Improved FMEA system based on MCDM and Catboost

Abstract

The invention provides an improved FMEA system based on MCDM and Catboost, and the system is characterized in that the system comprises 1 expert ends and an FMEA end, and the FMEA end comprises a degree calculation module which is used for obtaining centrality and cause degrees corresponding to N failure modes respectively; the key failure mode screening module is used for screening n key failure modes from the N failure modes according to the centrality and the reason degree; the weight calculation module is used for calculating a risk factor comprehensive weight and an expert comprehensive weight, the comprehensive score calculation module is used for calculating a comprehensive score of each key failure mode, and the feature calculation module is used for calculating a feature value and a close degree of a cloud model of each key failure mode; and the risk level analysis module is used for obtaining risk levels corresponding to the n key failure modes. In a word, the method can select a key failure mode from all failure modes and obtain a relatively accurate risk level.

Description

Technical Field

[0001] The present invention relates to the technical field of reliability and risk analysis methods, and in particular to an improved FMEA system based on MCDM and Catboost. Background Art

[0002] Failure Mode and Effect Analysis (FMEA) is an important tool for risk analysis and evaluation. To identify high-risk failure modes in the target system, FMEA uses three risk factors: severity (S), occurrence (Occurrence, O) and detection (Detection, D). Experts score the potential failure modes in the target system from 1 to 10. To compare the risk levels of different failure modes, the scores of each failure mode under the three indicators of severity, occurrence and detection are multiplied. The product is defined as the Risk Priority Number (RPN) to measure the risk level of each failure mode. The higher the score, the higher the risk level of the failure mode, and the more urgent it is to control. By comparing the RPN values, the failure modes are divided and handled accordingly; if there is a failure mode that exceeds the predetermined RPN value, it will be scored after manual control and improvement; when the RPN values ​​of all failure modes are within an acceptable range, the work of using FMEA to improve system reliability is completed.

[0003] The traditional FMEA method is highly subjective, and the decision-making information that can be made in the data mining model is at a medium level, and there are certain shortcomings. First, it is difficult to accurately express the expert's professional knowledge only through a score of 1-10, and a large amount of information is lost in the evaluation process. Secondly, the three risk factors and the evaluation information of different experts are equalized, ignoring the differences between experts and risk factors in different application backgrounds. Third, the uncertainty in the target system is difficult to be reflected in the traditional FMEA.

[0004] This shows that the risk identification capability of traditional FMEA is limited and there is still much room for improvement. Summary of the invention

[0005] The present invention is made to solve the above problems, and aims to provide an improved FMEA system based on MCDM and Catboost.

[0006] The present invention provides an improved FMEA system based on MCDM and Catboost, which has the following characteristics, including: 1 expert terminal, held by 1 expert respectively, for the expert to input the failure data of the cold chain logistics distribution system; FMEA terminal, which is respectively connected to each expert terminal for communication, and is used to obtain the key failure modes and corresponding risk levels of the cold chain logistics distribution system according to all the failure data, wherein the failure data includes the direct impact matrix of 1 expert on N failure modes of the cold chain logistics distribution system, and the evaluation matrix, risk factor scoring matrix, least important risk factor and mutual evaluation matrix of 1 expert; the FMEA terminal includes a communication module, a degree calculation module, a key failure mode screening module, a weight calculation module, a comprehensive score calculation module, a feature calculation module and a risk level analysis module; the communication module is used to receive the failure data sent by each expert terminal. The data, the degree calculation module is used to calculate according to l direct impact matrices to obtain the centrality and causal degree corresponding to N failure modes respectively, the key failure mode screening module is used to screen out n failure modes as key failure modes from N failure modes according to the centrality and causal degree, the weight calculation module is used to calculate according to all evaluation matrices, risk factor scoring matrix, the least important risk factor and mutual evaluation matrix to obtain the comprehensive weight of risk factors and the comprehensive weight of experts, the comprehensive score calculation module is used to calculate according to the comprehensive weight of risk factors and the comprehensive weight of experts to obtain the comprehensive scores corresponding to the four characteristics of each key failure mode, the feature calculation module is used to calculate the eigenvalue and closeness of the cloud model of each key failure mode according to the comprehensive score, and the risk level analysis module is used to obtain the risk level corresponding to n key failure modes according to the eigenvalue and closeness.

[0007] The improved FMEA system based on MCDM and Catboost provided by the present invention may also have the following characteristics: wherein the centrality c corresponding to the i-th failure mode i and the cause degree r i The calculation formula is: i =R i +R j , r i =R i -R j , Where R i is the impact degree corresponding to the i-th failure mode, R j is the impact degree corresponding to the i-th failure mode, is the element in the i-th row and j-th column of the normalized matrix R corresponding to the z-th expert, is the element in the jth row and ith column of the normalized matrix R corresponding to the zth expert. The calculation formula of the normalized matrix R corresponding to the zth expert is: R = Q*(EQ) -1 , Where E is the unit matrix, Q ij is the element in the i-th row and j-th column of the matrix Q, P ij It is the element in the i-th row and j-th column of the direct influence matrix corresponding to the z-th expert. The direct influence matrix is ​​an N*N matrix.

[0008] The improved FMEA system based on MCDM and Catboost provided by the present invention may also have the following features: wherein, in the critical failure mode screening module, the screening process is specifically as follows: failure modes with negative cause degrees are greatly affected by other failure modes and belong to result factors; failure modes with higher centralities are more important; combined with the 80 / 20 principle, failure modes with negative cause degrees and centralities that are below 80% of the overall failure mode centrality level are screened out, and all remaining failure modes are critical failure modes.

