Improved fmea apparatus considering uncertainty and failure mode interrelationships

Abstract

This invention provides an improved FMEA device that considers the relationship between uncertainty and failure modes. It features the following components: an SVTraNN scoring conversion module for obtaining the SVTraNN evaluation matrix; an expert weight calculation module for calculating expert weights; an expert evaluation summarization module for calculating the SVTraNN decision matrix; a subjective weight calculation module for calculating the final subjective weight of each risk factor; an objective weight calculation module for calculating the objective weight of each risk factor; a comprehensive weight calculation module for calculating the comprehensive weight of each risk factor; a comprehensive score calculation module for calculating the comprehensive score of each risk mode; a risk ranking module for obtaining the risk ranking result of each risk mode; and a risk level module for obtaining the risk level of each risk mode. In summary, this method can effectively assess the risk level of risk modes.

Description

Technical Field

[0001] This invention belongs to the field of reliability and risk analysis methods, and specifically relates to an improved FMEA device that considers the relationship between uncertainty and failure modes. Background Technology

[0002] Failure Mode and Effects Analysis (FMEA) is a proactive reliability and risk analysis technique. Before a failure occurs, it analyzes the causes and effects of each failure mode to identify system weaknesses and critical components, enabling decision-makers to prioritize risk management activities and implement preventative measures. Today, FMEA is widely used in reliability and risk management research in fields such as aerospace, equipment maintenance, and medicine.

[0003] Traditional FMEA methods use discrete values ​​from 1 to 10 to evaluate the severity (S), occurrence (O), and detectability (D) of failure modes, and then multiply these three values ​​to obtain the Risk Priority Number (RPN), which is used to measure the risk level of each failure mode. By comparing RPNs, failure modes can be prioritized to assist in risk management decisions. However, the completeness and robustness of traditional FMEA methods have been questioned by many scholars, which can be summarized in four points: (1) The evaluation information of experts on failure modes is fuzzy and uncertain, and it is difficult to express it through discrete values; (2) It does not consider the weight of evaluation experts and risk factors, and does not fully consider the differences between experts and the different impacts of risk factors on the research object; (3) The calculation of RPN lacks robustness and scientificity. RPN is sensitive to changes in risk factor evaluation, and different combinations of risk factor evaluations may have the same RPN value but different meanings; (4) It does not consider the interrelationships between failure modes. Therefore, traditional FMEA methods still have considerable room for improvement. Summary of the Invention

[0004] The present invention was made to solve the above-mentioned problems, and aims to provide an improved FMEA device that takes into account the relationship between uncertainty and failure modes.

[0005] This invention provides an improved FMEA device that considers the relationship between uncertainty and failure modes. It is used to obtain the risk ranking and risk level of each potential AI safety risk mode based on expert ratings and risk factor importance scores. The device includes: an SVTraNN rating conversion module storing a preset SVTraNN language variable conversion table, used to obtain the SVTraNN evaluation matrix for each expert based on the SVTraNN language variable conversion table and rating data; an expert weight calculation module, used to calculate the expert weight for each risk factor based on the SVTraNN scores corresponding to that risk factor in all SVTraNN evaluation matrices; an expert evaluation summarization module, used to calculate the SVTraNN decision matrix based on all SVTraNN evaluation matrices and expert weights; and a subjective weight calculation module, storing the SVTraNN language variable conversion table and the SVTraN-FUCOM programming model, used to calculate the subjective weight based on the SVTraNN... The system employs a language variable transformation table, an SVTraN-FUCOM programming model, and importance score data for all risk factors to calculate the final subjective weights for each risk factor. An objective weight calculation module stores a difference-based goal programming model and calculates the objective weights for each risk factor based on the SVTraNN decision matrix and the difference-based goal programming model. A comprehensive weight calculation module calculates the comprehensive weights for each risk factor based on all final subjective and objective weights. A comprehensive score calculation module calculates the comprehensive score for each potential AI safety risk pattern based on the SVTraN-CODAS method, the SVTraNN decision matrix, and all comprehensive weights. A risk ranking module ranks each potential AI safety risk pattern based on all comprehensive scores. A risk level module performs clustering and hierarchical analysis based on the comprehensive score, score data, and comprehensive weights to obtain the risk level of each potential AI safety risk pattern; the higher the risk ranking, the higher the risk of the corresponding potential AI safety risk pattern.

[0006] The improved FMEA apparatus considering the relationship between uncertainty and failure modes provided by this invention may also have the following features: the expert weight calculation module includes: an average similarity calculation unit, used to calculate the average similarity corresponding to each risk factor based on the SVTraNN evaluation matrix of each pair of experts; a confidence level calculation unit, used to calculate the confidence level corresponding to each expert for each risk factor based on the average similarity between each pair of experts; and an expert weight calculation unit, used to calculate the expert weight corresponding to each expert for each risk factor based on the confidence levels of all experts.

