpose is to verify the safety of the aircraft. However, the test flight needs to be carried out beyond the flight envelope, in special environments and under extreme weather conditions, which makes accidents very likely to occur during the test. In order to improve the safety of civil aircraft test flights, the National Aeronautics and Space Administration (NASA) has established a comprehensive test flight risk database based on years of model test flight experience, covering almost all civil aircraft test flight subjects. However, as a comprehensive system engineering, civil aircraft test flight not only includes complex test flight subjects, but its multi-task, high-load and high-risk flight conditions are bound to have an impact on the psychology and physiology of test pilots, and thus affect the safety of the flight process. However, the relevant risk database rarely mentions the impact of human factors on risks that cannot be ignored. Since the second half of the 20th century, due to the rapid development of aviation technology, the focus of aviation safety departments has gradually shifted from the technical stage to the human factor stage, and finally developed to the organizational and safety culture stage. The statistics of aviation accident reports in 2017 show that more than 70% of aviation accidents are affected by human factors. It can be seen that the increasingly complex aviation system has put forward higher requirements on the psychological quality of test pilots, but in the field of civil aircraft test flight, the influence of human factors has not received due attention. So far, there are few studies on human factors in civil aircraft test flight, and they are all aimed at evaluating risk factors in a specific aspect. Although they are highly targeted, they lack the systematic analysis of risk factors. The lack of relevant research may also lead to a slightly weak control of the influence of human factors in the actual civil aircraft test flight process. Therefore, in order to reduce the occurrence of civil aircraft test flight accidents and improve the safety of test flights, it is very important to systematically analyze the influence of human factors in the test flight process.

[0003] At present, the commonly used risk identification methods in the field of human factors include the Swiss cheese model (REASON model), the Human Factors Analysis and Classification System (HFACS model), etc. The HFACS model is a more comprehensive model that improves the REASON model. Its advantage is that it can understand the potential causal factors that lead to accidents. As a comprehensive human error analysis model, the HFACS model is widely used in various fields and can improve the safety and reliability of various fields. Therefore, it is valuable to analyze the relationship between human factors and civil aircraft test flight safety accidents within the framework of the HFACS model and then identify the risk source. However, although HFACS is a powerful tool for identifying and classifying human factors in accident investigations, it has been criticized for its limitations such as lack of quantitative analysis and reasoning under uncertain conditions and failure to consider the complex interdependence between factors.

[0004] Traditional quantitative risk assessment (QRA) calculation methods include Bayesian networks, fault tree analysis, event tree analysis, Monte Carlo analysis, etc. Among them, Bayesian networks (BN) have been considered to be an effective alternative to consider the dependencies between human factors identified by HFACS and analyze the overall impact of these dependencies on the occurrence of accidents. Bayesian networks have a strong ability to handle uncertainty problems. They use conditional probabilities to express the correlation between various information elements and can learn and reason under limited, incomplete, and uncertain information conditions. However, Bayesian network calculations require accurate probability information (including the prior probability of the root node and the conditional probability between nodes). However, due to the lack of historical data on test flight accidents and incomplete knowledge of the probability values ​​of human errors, it is difficult to calculate with accurate probability values. When assigning precise probabilities to ambiguous events, fuzzy set theory (FST) is an effective tool for dealing with cognitive uncertainty in failure probabilities, and can solve the problem that is difficult to calculate due to insufficient data. The fuzzy Bayesian network (FBN) that combines FST and BN is currently widely concerned and used. Therefore, integrating FST and BN in the HFACS framework helps to calculate more objective and accurate results.

[0005] Similarity Aggregation Method (SAM) is a probability calculation method based on FST. It uses the consensus index and the importance of each expert to summarize the fuzzy opinions of experts. Combining it with BN can accurately calculate the fuzzy failure probability of the root node of BN and reduce the deviation caused by subjective factors or individual differences. However, the traditional SAM also has disadvantages. Although it considers the influence of consensus and individual differences at the same time, it does not consider the influence relationship between the two. Guo et al. proposed an improved SAM. The improved SAM not only considers the degree of consensus, but also solves the consensus differences caused by different individual cognitive levels in the process of expert opinion aggregation, making the aggregation results more inclined to the estimates of highly reliable experts, thereby making the calculation results more accurate. Secondly, in order to solve the problem of conditional probability calculation between Bayesian network nodes, Fault Tree Analysis (FTA) can be well combined with Bayesian network for probability calculation due to its special logical correspondence structure and the mapping relationship between it and Bayesian network. Summary of the invention