[0009] The improved FMEA system based on MCDM and Catboost provided by the present invention may also have the following features: wherein the evaluation matrix is ​​a matrix of size n*3 obtained by experts performing interval direct fuzzy number scoring on n key failure modes under three risk factors of S, O, and D, and each element of the evaluation matrix includes μ L , μ U 、v L and v U The risk factor scoring matrix is ​​a 1*12 matrix obtained by experts performing interval direct fuzzy number scoring on the three risk factors S, O, and D. The least important risk factor is calculated according to the following formula: k ={a ik |max(count(a ik ),i=1,2,...,l)}, where a k is the least important risk factor, a ik is the least important risk factor given by the i-th expert, count(a ik ) is for a ik Counting is performed, and each expert evaluates all experts under z attributes to obtain z l*l mutual evaluation matrices. The element in the i-th row and j-th column of the s-th mutual evaluation matrix is ​​the score of the i-th expert on the j-th expert under the s-th attribute.

[0010] The improved FMEA system based on MCDM and Catboost provided by the present invention may also have the following features: wherein the weight calculation module includes: a risk factor objective weight calculation unit, which is used to calculate according to all evaluation matrices to obtain the objective risk factor weights of each expert for the three risk factors; an expert objective weight calculation unit, which is used to calculate according to all evaluation matrices and the objective risk factor weights to obtain the expert objective weights corresponding to l experts; a risk factor subjective weight calculation unit, which is used to calculate according to all risk factor scoring matrices and the least important risk factor to obtain the risk factor subjective weights corresponding to the three risk factors; an expert subjective weight calculation unit, which is used to calculate according to all mutual evaluation matrices to obtain the expert subjective weights corresponding to l experts; a risk factor comprehensive weight calculation unit, which is used to calculate according to the objective risk factor weights and the subjective risk factor weights combined with a genetic algorithm to obtain the risk factor comprehensive weights of l experts for the three risk factors; and an expert comprehensive weight calculation unit, which is used to calculate according to the expert objective weights and the expert subjective weights combined with a genetic algorithm to obtain the expert comprehensive weights corresponding to l experts.

[0011] The improved FMEA system based on MCDM and Catboost provided by the present invention may also have the following features: wherein the objective risk factor weight of the j-th risk factor corresponding to the i-th expert is The calculation formula is: In the formula is the anti-entropy value of the j-th risk factor corresponding to the i-th expert, is the element in the kth row and jth column of the evaluation matrix corresponding to the i-th expert The entropy value, μ L (x), μ U (x), v L (x) and v U (x) is the characteristic μ of the element L , characteristic μ U , Feature v L and feature v U , the expert objective weight α corresponding to the i-th expert i The calculation formula is: Where ξ ki is the grey correlation coefficient of the i-th expert for the k-th key failure mode, is the element in the kth row and jth column of the evaluation matrix corresponding to the i-th expert, is the comprehensive scoring matrix of the i-th expert, pki The comprehensive scoring matrix For each column in p k0 is the matrix obtained by taking the arithmetic average of the comprehensive scoring matrices of l experts For each column in , the risk factor subjective weight a corresponding to the j-th risk factor j The calculation formula is: Where μ L (y), μ U (y), v L (y) and v U (y) is the element in row i and column j of the total risk factor score matrix. The characteristic μ L , characteristic μ U , Feature v L and feature v U The total risk factor score matrix is ​​constructed based on all risk factor score matrices, x ij For elements The entropy value, x′ ij For element x ij The standardized entropy value, s k is the sum of the entropy values ​​of the corresponding elements of the l least important risk factors in the total risk factor scoring matrix, and the expert subjective weight δ corresponding to the i-th expert ib The calculation formula is: ic≠ib, where is the score given by the ia-th expert to the ib-th expert on the s-th attribute, is the score of the ia-th expert on the ic-th expert under the s-th attribute. The expression of the constraints and functions of the genetic algorithm in the risk factor comprehensive weight calculation unit is: In the formula is the comprehensive risk factor weight of the i-th expert for the g1-th risk factor, is the matrix corresponding to the entropy value of each element in the evaluation matrix of the i-th expert, g = 1, 2, ..., n, and the expression of the constraints and functions of the genetic algorithm in the expert comprehensive weight calculation unit is: min(α i ,δ ib )≤w g2 ≤max(α i ,δ ib ), Where w g2is the expert comprehensive weight corresponding to the h2th expert, Q g Based on The matrix obtained by calculating the entropy value.

[0012] The improved FMEA system based on MCDM and Catboost provided by the present invention may also have the following features: wherein, in the comprehensive score calculation module, the calculation expression of the comprehensive score is: In the formula It is the comprehensive score of each critical failure mode.

[0013] The improved FMEA system based on MCDM and Catboost provided by the present invention may also have the following features: wherein, in the feature calculation module, the feature values ​​are Ex1, En1, He1, Ex2, En2 and He2, (Ex1, En1, He1) is the support cloud of the key failure mode, (Ex2, En2, He2) is the opposition cloud of the key failure mode, and the closeness C of the kth key failure mode is k The calculation formula is: Similarity of positive shape in the formula To support cloud and positive ideal cloud o + =(maxEx 1k ,minEn 1k ,minHe 1k ) The calculated sim sk , positive distance similarity To support cloud and positive ideal cloud o + The calculated sim dk , maxEx 1k is the maximum value of the characteristic values ​​Ex1 of all critical failure modes, minEn 1k is the minimum value of the characteristic values ​​En1 of all critical failure modes, minHe 1k is the minimum value of the eigenvalues ​​He1 of all key failure modes, negative shape similarity For opposing clouds and negative ideal clouds o - =(maxEx 2k ,minEn 2k ,minHe 2k ) The calculated sim sk , negative distance similarity For opposing clouds and negative ideal clouds o - Respectively o k1 and k2 The calculated sim dk , maxEx 2kis the maximum value of the characteristic values ​​Ex2 of all critical failure modes, minEn 2k is the minimum value of the characteristic values ​​En2 of all critical failure modes, minHe 2k is the minimum value of the characteristic values ​​He2 of all critical failure modes, is a positive similarity, is a negative similarity.