[0007] The improved FMEA apparatus provided by this invention, which considers the relationship between uncertainty and failure modes, may also have the following feature: wherein the average similarity AS between expert k and expert l for risk factor S is... S (e k ,e l The calculation expression for ) is: f = <(a f ,b f ,c f ,d f );T f ,I f ,F f >,g=<(a g ,b g ,c g ,d g );T g ,I g ,F g >, In the formula, m represents the total number of potential AI security risk patterns, and f and g are two SVTraNNs, respectively. Let SVTraNN be the SVTraNN corresponding to the i-th potential AI security risk pattern in the SVTraNN evaluation matrix corresponding to expert k. Let x be the SVTraNN corresponding to the i-th potential AI security risk pattern in the SVTraNN evaluation matrix corresponding to expert l. fT x gT y fT y gT Corresponding to true membership degree, x fI x gI y fI y gI Corresponding to uncertain membership degree, x fF x gF y fF y gF Corresponding to the pseudo-membership degree, the confidence level CL of expert k under risk factor S. S (e k The calculation expression for ) is: In the formula, t represents the total number of experts, and ws represents the expert weight ws corresponding to expert k under risk factor S. k The calculation expression is:

[0008] The improved FMEA apparatus provided by this invention, which considers the relationship between uncertainty and failure modes, may also have the following feature: where the SVTraNN evaluation matrix H corresponding to expert k... k The expression is: In the formula, columns 1, 2, and 3 correspond to the scores of risk factors S, O, and D, respectively. The expression for the SVTraNN decision matrix is: In the formula, i = 1, ..., m, where m is the total number of potential AI security risk patterns. and These represent the scores of the i-th potential AI safety risk pattern in the SVTraNN evaluation matrix corresponding to the k-th expert, under risk factors S, O, and D, respectively, and ws k wo k and wd k These represent the expert weights for the k-th expert under risk factors S, O, and D, respectively.

[0009] The improved FMEA apparatus considering the relationship between uncertainty and failure modes provided by this invention may also have the following features: the subjective weight calculation module includes: an SVTraNN transformation unit, used to transform the risk factor importance score data according to the SVTraNN linguistic variable transformation table to obtain the SVTraNN data corresponding to each expert; a subjective weight solving unit, used to solve the SVTraN-FUCOM programming model according to the SVTraNN data corresponding to each expert, respectively, to obtain the subjective weight of each expert for each risk factor; and a final subjective weight calculation unit, used to calculate the average of the subjective weights corresponding to all experts for each risk factor, and use the calculation result as the final subjective weight corresponding to that risk factor.

[0010] The improved FMEA apparatus considering the relationship between uncertainty and failure modes provided by this invention may also have the following feature: When solving for the subjective weights of expert e1, the expression of the corresponding SVTraN-FUCOM programming model is: minχ, st

[0011]

[0012] In the formula Let SVTraNN data be the score given by the i-th expert for the importance of the j-th risk factor. The SVTraNN data represents the subjective weights of the i-th expert on the j-th risk factor. Let be the subjective weight of the i-th expert for the j-th risk factor.

[0013] The improved FMEA device considering the relationship between uncertainty and failure modes provided by this invention may also have the following features: the objective weight calculation module includes: a score matrix generation unit, used to calculate and normalize the SVTraNN score of the SVTraNN decision matrix to obtain the SVTraNN normalized score matrix; a parameter calculation unit, used to calculate the standard deviation of each risk factor and the Pearson correlation coefficient between each pair of risk factors based on the SVTraNN normalized score matrix; and an objective weight solving unit, used to solve the objective weight corresponding to each risk factor by constructing a target programming model based on the standard deviation and the Pearson correlation coefficient.

[0014] The improved FMEA apparatus provided by this invention, which considers the relationship between uncertainty and failure modes, may also have the following feature: wherein the calculation expression for the SVTraNN score is: In the formula <(a f ,b f ,c f ,d f );T f ,I f ,F f > represents the expression corresponding to element f in the SVTraNN decision matrix, and ξ(f) is the result of calculating the SVTraNN score for element f. The expression for constructing the difference-based goal programming model is: minλ b +0.5λ c st ωo j >0, j=1,2,3. In the formula, ωo j Let σ be the objective weight of the j-th risk factor. j Let r be the standard deviation of all normalized scores for the j-th risk factor. jl Let be the Pearson correlation coefficient between the j-th risk factor and all normalized scores of the l-th risk factor.

[0015] The improved FMEA device considering the relationship between uncertainty and failure modes provided by this invention may also have the following features: the comprehensive score calculation module includes: a normalization unit, used to normalize the SVTraNN decision matrix using the RAFSI normalization method to obtain the SVTraNN normalized decision matrix; a weighting unit, used to obtain a weighted normalized matrix based on the comprehensive weights and the SVTraNN normalized decision matrix; a negative ideal solution unit, used to obtain the negative ideal solution for each risk factor based on the SVTraNN normalized decision matrix and the corresponding SVTraNN score; a distance calculation unit, used to calculate the Euclidean distance and Manhattan distance corresponding to each potential AI safety risk mode based on the SVTraNN normalized decision matrix and the negative ideal solution; a correlation evaluation matrix construction unit, used to construct a correlation evaluation matrix based on all Euclidean distances and Manhattan distances; and a comprehensive score calculation unit, used to calculate the comprehensive score for each potential AI safety risk mode based on the correlation evaluation matrix.