[0006] In order to solve the problem that it is difficult to conduct quantitative risk probability calculation research due to insufficient test flight accident data, and because the current risk factor analysis level of human factors related to civil aircraft test flight is relatively single, a method for quantitatively evaluating the human factors affecting the risk of civil aircraft test flight is proposed to improve the quality and comprehensiveness of human risk assessment in civil aircraft test flight, effectively find the key links of accidents, carry out targeted prevention and control, and reduce the occurrence of accidents.

[0007] To achieve the above object, the present invention provides a method for quantitatively evaluating human factors affecting the risk of civil aircraft flight test, comprising:

[0008] Adaptively improve the HFACS model of the human factor analysis and classification system based on test flight data to identify human risk factors that affect civil aircraft test flight safety;

[0009] By identifying the human risk factors, a fault tree model is constructed to obtain conditional probability;

[0010] Based on the mapping relationship between the fault tree model and the Bayesian network model, constructing a Bayesian network model;

[0011] The expert survey method is adopted, and the fuzzy opinions of experts with consensus differences are comprehensively considered based on the improved similarity aggregation method. The prior probability of the root node is obtained by aggregating the fuzzy opinions of experts and defuzzifying the experts' opinions.

[0012] The conditional probability and the root node prior probability are calculated and analyzed through the Bayesian network model to obtain key risk factors.

[0013] Preferably, the human risk factors affecting the safety of civil aircraft test flights include: unsafe behavior factors, unsafe supervision factors and organizational influence factors.

[0014] Preferably, the mapping relationship between the fault tree model and the Bayesian network model is:

[0015] The top event is converted into a leaf node, the intermediate event is converted into an intermediate node in the Bayesian network, the bottom event is converted into a root node in the Bayesian network, and the logic gate of the fault tree corresponds to the conditional probability table in the Bayesian network; wherein the logic gate includes an OR gate and an AND gate.

[0016] Preferably, obtaining the root node prior probability includes:

[0017] The relative importance of each expert's opinion is represented by assigning a weight coefficient to the expert, wherein the Delphi method is used to assign the weight coefficient of each expert, and the failure probability of the risk factor is represented based on the trapezoidal fuzzy number, that is, the root node prior probability.

[0018] Preferably, the membership function of the trapezoidal fuzzy number is:

[0019]

[0020] In the formula, x is a fuzzy number, r1, r2, r3, and r4 are comprehensive evaluation fuzzy numbers. The numerical value corresponding to the fuzzy set, F (x) is the membership function of the trapezoidal fuzzy number.

[0021] Preferably, the process of aggregating the expert fuzzy opinions and defuzzifying the expert opinions includes:

[0022] The consistency degree of experts' fuzzy ratings, the weighted consistency of experts' fuzzy evaluations, and the relative consistency of experts' fuzzy evaluations were calculated respectively;

[0023] Based on the consistency of the expert fuzzy rating, the weighted consistency of the expert fuzzy evaluation, and the relative consistency of the expert fuzzy evaluation, the expert consensus coefficient is predicted, and the expert's node status is calculated as The comprehensive evaluation fuzzy number when ;

[0024] The comprehensive evaluation fuzzy number is defuzzified based on the area centroid method.

[0025] Preferably, the method for calculating the consistency degree of the expert fuzzy rating is:

[0026]

[0027] In the formula, S(R u , R v ) indicates expert Zu and Expert Z v Consistency in the evaluation of the same event, ; a i For Expert Z u The fuzzy number corresponding to the evaluation result, b i For Expert Z v The fuzzy number corresponding to the evaluation result, i is the i-th fuzzy number;

[0028] S(R u , R v ) value, the greater the agreement between the two experts. u , R v ) = 0, indicating that the opinions of the two experts have no overlap;

[0029] The method for calculating the weighted consistency of the expert fuzzy evaluation is:

[0030]

[0031] Where N represents the number of evaluation experts, W(E v ) is expert Z v The weight of is the weighted consistency of the expert fuzzy evaluation, v is the expert Z v ;

[0032] The method for calculating the relative consistency of the expert fuzzy evaluation is:

[0033]

[0034] In the formula, is the relative consistency of the expert fuzzy evaluation, u is the expert Z u .