[0014] The improved FMEA system based on MCDM and Catboost provided by the present invention may also have the following features: wherein the risk level analysis module stores a preset risk level prediction model, and the characteristic values ​​and proximity corresponding to each key failure mode are input into the risk level prediction model to obtain the risk level corresponding to each key failure mode. The risk level prediction model is constructed by taking the characteristic values ​​and proximity corresponding to the existing failure model as the input of the Catboost classification model, and taking the k-means clustering results corresponding to the characteristic values ​​and proximity as the labels of the failure model.

[0015] Functions and Effects of the Invention

[0016] According to the improved FMEA system based on MCDM and Catboost involved in the present invention, firstly, interval intuitionistic fuzzy numbers are used to evaluate failure modes, so as to improve the clarity of experts' evaluation information on failure modes, and failure modes are screened according to causality and centrality to obtain key failure modes, thereby saving resources spent on secondary failure modes; secondly, the improved G2 method and anti-entropy weight method are used to assign subjective and objective weights to risk factors, and the decision consensus model based on deviation and GRA are used to assign subjective and objective weights to experts, and the weights are combined by maximizing the difference, thereby solving the problem that traditional FMEA ignores the differences between risk factors and experts; then, the comprehensive scores of key failure modes are converted into cloud models, and the closeness of key failure modes is calculated, so as to sort the key failure modes, thereby solving the problem of insufficient RPN resolution accuracy; finally, the k-means algorithm is used to perform cluster analysis on failure modes, so as to obtain different risk levels for each key failure mode, and the Catboost classification model is used to predict the risk level of the failure mode. Therefore, the improved FMEA system based on MCDM and Catboost of the present invention can select key failure modes from all failure modes, obtain a more accurate risk level, and predict the risk level of potential failure modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A block diagram of an improved FMEA system in an embodiment of the present invention;

[0018] Figure 2 is a block diagram of the FMEA end in an embodiment of the present invention;

[0019] Figure 3 is a schematic diagram of k-means clustering analysis results of supporting clouds of various key failure modes when k=3 in an embodiment of the present invention;

[0020] Figure 4 is a schematic diagram of k-means clustering analysis results of the opposing clouds of various key failure modes when k=3 in an embodiment of the present invention;

[0021] Figure 5 Schematic diagram of a confusion matrix output by a Catboost classification model under leave-one-out cross validation in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments and accompanying drawings specifically illustrate the improved FMEA system based on MCDM and Catboost of the present invention.

[0023] The improved FMEA system based on MCDM and Catboost in this embodiment is used to analyze and evaluate the cold chain logistics distribution system.

[0024] Figure 1 A block diagram of an improved FMEA system in an embodiment of the present invention.

[0025] like Figure 1 As shown, the improved FMEA system 1 of this embodiment includes an expert terminal 10 and a FMEA terminal 20.

[0026] The l expert terminals 10 are held by l experts respectively and are used for the experts to input the failure data of the cold chain logistics distribution system. The failure data includes the direct impact matrix of the l experts on the N failure modes of the cold chain logistics distribution system, as well as the evaluation matrix of the l experts, the risk factor scoring matrix, the least important risk factor and the mutual evaluation matrix.

[0027] In this embodiment, l=4 experts are respectively A, a university scholar researching cold chain logistics and distribution, B, the person in charge of logistics technology research and development of a cold chain logistics company, C, the person in charge of logistics management of the cold chain logistics company, and D, the person in charge of distribution management of the cold chain logistics company.

[0028] In this embodiment, four experts determined N=18 failure modes of the cold chain logistics distribution system through literature review, brainstorming and expert interviews, as shown in Table 1:

[0029] Table 1 Failure mode

[0030]

[0031] In Table 1, the first column is the number corresponding to each failure mode, the second column is the specific content of each failure mode, the third column is the failure cause corresponding to each failure mode, and the fourth column is the failure consequence corresponding to each failure mode.

[0032] Figure 2 It is a block diagram of the FMEA terminal in an embodiment of the present invention.

[0033] like Figure 2 As shown, the FMEA terminal 20 is respectively connected to each expert terminal 10 for communication, and is used to obtain the key failure modes and corresponding risk levels of the cold chain logistics distribution system according to all failure data, including a communication module 201, a degree calculation module 202, a key failure mode screening module 203, a weight calculation module 204, a comprehensive score calculation module 205, a feature calculation module 206, a risk level analysis module 207 and a FMEA control module 208 for controlling the above modules.

[0034] The communication module 201 is used to receive failure data sent by each expert terminal.

[0035] The degree calculation module 202 is used to calculate according to l direct impact matrices to obtain the centrality and cause degree corresponding to N failure modes. In this embodiment, 4 experts use 5 granularity scores, namely M ij (M ij =0,1,2,3,4; j,j=1,...,N) to characterize the mutual influence degree among N=18 failure modes, M ij is the influence of the i-th failure mode on the j-th failure mode. When i=j, M ij = 0, and then construct the ij An 18*18 matrix with elements is used as the direct influence matrix.

[0036] Among them, the centrality c corresponding to the i-th failure mode i and the cause degree r i The calculation formula is:

[0037] c i =R i +R j ,

[0038] r i =R i -R j ,

[0039]

[0040]

[0041] Where R i is the impact degree corresponding to the i-th failure mode, Rj is the impact degree corresponding to the i-th failure mode, is the element in the i-th row and j-th column of the normalized matrix R corresponding to the z-th expert, is the element in the j-th row and i-th column of the normalized matrix R corresponding to the z-th expert.