[0016] The improved FMEA apparatus provided by this invention, which considers the relationship between uncertainty and failure modes, may also have the following feature: The expression for the weighted normalization matrix is: V m×3 =[v ij ] m×3 , In the formula V m×3 For the weighted normalized matrix, v ij For the weighted normalized matrix V m×3 The element in the i-th row and j-th column, ω j The comprehensive weight corresponding to the j-th risk factor is... Let E be the element in the i-th row and j-th column of the SVTraNN normalized decision matrix, and let E be the Euclidean distance E corresponding to the i-th potential AI security risk pattern. i The calculation expression is: In the formula and Let be the true membership degree corresponding to the i-th potential AI security risk pattern under the j-th risk factor. and Let represent the uncertain membership degree corresponding to the i-th potential AI safety risk pattern under the j-th risk factor. and Let be the pseudo-membership degree corresponding to the i-th potential AI security risk pattern under the j-th risk factor. and Let be the true membership degree corresponding to the negative ideal solution of the j-th risk factor. and Let be the uncertain membership degree corresponding to the negative ideal solution of the j-th risk factor. and Let T be the pseudo-membership degree corresponding to the negative ideal solution of the j-th risk factor, and let T be the Manhattan distance T corresponding to the i-th potential AI security risk pattern. i The calculation expression is: The expression for the relevant evaluation matrix is: RA = [o ik ] m×m o ik =(E i -E k )+Ψ(E i -E k )*(T i -T k ), In the formula, τ is a preset value, and O is the comprehensive score corresponding to the i-th potential AI security risk mode. i The calculation expression is: In the formula, m represents the total number of potential risk patterns for artificial intelligence security.

[0017] The role and effect of invention

[0018] The improved FMEA apparatus of this invention, which considers the relationship between uncertainty and failure modes, firstly transforms the scoring data using SVTraNN to account for the uncertainty and ambiguity of expert evaluation information; secondly, it calculates expert weights based on confidence levels, addressing the shortcomings of traditional FMEA that do not consider expert differences; thirdly, it solves the final subjective weights of risk factors using the SVTraN-FUCOM programming model and constructs a goal programming model based on the differences to solve the objective weights, thus solving the problems of strong subjectivity and failure to consider the differences in the impact of different risk factors on the research object in traditional FMEA methods; fourthly, it calculates the comprehensive weights of risk factors using objective weights and final subjective weights, further calculates the comprehensive score and ranks them to obtain the risk ranking result. This addresses the shortcomings of traditional FMEA methods, such as the lack of robustness and scientific rigor in RPN calculation, while also considering the relationship between risk modes. Therefore, the improved FMEA apparatus of this invention, which considers the relationship between uncertainty and failure modes, can effectively assess the risk level of risk modes. Attached Figure Description

[0019] Figure 1 This is a block diagram of the improved FMEA device in an embodiment of the present invention;

[0020] Figure 2 This is a schematic diagram illustrating the calculation results of the relevant evaluation matrix and the comprehensive score in an embodiment of the present invention;

[0021] Figure 3This is a schematic diagram of the process of generating risk ranking results by improving the FMEA device in an embodiment of the present invention;

[0022] Figure 4 This is a schematic diagram of the ISODATA clustering results in an embodiment of the present invention. Detailed Implementation

[0023] To make the technical means, creative features, objectives and effects of the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the improved FMEA device of the present invention that considers the relationship between uncertainty and failure modes.

[0024] This embodiment provides an improved FMEA device that considers the relationship between uncertainty and failure modes, hereinafter referred to as the improved FMEA device. It is used to obtain the risk ranking results and risk levels of each potential artificial intelligence safety risk mode based on the scoring data of each expert on each potential artificial intelligence safety risk mode and the score data of the importance of risk factors.

[0025] In this embodiment, the experts include a team of experts consisting of university scholars (e1), senior risk strategy analysts (e2), artificial intelligence risk assessment experts (e3), and senior artificial intelligence product managers (e4).

[0026] This embodiment contains 19 potential AI security risk patterns, as detailed in the table below:

[0027]

[0028]

[0029] The first column in the table above represents each risk module, the second column represents each potential AI security risk mode, and the third to fifth columns represent the risk causes, risk consequences, and numbers corresponding to each potential AI security risk mode.

[0030] Based on risk factors S, O, and D, the scoring criteria are established as shown in the table below:

[0031]

[0032] The first column in the table above shows the various scores, and the second to fourth columns are the descriptions of risk factor S (severity), risk factor O (occurrence), and risk factor D (detectability) under each score, respectively.

[0033] The scoring data generated by various experts for each potential AI security risk pattern are shown in the table below:

[0034]

[0035]

[0036] The first column in the table above is the number of each potential AI safety risk pattern, and the second to thirteenth columns are the scores given by each expert on risk factors S, O, and D, respectively.

[0037] The risk factor importance scoring criteria in this embodiment are shown in the table below:

[0038] score Importance 1 Almost no impact 2 Slight impact 3，4 It has a low degree of impact. 5，6 Relatively important 7，8 Significant impact 9 Plays a decisive role 10 Extremely crucial

[0039] The risk factor importance scores for each expert are shown in the table below:

[0040] Risk factors <![CDATA[e1]]> <![CDATA[e2]]> <![CDATA[e3]]> <![CDATA[e4]]> S 9 8 9 8 O 8 7 7 6 D 6 6 5 5

[0041] Figure 1 This is a block diagram of an improved FMEA device in an embodiment of the present invention.