[0035] Preferably, the method for predicting the expert consensus coefficient is:

[0036]

[0037] In the formula, W(E u ) represents expert Z u The weight of is the relaxation factor, where ∈[0,1], when =0 means no preference; =1 indicates that the consensus of expert opinions is the same as the expert weight. is the expert consensus coefficient;

[0038] The method for calculating the comprehensive evaluation fuzzy number of experts when the node status is False is:

[0039]

[0040] In the formula, is the comprehensive evaluation fuzzy number, It is the product of the consensus coefficient of the Nth expert and his evaluation fuzzy number.

[0041] Preferably, the method for defuzzifying the comprehensive evaluation fuzzy number is:

[0042]

[0043] Where FPS is the fuzzy probability score, r1, r2, r3, and r4 are the comprehensive evaluation fuzzy numbers. The value corresponding to the fuzzy set of , x is a fuzzy number.

[0044] Preferably, the method for obtaining the root node prior probability is:

[0045]

[0046] Where FPS is the fuzzy probability score, and FFP is the root node prior probability of each event.

[0047] Compared with the prior art, the present invention has the following advantages and technical effects:

[0048] This method not only analyzes factors limited to the workload and training volume of test flight staff, but also analyzes the many risk factors that cause test flight human accidents covering various aspects such as operation, psychology, and supervision, so as to achieve a more comprehensive and integrated analysis of test flight human risks. The improved similarity aggregation method is used to aggregate expert opinions. The improved SAM not only considers the degree of consensus, but also solves the consensus differences caused by different individual cognitive levels during the aggregation process, so that the aggregation results are more inclined to the estimates of highly reliable experts. Combined with the Bayesian network, it can handle the cognitive uncertainty caused by the lack of data and make probabilistic predictions, thereby solving the problem of difficulty in computational analysis due to insufficient test flight data;

[0049] The present invention improves the HFACS model, and more scientifically and comprehensively identifies the numerous risk factors that cause human accidents in test flights, covering various aspects such as operation, psychology, and supervision; by improving the similarity aggregation method, the mutual influence of expert consistency and individual differences is considered, and the proportion of high-weight experts in the overall estimation results is increased, making the calculation results more objective and reliable; the Bayesian network can be used to deal with cognitive uncertainty caused by lack of data, perform probabilistic prediction, and intuitively find out key risk factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0051] Figure 1 A flow chart of a method for quantitatively evaluating human factors affecting civil aircraft flight test risks according to an embodiment of the present invention;

[0052] Figure 2 An example diagram illustrating an embodiment of the present invention;

[0053] Figure 3 The fault tree of the embodiment of the present invention is mapped into a Bayesian network schematic diagram. DETAILED DESCRIPTION

[0054] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0055] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0056] In order to solve the problem that it is difficult to conduct quantitative risk probability calculation research due to insufficient test flight accident data, and because the risk factor analysis level related to human factors in civil aircraft test flights is relatively single at this stage, a method for quantitatively evaluating human factors affecting civil aircraft test flight risks is proposed. This invention will focus on the many risk factors that cause test flight human accidents covering various aspects such as operation, psychology, and supervision, so as to achieve a more comprehensive and comprehensive analysis of test flight human risks; the improved similarity aggregation method is used to calculate fuzzy probability, and by considering the influence of individual differences of experts on the consistency of their opinions, the influence of high-credibility expert opinions on the calculation results is deepened, making the prediction results more accurate and reliable. Through the two-way reasoning analysis and sensitivity analysis of the Bayesian network, the key risk factors are intuitively found, which is helpful to formulate effective preventive measures and reduce the incidence of civil aircraft test flight accidents.

[0057] The present invention proposes a method for quantitatively evaluating the human factors affecting the risk of civil aircraft test flight, such as Figure 1 ,include:

[0058] Combined with the test flight data, the HFACS model of human factor analysis and classification system is adaptively improved, the logical correspondence of certain factors in HFACS is changed, so that the correlation of various factors is more suitable for the civil aircraft test flight process, and the human risk factors that affect the safety of civil aircraft test flight are identified;

[0059] By identifying the human risk factors, a fault tree model is constructed to obtain conditional probability;

[0060] Based on the mapping relationship between the fault tree model and the Bayesian network model, constructing a Bayesian network model;

[0061] The expert survey method is adopted, and the fuzzy opinions of experts with consensus differences are comprehensively considered based on the improved similarity aggregation method. The prior probability of the root node is obtained by aggregating the fuzzy opinions of experts and defuzzifying the experts' opinions.