[0042] The calculation formula of the normalized matrix R corresponding to the z-th expert is:

[0043] R=Q*(EQ) -1 ,

[0044]

[0045] Where E is the unit matrix, Q ij is the element in the i-th row and j-th column of the matrix Q, P ij is the element in the i-th row and j-th column of the direct influence matrix corresponding to the z-th expert.

[0046] In this embodiment, the influence degree, affected degree, centrality and cause degree corresponding to each failure mode are shown in Table 2:

[0047] Table 2 The influence, influence, centrality and cause degree of failure modes

[0048] Failure Mode Impact Influence Centrality Cause FM1 0.2903 1.1796 1.4698 -0.8893 FM2 0.5295 2.2160 2.7455 -1.6864 FM3 0.3113 2.9501 3.2614 -2.6389 FM4 0.6359 1.3276 1.9635 -0.6917 FM5 0.6365 1.6136 2.2501 -0.9771 FM6 0.7828 0.0000 0.7828 0.7828 FM7 0.5352 0.2563 0.7915 0.2790 FM8 1.0148 0.3984 1.4132 0.6165 FM9 0.6688 0.7290 1.3979 -0.0602 FM10 0.6476 0.0000 0.6476 0.6476 FM11 1.0209 0.7817 1.8026 0.2391 FM12 1.3692 0.0000 1.3692 1.3692 FM13 0.9412 0.1623 1.1035 0.7789 FM14 1.3358 0.0197 1.3555 1.3161 FM15 0.7476 0.6522 1.3998 0.0954 FM16 0.9448 1.0272 1.9720 -0.0824 FM17 1.1547 1.9847 3.1394 -0.8300 FM18 1.7314 0.0000 1.7314 1.7314

[0049] The first column in Table 2 is the number corresponding to each failure mode, and the second to fifth columns are the influence, influence, centrality and cause degree corresponding to each failure mode. For example, the cell in the fourth column of the second row indicates that the centrality corresponding to the failure mode numbered FM1 is 1.4698.

[0050] The critical failure mode screening module 203 is used to screen out n failure modes from N failure modes according to the centrality and the cause degree to obtain n failure modes as critical failure modes.

[0051] In this embodiment, the failure mode with negative cause degree is greatly affected by other failure modes and belongs to the result factor. The higher the centrality, the more important the failure mode. Combined with the 80 / 20 principle, the failure mode with the centrality level in the top 20% is more worthy of attention. Therefore, the screening process is as follows:

[0052] Failure modes with negative causal degrees and centralities below 80% of the overall failure mode centrality level are screened out, and all remaining failure modes are critical failure modes.

[0053] In this embodiment, n=13 key failure modes are obtained through the above screening as shown in Table 3:

[0054] Table 3 Key failure modes

[0055]

[0056]

[0057] The first column in Table 3 is the number corresponding to each key failure mode, and the second to fourth columns are the content, failure cause and failure consequence corresponding to each key failure mode.

[0058] The weight calculation module 204 is used to calculate according to all evaluation matrices, risk factor scoring matrices, the least important risk factors and mutual evaluation matrices to obtain risk factor comprehensive weights and expert comprehensive weights, including a risk factor objective weight calculation unit 2041, an expert objective weight calculation unit 2042, a risk factor subjective weight calculation unit 2043, an expert subjective weight calculation unit 2044, a risk factor comprehensive weight calculation unit 2045 and an expert comprehensive weight calculation unit 2046.

[0059] The risk factor objective weight calculation unit 2041 is used to perform calculations based on all evaluation matrices to obtain the risk factor objective weights of the three risk factors for each expert.

[0060] The evaluation matrix is ​​an n*3 matrix obtained by experts performing interval direct fuzzy number scoring on n key failure modes under the three risk factors of S, O, and D. Each element of the evaluation matrix includes μ L , μ U 、v L and v U Four features, in this embodiment, feature μ L is the minimum membership degree of the risk factor, μ U is the maximum membership degree of the risk factor, v L is the minimum non-membership degree of the risk factor, v U is the maximum non-membership degree of the risk factor, μ L , μ U 、v L and v U The value range of is [0,1].

[0061] The scores of each element in the evaluation matrix corresponding to the four experts in this embodiment are shown in Tables 4 to 7:

[0062] Table 4 A rating of university scholars

[0063]

[0064] Table 5 B score of the person in charge of logistics technology R&D

[0065]

[0066]

[0067] Table 6 C score of logistics management person in charge

[0068]

[0069] Table 7 D score of the person in charge of distribution management

[0070]

[0071]

[0072] Among them, the objective risk factor weight of the j-th risk factor corresponding to the i-th expert is The calculation formula is:

[0073]

[0074]

[0075]

[0076]

[0077] In the formula is the anti-entropy value of the j-th risk factor corresponding to the i-th expert, is the element in the kth row and jth column of the evaluation matrix corresponding to the i-th expert The entropy value, μ L (x), μ U (x), v L (x) and v U (x) is the characteristic μ of the element L , characteristic μ U , Feature v L and feature v U .

[0078] In this embodiment, the objective weight calculation results of the risk factors corresponding to each expert are: and

[0079] The expert objective weight calculation unit 2042 is used to calculate the expert objective weights corresponding to l experts according to all evaluation matrices and risk factor objective weights.

[0080] Among them, the expert objective weight α corresponding to the i-th expert i The calculation formula is:

[0081]

[0082]

[0083]

[0084]

[0085] Where ξ ki is the grey correlation coefficient of the i-th expert for the k-th key failure mode, is the element in the kth row and jth column of the evaluation matrix corresponding to the i-th expert, is the comprehensive scoring matrix of the i-th expert, p ki The comprehensive scoring matrix For each column in p k0 is the matrix obtained by taking the arithmetic average of the comprehensive scoring matrices of l experts In each column, ρ is the resolution coefficient, and the value of ρ in this embodiment is 0.5.