[0042] like Figure 1 As shown, the improved FMEA device 100 includes an SVTraNN score conversion module 11, an expert weight calculation module 12, an expert evaluation summary module 13, a subjective weight calculation module 14, an objective weight calculation module 15, a comprehensive weight calculation module 16, a comprehensive score calculation module 17, a risk ranking module 18, a risk level module 19, and a control module 20 that controls the operation of the above modules.

[0043] The SVTraNN rating conversion module 11 stores a preset SVTraNN language variable conversion table, which is used to obtain the SVTraNN evaluation matrix corresponding to each expert based on the SVTraNN language variable conversion table and the rating data.

[0044] Wherein, the SVTraNN evaluation matrix H corresponds to expert k. k The expression is:

[0045]

[0046] In the formula, columns 1, 2, and 3 correspond to the scores of risk factors S, O, and D, respectively.

[0047] The SVTraNN language variable transformation table for this implementation is shown below:

[0048] score SVTraNNs Score 1 <(0,0.05,0.05,0.1)；0.1,0.9,0.9> 0.0440 2 <(0.1,0.15,0.15,0.2)；0.1,0.8,0.75> 0.1256 3 <(0.2,0.25,0.25,0.3)；0.25,0.7,0.6> 0.2917 4 <(0.3,0.35,0.35,0.4)；0.4,0.6,0.5> 0.3789 5 <(0.4,0.45,0.45,0.5)；0.5,0.5,0.4> 0.4766 6 <(0.5,0.55,0.55,0.6)；0.6,0.4,0.3> 0.5755 7 <(0.6,0.65,0.65,0.7)；0.7,0.3,0.2> 0.6750 8 <(0.7,0.75,0.75,0.8)；0.8,0.2,0.1> 0.7748 9 <(0.8,0.85,0.85,0.9)；0.9,0.1,0.1> 0.8747 10 <(0.9,0.95,0.95,1)；1,0.1,0> 0.9746

[0049] The first column in the table above represents the rating, and the second column represents the SVTraNNs data corresponding to the rating. The third column shows the scores corresponding to the SVTraNNs data.

[0050] The expert weight calculation module 12 is used to calculate the expert weight of each expert for each risk factor based on the SVTraNNs score of the risk factor in all SVTraNNs evaluation matrices.

[0051] The expert weight calculation module 12 includes an average similarity calculation unit 121, a confidence level calculation unit 122, and an expert weight calculation unit 123.

[0052] The average similarity calculation unit 121 is used to calculate the average similarity corresponding to each risk factor based on the SVTraNN evaluation matrix of the pairwise experts.

[0053] Among them, the average similarity AS between expert k and expert l regarding risk factor S. S (e k ,e l The calculation expression for ) is:

[0054]

[0055] f = <(a f ,b f ,c f ,d f );T f ,I f ,F f >,

[0056] g=<(a g ,b g ,c g ,d g );T g ,I g ,F g >,

[0057]

[0058] In the formula, m represents the total number of potential AI security risk patterns, and f and g are two SVTraNNs, respectively. Let SVTraNN be the SVTraNN corresponding to the i-th potential AI security risk pattern in the SVTraNN evaluation matrix corresponding to expert k. Let x be the SVTraNN corresponding to the i-th potential AI security risk pattern in the SVTraNN evaluation matrix corresponding to expert l. fT x gT y fT y gT Corresponding to true membership degree, x fI x gI y fI y gI Corresponding to uncertain membership degree, xfF x gF y fF y gF Corresponding to pseudo-membership degree.

[0059] In this embodiment, x is used. fT x gT y fT y gT For example, the formula is defined as:

[0060]

[0061] In this embodiment, the average pairwise similarity between experts under risk factor S is shown in the table below:

[0062]

[0063]

[0064] The confidence level calculation unit 122 is used to calculate the confidence level of each expert for each risk factor based on the average similarity between each pair of experts.

[0065] Among them, the confidence level CL corresponding to expert k under risk factor S. S (e k The calculation expression for ) is:

[0066]

[0067] In the formula, t represents the total number of experts.

[0068] The confidence levels of each expert under risk factor S in this embodiment are shown in the table below:

[0069] Confidence level <![CDATA[e1]]> <![CDATA[e2]]> <![CDATA[e3]]> <![CDATA[e4]]> <![CDATA[CL S (and k )]]> 0.8463 0.7919 0.8599 0.8532

[0070] The first row in the table above represents each expert, and the second row represents the confidence level for each expert.

[0071] The expert weight calculation unit 123 is used to calculate the expert weight corresponding to each expert for each risk factor based on the confidence level of all experts.

[0072] Among them, the expert weight ws corresponding to expert k under risk factor S. k The calculation expression is:

[0073]

[0074] In this embodiment, the expert weights for each risk factor are shown in the table below:

[0075] Expert Weight <![CDATA[e1]]> <![CDATA[e2]]> <![CDATA[e3]]> <![CDATA[e4]]> <![CDATA[ws k ]]> 0.2525 0.2363 0.2566 0.2546 <![CDATA[wo k ]]> 0.2533 0.2411 0.2543 0.2513 <![CDATA[wd k ]]> 0.2402 0.2515 0.2567 0.2516

[0076] The first row in the table above represents each expert, and the second to fourth rows represent the expert weights corresponding to each expert under risk factor S, risk factor O, and risk factor D, respectively.