[0062] The conditional probability and the root node prior probability are calculated and analyzed through the Bayesian network model to obtain key risk factors.

[0063] Furthermore, the human risk factors that affect the safety of civil aircraft test flights include: unsafe behavior factors, unsafe supervision factors, and organizational influence factors. See Table 1 for details.

[0064] Table 1

[0065]

[0066] Specifically, the mapping relationship between the fault tree model and the Bayesian network model is:

[0067] The top event is converted into a leaf node, the intermediate event is converted into an intermediate node in the Bayesian network, the bottom event is converted into a root node in the Bayesian network, and the logic gates of the fault tree correspond to the conditional probability table in the Bayesian network; wherein the logic gates include OR gates and AND gates.

[0068] The mapping relationship between fault tree and Bayesian network is as follows: Figure 3 ,During the mapping process, the top event is converted into a leaf node, the intermediate event is converted into an intermediate node in the Bayesian network, and the bottom event is converted into a root node in the Bayesian network. The logic gates of the fault tree correspond to the conditional probability table in the Bayesian network, where, It represents an "OR" gate, which means that two events are independent, and the occurrence of one of them will cause the upper event to occur; It represents an “AND” gate, which means that the upper event will occur only when two events occur at the same time.

[0069] Furthermore, the prior probability of the root node is obtained, including:

[0070] The relative importance of each expert's opinion is represented by assigning a weight coefficient to the expert, wherein the Delphi method is used to assign the weight coefficient of each expert, and the failure probability of the risk factor is represented based on the trapezoidal fuzzy number, that is, the root node prior probability.

[0071] The membership function of trapezoidal fuzzy number is:

[0072]

[0073] In the formula, x is a fuzzy number, r1, r2, r3, and r4 are comprehensive evaluation fuzzy numbers. The numerical value corresponding to the fuzzy set, F (x) is the membership function of the trapezoidal fuzzy number.

[0074] The process of aggregating expert fuzzy opinions and defuzzifying expert opinions includes:

[0075] The consistency degree of experts' fuzzy ratings, the weighted consistency of experts' fuzzy evaluations, and the relative consistency of experts' fuzzy evaluations were calculated respectively;

[0076] Based on the consistency of the expert fuzzy rating, the weighted consistency of the expert fuzzy evaluation, and the relative consistency of the expert fuzzy evaluation, the expert consensus coefficient is predicted, and the comprehensive evaluation fuzzy number of the expert when the node state is False is calculated;

[0077] The comprehensive evaluation fuzzy number is defuzzified based on the area centroid method.

[0078] The method for calculating the consistency of experts' fuzzy ratings is:

[0079]

[0080] In the formula, S(R u , R v ) indicates expert Z u and Expert Z v Consistency in the evaluation of the same event, ; a i For Expert Z u The fuzzy number corresponding to the evaluation result, b i For Expert Z v The fuzzy number corresponding to the evaluation result, i is the i-th fuzzy number;

[0081] S(R u , R v ) value, the greater the agreement between the two experts. u , R v ) = 0, indicating that the opinions of the two experts have no overlap;

[0082] The method for calculating the weighted consistency of expert fuzzy evaluation is:

[0083]

[0084] Where N represents the number of evaluation experts, W(E v ) is expert Z v The weight of is the weighted consistency of the experts’ fuzzy evaluation;

[0085] The method for calculating the relative consistency of expert fuzzy evaluation is:

[0086]

[0087] In the formula, is the relative consistency of experts’ fuzzy evaluation.

[0088] The method for predicting the expert consensus coefficient is:

[0089]

[0090] In the formula, W(E u ) represents expert Z u The weight of is the relaxation factor, where ∈[0,1], when =0 means no preference; =1 indicates that the consensus of expert opinions is the same as the expert weight. is the expert consensus coefficient;

[0091] The method for calculating the comprehensive evaluation fuzzy number of experts when the node status is False is:

[0092]

[0093] In the formula, is the comprehensive evaluation fuzzy number, It is the product of the consensus coefficient of the Nth expert and his evaluation fuzzy number.