[0086] In this embodiment, the calculation results of the expert objective weights corresponding to the experts are α1=0.2698, α2=0.2774, α3=0.1855 and α4=0.2673.

[0087] The risk factor subjective weight calculation unit 2043 is used to calculate the risk factor subjective weights corresponding to the three risk factors according to all risk factor scoring matrices and the least important risk factor.

[0088] Among them, the risk factor scoring matrix is ​​a 1*12 matrix obtained by experts performing interval direct fuzzy number scoring on the three risk factors S, O, and D. The least important risk factor is calculated according to the following formula:

[0089] a k ={a ik |max(count(a ik ),i=1,2,...,l)},

[0090] Where a k is the least important risk factor, a ik is the least important risk factor given by the i-th expert, count(a ik ) is for a ik Count.

[0091] The scores of the experts on each risk factor in this embodiment are shown in Table 8:

[0092] Table 8 Risk factor scores

[0093]

[0094] In Table 8, the first column is for each expert, the second to thirteenth columns are the risk factor scores of each expert on the three risk factors, and the fourteenth column is the least important risk factor given by each expert.

[0095] The subjective weight a of the risk factor corresponding to the jth risk factor j The calculation formula is:

[0096]

[0097]

[0098]

[0099]

[0100]

[0101] Where μ L (y), μ U (y), v L (y) and v U (y) is the element in row i and column j of the total risk factor score matrix. The characteristic μ L , characteristic μ U , Feature v L and feature v U The total risk factor score matrix is ​​constructed based on all risk factor score matrices, x ij For elements The entropy value, x′ ij For element x ij The standardized entropy value, s k It is the sum of the entropy values ​​of the corresponding elements of the l least important risk factors in the total risk factor scoring matrix.

[0102] In this embodiment, the calculation results of the risk factor subjective weights a1, a2 and a3 are 0.2693, 0.4614 and 0.2693 respectively.

[0103] The expert subjective weight calculation unit 2044 is used to calculate the expert subjective weights corresponding to l experts according to all mutual evaluation matrices.

[0104] Among them, each expert evaluates all experts under z attributes to obtain z l*l mutual evaluation matrices. The element in the i-th row and j-th column of the s-th mutual evaluation matrix is ​​the score of the i-th expert on the j-th expert under the s-th attribute.

[0105] In this embodiment, four experts perform 9-granularity scoring under z=2 attributes, namely, influence e1 and intimacy e2. The nine-granularity scoring includes s0, s1, s2, s3, s4, s5, s6, s7 and s8, which represent none, very low, low, relatively low, medium, relatively high, high, very high and complete respectively. The expert's score of his influence e1 is set to medium, i.e., s4, and the score of his intimacy e2 is set to complete, i.e., s8. The results of all mutual evaluations are shown in Table 9:

[0106] Table 9 All mutual evaluation results

[0107]

[0108] The first column in the table shows the experts who made the scores. The second to fifth columns show the scores of experts A, B, C, and D given by each expert at influence e1. The sixth to ninth columns show the scores of experts A, B, C, and D given by each expert at intimacy e2. For example, the cell in the third row and third column indicates that expert A's score for expert B's influence e1 is s7.

[0109] The expert subjective weight δ corresponding to the i-th expert ib The calculation formula is:

[0110]

[0111]

[0112]

[0113]

[0114]

[0115] In the formula is the score given by the ia-th expert to the ib-th expert on the s-th attribute, is the score given by the ia-th expert to the ic-th expert on the s-th attribute.

[0116] In this embodiment, the calculation results of the expert subjective weights corresponding to the experts are δ1=0.2496, δ2=0.2494, δ3=0.2579 and δ4=0.2431.

[0117] The risk factor comprehensive weight calculation unit 2045 is used to calculate the risk factor comprehensive weights of the three risk factors by l experts respectively according to the objective weights of the risk factors and the subjective weights of the risk factors in combination with the genetic algorithm.

[0118] Among them, the expression of the constraint and function of the genetic algorithm in the risk factor comprehensive weight calculation unit 2045 is:

[0119]

[0120]

[0121]

[0122]

[0123] In the formula is the comprehensive risk factor weight of the i-th expert for the g1-th risk factor, is the matrix corresponding to the entropy value of each element in the evaluation matrix of the i-th expert, g = 1, 2, ..., n.

[0124] In this example, the calculation results of the comprehensive weights of the risk factors of the three risk factors by the experts are: and

[0125] The expert comprehensive weight calculation unit 2046 is used to calculate the expert comprehensive weights corresponding to l experts based on the expert objective weights and the expert subjective weights in combination with the genetic algorithm.

[0126] Among them, the expression of the constraint and function of the genetic algorithm in the expert comprehensive weight calculation unit 2046 is:

[0127]

[0128]

[0129] min(α i ,δ ib )≤w g2 ≤max(α i ,δ ib ),

[0130]

[0131] Where w g2 is the comprehensive expert weight corresponding to the g2th expert, Q g Based on The matrix obtained by calculating the entropy value.

[0132] In this embodiment, the calculation results of the expert comprehensive weights corresponding to the experts are w1=0.2698, w2=0.2774, w3=0.1855 and w4=0.2673.

[0133] The comprehensive score calculation module 205 is used to calculate according to the comprehensive weight of the risk factors and the comprehensive weight of the experts to obtain the comprehensive scores corresponding to the four characteristics of each key failure mode.

[0134] The calculation expression of the comprehensive score is:

[0135]

[0136] In the formula It is the comprehensive score of each critical failure mode.

[0137] The comprehensive scores of each key failure mode in this embodiment are shown in Table 10:

[0138] Table 10 Comprehensive scores of each key failure mode

[0139]

[0140]

[0141] The first column in Table 10 is the number of each key failure mode, and the second to fifth columns are the key failure modes in μ L , μ U 、v L and v U The comprehensive score below.