[0077] The expert evaluation summary module 13 is used to calculate the SVTraNN decision matrix based on the evaluation matrices of all SVTraNNs and the expert weights.

[0078] The expression for the SVTraNN decision matrix is ​​as follows:

[0079]

[0080]

[0081] In the formula, i = 1, ..., m, where m is the total number of potential AI security risk patterns. and These represent the scores of the i-th potential AI safety risk pattern in the SVTraNN evaluation matrix corresponding to the k-th expert, under risk factors S, O, and D, respectively, and ws k wo k and wd k These represent the expert weights for the k-th expert under risk factors S, O, and D, respectively.

[0082] The calculation results of the SVTraNN decision matrix in this embodiment are shown below:

[0083]

[0084] The subjective weight calculation module 14 stores the SVTraNN linguistic variable transformation table and the SVTraN-FUCOM programming model, which is used to calculate the final subjective weight of each risk factor based on the SVTraNN linguistic variable transformation table, the SVTraN-FUCOM programming model and the importance score data of all risk factors.

[0085] The subjective weight calculation module 14 includes an SVTraNN transformation unit 141, a subjective weight solving unit 142, and a final subjective weight calculation unit 143.

[0086] The SVTraNN transformation unit 141 is used to transform the risk factor importance score data according to the SVTraNN linguistic variable transformation table to obtain the SVTraNN data corresponding to each expert. For example, the scores of expert e1 for risk factors S, O, and D are transformed into <(0.8,0.85,0.85,0.9);0.9,0.1,0.1>, <(0.7,0.75,0.75,0.8);0.8,0.2,0.1>, and <(0.5,0.55,0.55,0.6);0.6,0.4,0.3>, respectively.

[0087] The subjective weight solving unit 142 is used to solve the SVTraN-FUCOM planning model based on the SVTraNN data corresponding to each expert, and obtain the subjective weights of each expert for each risk factor.

[0088] When solving for the subjective weights of expert e1, the expression for the corresponding SVTraN-FUCOM programming model is:

[0089] minχ

[0090] st

[0091]

[0092]

[0093] In the formula Let SVTraNN data be the score given by the i-th expert for the importance of the j-th risk factor. The SVTraNN data represents the subjective weights of the i-th expert on the j-th risk factor. Let be the subjective weight of the i-th expert for the j-th risk factor.

[0094] In this embodiment, the subjective weights of experts e1, e2, e3, and e4 are calculated to be {0.3916, 0.3479, 0.2605}, {0.3845, 0.3333, 0.2821}, {0.4359, 0.3333, 0.2308}, and {0.4067, 0.3175, 0.2758}, respectively.

[0095] The final subjective weight calculation unit 143 is used to calculate the average of the subjective weights corresponding to all experts for each risk factor, and use the calculation result as the final subjective weight corresponding to the risk factor.

[0096] In this embodiment, the final subjective weights calculated are {0.4047, 0.3330, 0.2623}.

[0097] The objective weight calculation module 15 stores the difference-based goal programming model, which is used to calculate the objective weights corresponding to each risk factor based on the SVTraNN decision matrix and the difference-based goal programming model.

[0098] The objective weight calculation module 15 includes a score matrix generation unit 151, a parameter calculation unit 152, and an objective weight solution unit 153.

[0099] The score matrix generation unit 151 is used to calculate and normalize the SVTraNN score of the SVTraNN decision matrix to obtain the SVTraNN normalized score matrix.

[0100] The calculation expression for the SVTraNN normalized score is as follows:

[0101]

[0102] In the formula <(a f ,b f ,c f ,d f );T f ,I f ,F f > represents the expression corresponding to element f in SVTraNN, and ξ(f) is the result obtained by calculating the SVTraNN score of element f.

[0103] The SVTraNN normalized score matrix calculated in this embodiment is shown in the table below:

[0104]

[0105] The first column in the table above shows the potential AI security risk patterns, and the second to fourth columns show the SVTraNN score normalization results for each potential AI security risk pattern under risk factors S, O, and D, respectively.

[0106] The parameter calculation unit 152 is used to calculate the standard deviation of each risk factor and the Pearson correlation coefficient between each pair of risk factors based on the SVTraNN normalized score matrix.

[0107] The objective weight solving unit 153 is used to solve the objective weights corresponding to each risk factor by constructing a target programming model based on the standard deviation and Pearson correlation coefficient.

[0108] The expression for the difference-based goal programming model is as follows:

[0109] minλ b +0.5λ c

[0110] st

[0111]

[0112] ωo j >0, j=1,2,3.

[0113] In the formula ωo j Let σ be the objective weight of the j-th risk factor. j Let r be the standard deviation of all normalized scores for the j-th risk factor. jl Let be the Pearson correlation coefficient between the j-th risk factor and all normalized scores of the l-th risk factor. In this embodiment, the standard deviation {σ} j |j=1,2,3}={0.2563,0.2381,0.1892}, Pearson correlation coefficient r 12 =0.3569, r 13 =0.7691, r 23 =0.2639.

[0114] In this embodiment, the calculated objective weight is {ωo}. j |j=1,2,3}={0.3404,0.3749,0.2846}.