[0094] The method for defuzzifying the comprehensive evaluation fuzzy number is:

[0095]

[0096] Where FPS is the fuzzy probability score, r1, r2, r3, and r4 are the comprehensive evaluation fuzzy numbers. The value corresponding to the fuzzy set of , x is a fuzzy number.

[0097] The method to obtain the prior probability of the root node is:

[0098]

[0099] Where FPS is the fuzzy probability score, and FFP is the root node prior probability of each event.

[0100] The present invention adopts the expert survey method to calculate the probability, and adopts the improved similarity aggregation method to comprehensively consider the expert fuzzy opinions of consensus differences to better deal with various types of uncertainties. Through the calculation processes of aggregating expert opinions and defuzzifying expert opinions, a more accurate fuzzy failure probability is finally obtained, making the prediction results more accurate and reliable. Finally, through the Bayesian network calculation and analysis, the key risk factors are obtained, and then targeted prevention and control are carried out to improve safety and reduce the occurrence of accidents.

[0101] In order to more clearly express the technical solution of the present invention, the following specific embodiments are provided to introduce the solution:

[0102] like Figure 2 Firstly, based on the risk factors identified by the present invention and combined with literature research, a fault tree model is established. Secondly, according to the mapping relationship between the fault tree model and the Bayesian network model, a Bayesian network model is established.

[0103] An expert survey was conducted based on the fuzzy theory set, and 20 experts in the field of flight testing were invited to use seven levels of natural language to evaluate. The natural language variables are shown in Table 2. Since experts have different professional titles, working time and age, the scores also have different reference values. Therefore, the weight coefficient of the experts should be assigned to represent the relative importance of each expert's opinion. This embodiment uses the Delphi method to assign the weight W(E u ), the expert weight scoring standard is shown in Table 3. Since the trapezoidal shape of the membership function is more effective for risk assessment, the present invention uses trapezoidal fuzzy numbers to represent the probability of occurrence of the bottom event. The trapezoidal fuzzy number uses = (r1, r2, r3, r4) means that, by definition, its membership function is:

[0104]

[0105] Table 2

[0106]

[0107] Table 3

[0108]

[0109] In order to ensure the objectivity and accuracy of the evaluation results, this embodiment uses the Improved Similarity Aggregation Method (ISAM) to test the consistency of the experts' language opinions, and performs comprehensive operations to reduce the impact of the differences in expert evaluation results on the probability of human risk events in civil aircraft test flights. Assume that experts Zi (i = 1, 2, . . . , n) use natural language to express their evaluation of various human risk events in test flights, and then convert the natural language into corresponding fuzzy numbers. The expert fuzzy opinion aggregation comprehensive operation and defuzzification process is as follows:

[0110] (1) Calculate the degree of consistency (Degree of Similarity) of experts’ fuzzy ratings.

[0111] Let Z u and Z v represents a group of experts, R u = (a1, a2, a3, a4) represents the fuzzy number corresponding to the evaluation result of expert Zu, R v = (b1, b2, b3, b4) represents the fuzzy number corresponding to the expert Zv evaluation result, S(R u , R v ) indicates expert Z u and Z v The consistency of the evaluation of the same event, where The calculation formula for the consistency level S of each group of expert evaluation results is:

[0112]

[0113] S(R u , R v ) value, the greater the agreement between the two experts. u , R v )=0, indicating that the opinions of the two experts have no intersection. The improved method will aggregate the opinions according to the credibility of the experts in the next step to improve the reliability of the aggregation results.

[0114] (2) Calculate the weighted agreement of experts’ fuzzy evaluation:

[0115]

[0116] Where N represents the number of evaluation experts, W(E v ) is expert Z v The weight of .

[0117] The improved SAM differs from the original SAM in that it incorporates the expert weights into the calculation of absolute agreement, namely weighted agreement (WA), thereby obtaining a more reasonable and reliable summary estimate.

[0118] (3) Calculate the relative agreement of experts’ fuzzy evaluation:

[0119]

[0120] (4) Expert consensus coefficient prediction:

[0121]

[0122] Where: W(E u ) represents expert Z u The weight of is the relaxation factor, which means W(E u ) relative to RA(E u ), among which ∈[0, 1], when = 0 means no preference; = 1 indicates that the consensus of expert opinions is the same as the expert weight.