[0142] The feature calculation module 206 is used to calculate the feature value and closeness of the cloud model of each key failure mode according to the comprehensive score.

[0143] Among them, the eigenvalues ​​are Ex1, En1, He1, Ex2, En2 and He2, (Ex1, En1, He1) is the supporting cloud of the critical failure mode, and (Ex2, En2, He2) is the opposing cloud of the critical failure mode.

[0144] The calculation formulas for each characteristic value in this embodiment are as follows:

[0145]

[0146]

[0147]

[0148]

[0149]

[0150]

[0151] Where μ L(x) is the critical failure mode in μ L The comprehensive score under U (x) is the critical failure mode in μ U The comprehensive score under v L (x) is the critical failure mode at v L The comprehensive score under v U (x) is the critical failure mode at v U The comprehensive score below.

[0152] The characteristic values ​​corresponding to the key failure modes in this embodiment are shown in Table 11:

[0153] Table 11 Characteristic values ​​corresponding to each key failure mode

[0154]

[0155] The first column in Table 11 is the number corresponding to each key failure mode, and the second to seventh columns are the eigenvalues ​​Ex1, En1, He1, Ex2, En2 and He2 corresponding to each key failure mode.

[0156] The closeness C of the kth critical failure mode k The calculation formula is:

[0157]

[0158]

[0159]

[0160]

[0161]

[0162] The positive shape similarity To support cloud and positive ideal cloud o + =(maxEx 1k ,minEn 1k ,minHe 1k ) The calculated sim sk , positive distance similarity To support cloud and positive ideal cloud o + The calculated sim dk , maxEx 1k is the maximum value of the characteristic values ​​Ex1 of all critical failure modes, minEn 1k is the minimum value of the characteristic values ​​En1 of all critical failure modes, minHe 1kis the minimum value of the eigenvalues ​​He1 of all key failure modes, negative shape similarity For opposing clouds and negative ideal clouds o - =(maxEx 2k ,minEn 2k ,minHe 2k ) The calculated sim sk , negative distance similarity For opposing clouds and negative ideal clouds o - The calculated sim dk ,maxEn 2k is the maximum value of the characteristic values ​​Ex2 of all critical failure modes, minEn 2k is the minimum value of the characteristic values ​​En2 of all critical failure modes, minHw 2k is the minimum value of the characteristic values ​​He2 of all critical failure modes, is a positive similarity, is a negative similarity.

[0163] The similarity and closeness of each key failure mode in this embodiment are shown in Table 12:

[0164] Table 12 Similarity and closeness of each key failure mode

[0165]

[0166] The first column in Table 12 is the number corresponding to each key failure mode, the second to eighth columns are the positive shape similarity, positive distance similarity, positive similarity, negative shape similarity, negative distance similarity, negative similarity and closeness corresponding to each key failure mode, and the ninth column is the sorting sequence of the closeness of each key failure mode from large to small.

[0167] The risk level analysis module 207 is used to obtain the risk level corresponding to the critical failure mode according to the characteristic value and the closeness.

[0168] Among them, the risk level analysis module stores a preset risk level prediction model, and inputs the characteristic values ​​and proximity corresponding to each key failure mode into the risk level prediction model to obtain the risk level corresponding to each key failure mode. The risk level prediction model is constructed by taking the characteristic values ​​and proximity corresponding to the existing failure model as the input of the Catboost classification model, and taking the k-means clustering results corresponding to the characteristic values ​​and proximity as the labels of the failure model.

[0169] Figure 3 3 is a schematic diagram of the k-means clustering analysis results of the support clouds of each key failure mode when k=3 in an embodiment of the present invention.

[0170] like Figure 3 As shown, the three coordinate axes represent the eigenvalue Ex1, eigenvalue En1 and eigenvalue He1 of each key failure mode, the three circles are the three cluster centers when k=3, representing low risk, medium risk and high risk respectively, the rectangle indicates that the cluster analysis results of the key failure modes corresponding to FM4 and FM13 are high risk, the triangle indicates that the cluster analysis results of the key failure modes corresponding to FM2, FM3, FM5, FM11 and FM12 are medium risk, and the diamond indicates that the cluster analysis results of the key failure modes corresponding to FM1, FM6, FM7, FM8, FM9 and FM10 are low risk.

[0171] Figure 4 3 is a schematic diagram of the k-means clustering analysis results of the opposition clouds of various key failure modes when k=3 in an embodiment of the present invention.

[0172] like Figure 4 As shown, the three coordinate axes represent the eigenvalue Ex2, eigenvalue En2 and eigenvalue He2 of each key failure mode respectively. The three circles are the three cluster centers when k=3, representing low risk, medium risk and high risk respectively. The rectangle indicates that the cluster analysis results of the key failure modes corresponding to FM4 and FM13 are high risk. The triangle indicates that the cluster analysis results of the key failure modes corresponding to FM2, FM3, FM5, FM11 and FM12 are medium risk. The diamond indicates that the cluster analysis results of the key failure modes corresponding to FM1, FM6, FM7, FM8, FM9 and FM10 are low risk.

[0173] Depend on Figure 3 and Figure 4 It can be seen that the risk levels corresponding to the key failure modes of FM1, FM6, FM7, FM8, FM9 and FM10 are all low risks, that is, the impact on the cold chain logistics distribution system is not significant, and only a plan needs to be made or less resources need to be invested for prevention and control; the risk levels corresponding to the key failure modes of FM2, FM3, FM5, FM11 and FM12 are medium risks, that is, they have a certain impact on the cold chain logistics distribution system, and medium resources need to be invested for management and control; the risk levels corresponding to the key failure modes of FM4 and FM13 are high risks, that is, they have a greater impact on the cold chain logistics distribution system, which may affect all aspects of the entire system and cause various chain risks, so more resources need to be invested for prevention and monitoring.