[0115] The comprehensive weight calculation module 16 is used to calculate the comprehensive weight corresponding to each risk factor based on all final subjective and objective weights. The calculation formula is as follows:

[0116]

[0117] In this embodiment, the calculated comprehensive weight is {ω} j |j=1,2,3}={0.4085,0.3702,0.2213}.

[0118] The comprehensive score calculation module 17 is used to calculate the comprehensive score of each artificial intelligence security potential risk pattern based on the SVTraN-CODAS method, the SVTraNN decision matrix and all comprehensive weights.

[0119] The comprehensive score calculation module 17 includes a normalization unit 171, a weighting unit 172, a negative ideal solution unit 173, a distance calculation unit 174, a relevant evaluation matrix construction unit 175, and a comprehensive score calculation unit 176.

[0120] The specific implementation method of the SVTraN-CODAS method in this embodiment is as follows:

[0121] Normalization unit 171 is used to normalize the SVTraNN decision matrix using the RAFSI normalization method to obtain the SVTraNN normalized decision matrix.

[0122] In this embodiment, the SVTraNN score matrix is ​​used as a metric to determine the ideal solution for each risk factor. and non-ideal solutions The result is as follows:

[0123]

[0124] Secondly, through functions H m×3 All elements are mapped to the standard interval [h1, h 2c Let h1 = <(0.49, 0.64, 0.69, 0.82); 0.2, 0.8, 0.75>, h 2c =<(1.79,2.06,2.14,2.27);0.7,0.3,0.2>, ξ(h1)=0.8410, ξ(h 2c )=7.5089,ξ(h 2c The result is approximately 9 times ξ(h1), expressed as G. m×3 =[g ij ] m×3 .

[0125] Next, calculate h1 and h 2c The harmonic mean HM and the arithmetic mean AM,

[0126] Finally, obtain the SVTraNN normalized decision matrix. For the largest metric, For the smallest metric, In FMEA, all three risk factors are the largest indicators.

[0127] The weighting unit 172 is used to obtain the weighted normalized matrix based on the comprehensive weights and the SVTraNN normalized decision matrix.

[0128] The expression for the weighted normalized matrix is:

[0129] V m×3 =[v ij ] m×3 ,

[0130]

[0131] In the formula V m×3 For the weighted normalized matrix, vij For the weighted normalized matrix V m×3 The element in the i-th row and j-th column, ω j The comprehensive weight corresponding to the j-th risk factor is... Let be the element in the i-th row and j-th column of the SVTraNN normalized decision matrix.

[0132] The weighted normalized matrix calculated in this embodiment is:

[0133]

[0134] The negative ideal solution unit 173 is used to obtain the negative ideal solution for each risk factor based on the SVTraNN normalized decision matrix and the corresponding SVTraNN score. In this embodiment, the negative ideal solution consists of the SVTraNN data corresponding to the lowest score in each column, and the negative ideal solution ns = {<(0.09,0.10,0.10,0.11); 0.11,0.89,0.88>,<(0.08,0.09,0.09,0.10); 0.10,0.90,0.89>,<(0.05,0.05,0.05,0.06); 0.06,0.94,0.93>}.

[0135] The distance calculation unit 174 is used to calculate the Euclidean distance and Manhattan distance corresponding to each potential AI security risk pattern based on the SVTraNN normalized decision matrix and the negative ideal solution.

[0136] Among them, the Euclidean distance E corresponding to the i-th potential AI security risk pattern i The calculation expression is:

[0137]

[0138] In the formula and Let be the true membership degree corresponding to the i-th potential AI security risk pattern under the j-th risk factor. and Let represent the uncertain membership degree corresponding to the i-th potential AI safety risk pattern under the j-th risk factor. and Let be the pseudo-membership degree corresponding to the i-th potential AI security risk pattern under the j-th risk factor. and Let be the true membership degree corresponding to the negative ideal solution of the j-th risk factor. and Let be the uncertain membership degree corresponding to the negative ideal solution of the j-th risk factor. and Let be the pseudo-membership degree corresponding to the negative ideal solution of the j-th risk factor.

[0139] The Manhattan distance T corresponding to the i-th potential AI security risk pattern i The calculation expression is:

[0140]

[0141] The calculated Euclidean and Manhattan distances for various potential AI security risk patterns are shown in the table below:

[0142]

[0143]

[0144] The correlation evaluation matrix construction unit 175 is used to construct the correlation evaluation matrix based on all Euclidean distances and Manhattan distances.

[0145] The expression for the relevant evaluation matrix is ​​as follows:

[0146] RA = [o ik ] m×m ,

[0147] o ik =(E i -E k )+Ψ(E i -E k )*(T i -T k ),

[0148]

[0149] In the formula, τ is a preset value. In this embodiment, τ is taken as 0.03.

[0150] The comprehensive score calculation unit 176 is used to calculate the comprehensive score of each potential AI security risk mode based on the relevant evaluation matrix.

[0151] Among them, the comprehensive score O corresponding to the i-th potential AI security risk pattern is i The calculation expression is:

[0152]

[0153] In the formula, m represents the total number of potential risk patterns for artificial intelligence security.

[0154] Figure 2 This is a schematic diagram illustrating the calculation results of the relevant evaluation matrix and the comprehensive score in an embodiment of the present invention.