[0123] (5) Calculate the comprehensive evaluation fuzzy number of the expert when the node status is False:

[0124]

[0125] (6) Defuzzification of comprehensive evaluation fuzzy numbers is performed based on the area centroid method. The purpose of defuzzification is to aggregate the “overall” fuzzy numbers and convert them into fuzzy probability scores (FPS).

[0126]

[0127] Since the present embodiment adopts a trapezoidal fuzzy number, the defuzzification process of the trapezoidal fuzzy number is as follows:

[0128]

[0129] (7) The fuzzy probability of basic event X is converted into failure probability:

[0130] In the fuzzy fault tree analysis method, since the failure probability of each basic event is expressed by a fuzzy number, there is a difference between the actual probability data and the calculated probability. In order to solve this problem, a function is proposed to convert the failure probability of the basic event into the failure probability, namely:

[0131]

[0132] Wherein, FPS is the fuzzy possibility score, and FFP is the fuzzy failure probability of each event. Please note that the FFP in this embodiment is the result of fuzzy estimation, which is actually the defuzzified clarity value obtained by FST.

[0133] Finally, the present invention will use GeNIe software to construct a Bayesian network evaluation model, and assign and evaluate the node variables. GeNIe is a Bayesian network software with strong visualization, good user interface and intuitive and transparent assignment reasoning. Bayesian network calculation requires the simultaneous acquisition of the root node failure probability and the conditional probability between nodes. The root node failure probability is the fuzzy failure probability calculated by formula (2)-formula (9), and the conditional probability can be obtained by the mapping relationship between the fault tree model and the Bayesian network model. Through the rich analysis functions of the software, firstly, forward reasoning analysis is performed to obtain the prior probability of the leaf node (i.e., the target node), i.e., the total probability of the occurrence of human-caused accidents in test flights; secondly, reverse reasoning analysis is performed to obtain the posterior probability of the root node, and the probability of failure of each risk factor is found when the human-caused accident is certain to occur; finally, through sensitivity analysis, it is determined which factors have the greatest impact on the target node, and the root node with the greatest impact on the occurrence of human-caused risks in test flights is found. By combining the two-way reasoning analysis and sensitivity analysis of the Bayesian network, we can eventually identify the key risk factors and conduct targeted prevention and control to reduce the occurrence of accidents and improve the safety of civil aircraft test flights.

[0134] This method not only analyzes factors such as workload and training volume of test flight staff, but also analyzes many risk factors that cause test flight human factors accidents, including operation, psychology, supervision, etc., to achieve a more comprehensive analysis of test flight human factors risks. The improved similarity aggregation method is used to aggregate expert opinions. The improved SAM not only considers the degree of consensus, but also solves the consensus differences caused by different individual cognitive levels during the aggregation process, making the aggregation results more inclined to the estimates of highly reliable experts. Combined with the Bayesian network, it can handle the cognitive uncertainty caused by the lack of data and make probability predictions, thereby solving the problem of difficulty in computational analysis due to insufficient test flight data.

[0135] The present invention improves the HFACS model and more scientifically and comprehensively identifies the numerous risk factors that cause human-caused accidents in test flights, including operation, psychology, supervision, etc. The improved similarity aggregation method integrates the mutual influence of expert consistency and individual differences, increases the proportion of high-weight experts in the overall estimation results, and makes the calculation results more objective and reliable. The Bayesian network can be used to deal with cognitive uncertainty caused by lack of data, perform probability prediction, and intuitively find key risk factors.