[0174] Figure 5 Schematic diagram of a confusion matrix output by a Catboost classification model under leave-one-out cross validation in an embodiment of the present invention.

[0175] like Figure 5As shown in the figure, the horizontal axis is the risk level predicted by the Catboost classification model for each key failure mode, that is, the predicted value, and the vertical axis is the actual risk level of each key failure mode, that is, the actual value. It can be seen that the Catboost classification model can better predict the risk level of the key failure mode through the six eigenvalues ​​and closeness of the cloud model.

[0176] Functions and Effects of the Embodiments

[0177] According to the improved FMEA system based on MCDM and Catboost involved in this embodiment, firstly, interval intuitive fuzzy numbers are used to evaluate failure modes, improve the clarity of experts' evaluation information on failure modes, and screen failure modes according to cause degree and centrality to obtain key failure modes, saving resources spent on secondary failure modes; secondly, the improved G2 method and anti-entropy weight method are used to assign subjective and objective weights to risk factors, and the decision consensus model based on deviation and GRA are used to assign subjective and objective weights to experts, and the weights are combined by maximizing the difference, which solves the problem that traditional FMEA ignores the differences between risk factors and experts; then, the comprehensive score of the key failure mode is converted into a cloud model, and the closeness of the key failure mode is calculated to sort the key failure mode, thereby solving the problem of insufficient RPN resolution accuracy; finally, the k-means algorithm is used to cluster the failure mode to obtain different risk levels of each key failure mode, and the Catboost classification model is used to predict the risk level of the failure mode. In short, this method can select key failure modes from all failure modes, obtain more accurate risk levels, and predict the risk level of potential failure modes.

[0178] The above-mentioned embodiments are preferred examples of the present invention and are not intended to limit the protection scope of the present invention.

Claims

1. An improved FMEA system based on MCDM and Catboost, characterized in that: include: l expert terminals, each held by l experts, for the experts to input failure data of the cold chain logistics distribution system; The FMEA terminal is respectively connected to each of the expert terminals for obtaining the key failure modes and corresponding risk levels of the cold chain logistics distribution system according to all the failure data. The failure data includes the direct impact matrix of each of the experts on the N failure modes of the cold chain logistics distribution system, and the evaluation matrix, risk factor scoring matrix, least important risk factor and mutual evaluation matrix of each of the experts. The FMEA terminal includes a communication module, a degree calculation module, a key failure mode screening module, a weight calculation module, a comprehensive score calculation module, a feature calculation module and a risk level analysis module. The communication module is used to receive the failure data sent by each of the expert terminals, The degree calculation module is used to calculate according to the l direct impact matrices to obtain the centrality and cause degree corresponding to the N failure modes respectively. The key failure mode screening module is used to screen out n failure modes from the N failure modes according to the centrality and the cause degree to obtain n failure modes as the key failure modes, The weight calculation module is used to calculate based on all the evaluation matrices, the risk factor scoring matrix, the least important risk factor and the mutual evaluation matrix to obtain the risk factor comprehensive weight and the expert comprehensive weight. The comprehensive score calculation module is used to calculate according to the comprehensive weight of the risk factor and the comprehensive weight of the expert to obtain the comprehensive scores corresponding to the four characteristics of each of the key failure modes. The feature calculation module is used to calculate the feature value and closeness of the cloud model of each key failure mode according to the comprehensive score. The risk level analysis module is used to obtain the risk levels corresponding to the n key failure modes according to the characteristic values ​​and the closeness.

2. The improved FMEA system based on MCDM and Catboost according to claim 1, characterized in that: in, The centrality c corresponding to the failure mode described in the ith item i and the cause degree r i The calculation formula is: c i =R i +R j , r i =R i -R j , Where R i is the impact degree corresponding to the failure mode described in the ith item, R j is the impact degree corresponding to the failure mode described in the ith item, is the element in the i-th row and j-th column of the normalized matrix R corresponding to the z-th expert, is the element in the jth row and ith column of the normalized matrix R corresponding to the zth expert, The calculation formula of the normalized matrix R corresponding to the z-th expert is: R=Q*(E-Q) -1 , Where E is the unit matrix, Q ij is the element in the i-th row and j-th column of the matrix Q, P ij is the element in the i-th row and j-th column of the direct influence matrix corresponding to the z-th expert, The direct impact matrix is ​​an N*N matrix.

3. The improved FMEA system based on MCDM and Catboost according to claim 1, characterized in that: in, In the critical failure mode screening module, the screening process is specifically as follows: Failure modes with negative causal degrees are greatly affected by other failure modes and belong to result factors; the higher the centrality, the more important the failure mode. Combined with the 80 / 20 principle, the failure modes with negative causal degrees and centralities below 80% of the overall failure mode centrality level are screened out, and all remaining failure modes are the critical failure modes.

4. The improved FMEA system based on MCDM and Catboost according to claim 1, characterized in that: in, The evaluation matrix is ​​a matrix of size n*3 obtained by the experts performing interval direct fuzzy number scoring on the n critical failure modes under the three risk factors of S, O, and D. Each element of the evaluation matrix includes μ L , μ U 、v L and v U Four characteristics, The risk factor scoring matrix is ​​a 1*12 matrix obtained by the experts performing interval direct fuzzy number scoring on the three risk factors S, O, and D. The least important risk factor is calculated according to the following formula: a k ={a ik |max(count(a ik ),i=1,2,...,l)}, Where a k is the least important risk factor, a ik is the least important risk factor given by the i-th expert, count(a ik ) is for a ik To count, Each of the experts evaluates all the experts under z attributes to obtain z l*l-sized mutual evaluation matrices, and the element in the i-th row and j-th column of the s-th mutual evaluation matrix is ​​the score of the i-th expert on the j-th expert under the s-th attribute.