[0155] like Figure 2As shown, rows 2 to 20 of the table represent the evaluation values ​​of AI security potential risk models FM1 to FM19 and other models, respectively. Row 21 of the table represents the comprehensive score of each AI security potential risk model FM1 to FM19.

[0156] The risk ranking module 18 is used to obtain the risk ranking results of each potential AI security risk mode based on all comprehensive scores. In this embodiment, the comprehensive scores are ranked from largest to smallest, and the higher the risk ranking result, the higher the risk of the corresponding potential AI security risk mode.

[0157] The risk ranking results in this embodiment are shown in the table below:

[0158] serial number Risk Model <![CDATA[O i ]]> Sort <![CDATA[FM1]]> Over-collection of training data -1.771 17 <![CDATA[FM2]]> Training data collection is non-compliant -1.277 14 <![CDATA[FM3]]> Training data poisoning -1.116 12 <![CDATA[FM4]]> Member reasoning attack -1.552 15 <![CDATA[FM5]]> Training data leakage 1.253 9 <![CDATA[FM6]]> Training data fabrication -1.051 10 <![CDATA[FM7]]> Algorithm design flaws 1.559 8 <![CDATA[FM8]]> Algorithm backdoor attack 2.319 1 <![CDATA[FM9]]> Weak robustness of the algorithm 1.946 4 <![CDATA[FM 10 ]]> Algorithm uninterpretability -1.101 11 <![CDATA[FM 11 ]]> Algorithmic bias and discrimination 1.604 7 <![CDATA[FM 12 ]]> Adversarial sample attacks 1.842 5 <![CDATA[FM 13 ]]> Model theft attack 2.179 2 <![CDATA[FM 14 ]]> Model Reverse Attack -1.120 13 <![CDATA[FM 15 ]]> Model feedback misleading -2.267 18 <![CDATA[FM 16 ]]> Software framework -1.683 16 <![CDATA[FM 17 ]]> Computing support facilities -3.338 19 <![CDATA[FM 18 ]]> Generative model abuse 1.965 3 <![CDATA[FM 19 ]]> Social engineering attack 1.611 6

[0159] The first column in the table above represents the ID of each potential AI security risk pattern, the second column represents the content of each potential AI security risk pattern, and the third and fourth columns represent the overall score and corresponding ranking number of each potential AI security risk pattern, respectively. It is evident that "algorithm backdoor attacks," "model theft attacks," "abuse of generative models," "weak algorithm robustness," and "adversarial example attacks" are relatively more likely to pose a high level of security risk to human society.

[0160] The process of obtaining risk ranking results using the improved FMEA device 100 is described below with reference to the accompanying drawings.

[0161] Figure 3 This is a schematic diagram of the process of generating risk ranking results by improving the FMEA device in an embodiment of the present invention.

[0162] like Figure 3 As shown, the risk ranking results are obtained using the improved FMEA device 100, including the following steps:

[0163] Step S1: Using the SVTraNN rating conversion module 11, the SVTraNN evaluation matrix corresponding to each expert is obtained based on the SVTraNN language variable conversion table and the rating data.

[0164] Step S2: Using the expert weight calculation module 12, for each risk factor, the expert weight of each expert for that risk factor is calculated based on the SVTraNNs score corresponding to that risk factor in all SVTraNNs evaluation matrices.

[0165] Step S3: The expert evaluation summary module 13 calculates the SVTraNN decision matrix based on the evaluation matrices of all SVTraNNs and the expert weights.

[0166] Step S4: The subjective weight calculation module 14 calculates the final subjective weight of each risk factor based on the SVTraNN language variable transformation table, the SVTraN-FUCOM programming model, and the importance score data of all risk factors.

[0167] Step S5: The objective weight calculation module 15 constructs a target programming model based on the SVTraNN decision matrix and differences, and calculates the objective weights corresponding to each risk factor.

[0168] Step S6: The comprehensive weight calculation module 16 calculates the comprehensive weight corresponding to each risk factor based on all final subjective and objective weights.

[0169] Step S7: The comprehensive score calculation module 17 calculates the comprehensive score of each artificial intelligence security potential risk mode based on the SVTraN-CODAS method, the SVTraNN decision matrix and all comprehensive weights.

[0170] Step S8: The risk ranking module 18 uses all comprehensive scores to obtain the risk ranking results of each potential AI security risk mode.

[0171] The risk level module 19 is used to perform clustering and hierarchical analysis using ISODATA based on the comprehensive score, scoring data, and comprehensive weight to obtain the risk level of each potential AI safety risk pattern.

[0172] In this embodiment, the risk level module 19 first uses the comprehensive score as the first feature value; secondly, it averages and synthesizes the scoring data to obtain the data corresponding to each risk factor. and Next, WARPN is calculated based on the comprehensive weight. i As the second eigenvalue, its calculation expression is: Then, the first and second eigenvalues ​​are normalized respectively, and their calculation expressions are as follows: This leads to the generation of ISODATA clustering input, namely two-dimensional feature data.

[0173] The data obtained from the above process are shown in the table below:

[0174]

[0175] The first column in the table above represents the numbers of the potential AI security risk patterns, and the second to seventh columns represent the corresponding AI security risk patterns, respectively. WARPN i , and

[0176] When performing clustering in ISODATA, the Risk Level Module 19 uses the silhouette coefficient, CH score, and Davidson-Bourdin Index (DBI) as metrics to obtain the best clustering results.