[0136] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for quantitatively evaluating the human factors affecting the risk of civil aircraft test flight. It is characterized in that include: Adaptively improve the HFACS model of the human factor analysis and classification system based on test flight data to identify human risk factors that affect civil aircraft test flight safety; By identifying the human risk factors, a fault tree model is constructed to obtain conditional probability; Based on the mapping relationship between the fault tree model and the Bayesian network model, constructing a Bayesian network model; The expert survey method is adopted, and the fuzzy opinions of experts with consensus differences are comprehensively considered based on the improved similarity aggregation method. The prior probability of the root node is obtained by aggregating the fuzzy opinions of experts and defuzzifying the experts' opinions. Obtaining the root node prior probability includes: The relative importance of each expert's opinion is represented by assigning a weight coefficient to the expert, wherein the Delphi method is used to assign the weight coefficient of each expert, and the failure probability of the risk factor is represented based on a trapezoidal fuzzy number, that is, the root node prior probability; The process of aggregating the expert fuzzy opinions and defuzzifying the expert opinions includes: The consistency of the expert fuzzy rating, the weighted consistency of the expert fuzzy evaluation, and the relative consistency of the expert fuzzy evaluation were calculated respectively; among them, the improved similarity aggregation method was used to test the consistency of the experts' language opinions, and a comprehensive calculation was performed to reduce the impact of the differences in the expert evaluation results on the probability of human risk events in civil aircraft test pilots; Based on the consistency of the expert fuzzy rating, the weighted consistency of the expert fuzzy evaluation, and the relative consistency of the expert fuzzy evaluation, the expert consensus coefficient is predicted, and the comprehensive evaluation fuzzy number of the expert when the node state is False is calculated; Defuzzifying the comprehensive evaluation fuzzy number based on the area centroid method; The method for calculating the consistency of the expert fuzzy rating is: In the formula, S(R u , R v ) indicates expert Z u and Expert Z v The consistency of the evaluation of the same event, S(R u , R v )∈[0,1]; a i For Expert Z u The fuzzy number corresponding to the evaluation result, b i For Expert Z v The fuzzy number corresponding to the evaluation result, i is the i-th fuzzy number; S(R u , R v ) value, the greater the agreement between the two experts. u , R v ) = 0, indicating that the opinions of the two experts have no overlap; The method for calculating the weighted consistency of the expert fuzzy evaluation is: Where N represents the number of evaluation experts, W(E v ) is expert Z v The weight of is the weighted consistency of the expert fuzzy evaluation, v is the expert Z v ; The method for calculating the relative consistency of the expert fuzzy evaluation is: In the formula, is the relative consistency of the expert fuzzy evaluation, u is the expert Z u ; The method for predicting the expert consensus coefficient is: In the formula, W(E u ) represents expert Z u β is the relaxation factor, where β∈[0,1]. When β=0, it means there is no preference. When β=1, it means the consensus of expert opinions is the same as the expert weight. is the expert consensus coefficient; The method for calculating the comprehensive evaluation fuzzy number of experts when the node status is False is: In the formula, is the comprehensive evaluation fuzzy number, It is the product of the consensus coefficient of the Nth expert and his evaluation fuzzy number; The method for obtaining the prior probability of the root node is: Where FPS is the fuzzy possibility score, and FFP is the root node prior probability of each event; The conditional probability and the root node prior probability are calculated and analyzed through the Bayesian network model to obtain key risk factors.

2. The method for quantitatively evaluating the human factors affecting the risk of civil aircraft flight test according to claim 1, It is characterized in that The human risk factors that affect the safety of civil aircraft test flights include: unsafe behavior factors, unsafe supervision factors and organizational influence factors.

3. The method for quantitatively evaluating the human factors affecting the risk of civil aircraft flight test according to claim 1, It is characterized in that The mapping relationship between the fault tree model and the Bayesian network model is: The top event is converted into a leaf node, the intermediate event is converted into an intermediate node in the Bayesian network, the bottom event is converted into a root node in the Bayesian network, and the logic gate of the fault tree corresponds to the conditional probability table in the Bayesian network; wherein the logic gate includes an OR gate and an AND gate.

4. The method for quantitatively evaluating the human factors affecting the risk of civil aircraft flight test according to claim 1, It is characterized in that The membership function of the trapezoidal fuzzy number is: In the formula, x is a fuzzy number, r 1 , r 2 , r 3 , r 4 are comprehensive evaluation fuzzy numbers The numerical value corresponding to the fuzzy set, F (x) is the membership function of the trapezoidal fuzzy number.

5. The method for quantitatively evaluating the human factors affecting the risk of civil aircraft flight test according to claim 1, It is characterized in that The method for defuzzifying the comprehensive evaluation fuzzy number is: Where FPS is the fuzzy probability score, r 1 , r 2 , r 3 , r 4 are comprehensive evaluation fuzzy numbers The value corresponding to the fuzzy set of , x is a fuzzy number.

Images