5. According to the improved FMEA system based on MCDM and Catboost according to claim 4, Features: in, The weight calculation module includes: A risk factor objective weight calculation unit, used to calculate according to all the evaluation matrices to obtain the risk factor objective weights of the three risk factors for each of the experts; An expert objective weight calculation unit, used to calculate according to all the evaluation matrices and the objective weights of the risk factors to obtain the expert objective weights corresponding to l of the experts; A risk factor subjective weight calculation unit, used to calculate according to all the risk factor scoring matrices and the least important risk factor to obtain the risk factor subjective weights corresponding to the three risk factors; An expert subjective weight calculation unit, used to calculate the expert subjective weights corresponding to the l experts according to all the mutual evaluation matrices; A risk factor comprehensive weight calculation unit, used to calculate the risk factor comprehensive weights of the three risk factors respectively by the experts according to the objective weights of the risk factors and the subjective weights of the risk factors in combination with a genetic algorithm; The expert comprehensive weight calculation unit is used to calculate the expert comprehensive weights corresponding to the l experts according to the expert objective weights and the expert subjective weights in combination with a genetic algorithm.

6. The improved FMEA system based on MCDM and Catboost according to claim 5, characterized in that: in, The objective risk factor weight of the j-th risk factor corresponding to the i-th expert The calculation formula is: In the formula is the anti-entropy value of the j-th risk factor corresponding to the i-th expert, is the element in the kth row and jth column of the evaluation matrix corresponding to the i-th expert The entropy value, μ L (x), μ U (x), v L (x) and v U (x) is the characteristic μ of the element L , characteristic μ U , Feature v L and feature v U , The objective weight α of the expert corresponding to the i-th expert i The calculation formula is: Where ξ ki is the grey correlation coefficient of the i-th expert for the k-th key failure mode, is the element in the kth row and jth column of the evaluation matrix corresponding to the i-th expert, is the comprehensive scoring matrix of the i-th expert, p ki The comprehensive scoring matrix For each column in p k0 is the matrix obtained by taking the arithmetic average of the comprehensive scoring matrices of l experts For each column in The subjective weight a of the risk factor corresponding to the jth risk factor j The calculation formula is: Where μ L (y), μ U (y), v L (y) and v U (y) is the element in row i and column j of the total risk factor score matrix. The characteristic μ L , characteristic μ U , Feature v L and feature v U The total risk factor score matrix is ​​constructed based on all the risk factor score matrices, x ij For elements The entropy value, x′ ij For element x ij The standardized entropy value, s k is the sum of the entropy values ​​of the corresponding elements of the l least important risk factors in the total risk factor scoring matrix, The expert subjective weight δ corresponding to the i-th expert ib The calculation formula is: In the formula is the score given by the ia-th expert to the ib-th expert on the s-th attribute, is the score given by the ia-th expert to the ic-th expert under the s-th attribute, The constraints and functions of the genetic algorithm in the risk factor comprehensive weight calculation unit are expressed as follows: In the formula is the comprehensive risk factor weight of the i-th expert for the g1-th risk factor, is the matrix corresponding to the entropy value of each element in the evaluation matrix of the i-th expert, g = 1, 2, ..., n, The expression of the constraints and functions of the genetic algorithm in the expert comprehensive weight calculation unit is: min(a i ,d ib )≤w g2 ≤max(α i ,d ib ), Where w g2 is the comprehensive expert weight corresponding to the g2th expert, Q g Based on The matrix obtained by calculating the entropy value.

7. The improved FMEA system based on MCDM and Catboost according to claim 6, characterized in that: in, In the comprehensive score calculation module, the calculation expression of the comprehensive score is: In the formula It is the comprehensive score of each critical failure mode.

8. The improved FMEA system based on MCDM and Catboost according to claim 1, characterized in that: in, In the feature calculation module, the feature values ​​are Ex1, En1, He1, Ex2, En2 and He2, (Ex1, En1, He1) is the support cloud of the key failure mode, (Ex2, En2, He2) is the opposition cloud of the key failure mode, The closeness C of the kth critical failure mode k The calculation formula is: The positive shape similarity To support the cloud and positive ideal cloud o + =(maxEx 1k ,minEn 1k ,minHe 1k ) The calculated sim sk , positive distance similarity To support the cloud and positive ideal cloud o + The calculated sim dk , maxEx 1k is the maximum value of the characteristic values ​​Ex1 of all critical failure modes, minEn 1k is the minimum value of the characteristic values ​​En1 of all critical failure modes, minHe 1k is the minimum value of the characteristic values ​​He1 of all critical failure modes, Negative shape similarity For opposing clouds and negative ideal clouds o - =(maxEx 2k ,minEn 2k ,minHe 2k ) The calculated sim sk , negative distance similarity For opposing clouds and negative ideal clouds o - The calculated sim dk , maxEx 2k is the maximum value of the characteristic values ​​Ex2 of all critical failure modes, minEn 2k is the minimum value of the characteristic values ​​En2 of all critical failure modes, minHe 2k is the minimum value of the characteristic values ​​He2 of all critical failure modes, is a positive similarity, is a negative similarity.

9. The improved FMEA system based on MCDM and Catboost according to claim 1, characterized in that: in, The risk level analysis module stores a preset risk level prediction model, and inputs the characteristic value and the closeness corresponding to each of the key failure modes into the risk level prediction model to obtain the risk level corresponding to each of the key failure modes. The risk level prediction model is constructed by taking the eigenvalues ​​and closeness corresponding to the existing failure model as the input of the Catboost classification model, and taking the k-means clustering results corresponding to the eigenvalues ​​and closeness as the labels of the failure model.

Images