[0177] Figure 4 This is a schematic diagram of the ISODATA clustering results in an embodiment of the present invention.

[0178] like Figure 4 As shown, the horizontal axis is The value of the ordinate is The results showed a silhouette coefficient of 0.7563, a CH score of 209.1, and a DBI of 0.1868, indicating that the clustering results were significantly superior. Ultimately, three risk levels were obtained: "high risk," "medium risk," and "low risk."

[0179] The "high-risk" modes include algorithm backdoor attacks (FM8) and model theft attacks (FM). 13 Generative models abuse FM 18 Weak robustness of the algorithm FM9, adversarial example attacks FM 12 Social engineering attacks FM 19 Algorithm bias and discrimination FM 11 Algorithm design flaw FM7, training data leak FM5.

[0180] The "medium risk" mode includes training data forgery FM6 and algorithm uninterpretability FM. 10 Training data poisoning FM3, model reverse engineering attack FM 14 1. Non-compliant training data collection (FM2); 2. Member inference attack (FM4); 3. Software framework (FM) 16 Over-collection of training data (FM1), misleading model feedback (FM) 15 .

[0181] The "low-risk" model includes computing power support facilities (FM). 17 .

[0182] The control module 20 stores the control program that controls the operation of each module.

[0183] The role and effect of the embodiments

[0184] According to the improved FMEA device considering the relationship between uncertainty and failure modes involved in this embodiment, firstly, the scoring data is transformed using SVTraNN to increase the consideration of the uncertainty and ambiguity of expert evaluation information; secondly, expert weights are calculated through confidence levels, solving the shortcoming of traditional FMEA that does not consider the differences among experts; thirdly, the final subjective weights are solved using the SVTraN-FUCOM programming model, and objective weights are solved by constructing a goal programming model based on the differences, thus solving the problem of strong subjectivity and failure to consider the differences in the impact of different risk factors on the research object in traditional FMEA methods; fourthly, a comprehensive weight is calculated using objective weights and final subjective weights, and then a comprehensive score is calculated and ranked using the SVTraN-CODAS method to obtain the risk ranking result. This addresses the shortcomings of traditional FMEA methods in the lack of robustness and scientific rigor in RPN calculation, while also considering the relationship between risk modes. In summary, this method can effectively assess the risk level of risk modes.

[0185] In addition, this embodiment also obtains the risk level of the risk pattern through ISODATA clustering using the risk level module, so that managers can implement corresponding solutions for risk patterns with different risk levels.

[0186] Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An improved FMEA device considering the uncertainty and failure mode correlation, for obtaining the risk ranking results and risk levels of each artificial intelligence security potential risk mode according to the scoring data of each expert on each artificial intelligence security potential risk mode and the risk factor importance scoring data, characterized in that, The device comprises: an SVTraNN score conversion module storing a preset SVTraNN language variable conversion table, configured to obtain an SVTraNN evaluation matrix corresponding to each expert according to the SVTraNN language variable conversion table and the score data; an expert weight calculation module configured to calculate an expert weight of each expert for each risk factor according to SVTraNN scores corresponding to the risk factor in all the SVTraNN evaluation matrices; an expert evaluation summary module configured to calculate an SVTraNN decision matrix according to all the SVTraNN evaluation matrices and the expert weights; a subjective weight calculation module storing an SVTraNN language variable conversion table and an SVTraN-FUCOM planning model, configured to calculate a final subjective weight corresponding to each risk factor according to the SVTraNN language variable conversion table, the SVTraN-FUCOM planning model, and all the risk factor importance score data; an objective weight calculation module storing a difference construction objective planning model, configured to calculate an objective weight corresponding to each risk factor according to the SVTraNN decision matrix and the difference construction objective planning model; a comprehensive weight calculation module configured to calculate a comprehensive weight corresponding to each risk factor according to all the final subjective weights and the objective weights; a comprehensive score calculation module configured to calculate a comprehensive score of each artificial intelligence security potential risk mode according to an SVTraN-CODAS method, the SVTraNN decision matrix, and all the comprehensive weights; a risk ranking module configured to obtain the risk ranking result of each artificial intelligence security potential risk mode according to all the comprehensive scores; a risk level module configured to obtain the risk level of each artificial intelligence security potential risk mode by clustering and layering according to the comprehensive scores, the score data, and the comprehensive weights, wherein the earlier the risk ranking result is, the higher the risk of the corresponding artificial intelligence security potential risk mode is, the expert weight calculation module comprises: an average similarity calculation unit configured to calculate an average similarity corresponding to each risk factor according to the SVTraNN evaluation matrix of each pair of experts; a confidence level calculation unit configured to calculate a confidence level corresponding to each expert according to the average similarity between each pair of experts for each risk factor; an expert weight calculation unit configured to calculate the expert weight corresponding to each expert for each risk factor according to the confidence level of all the experts, expert and expert between the expert The calculation expression is: ， ， ， ， ， wherein is the total number of potential risks of artificial intelligence security, and are two SVTraNNs respectively, is an expert is the first column of the corresponding SVTraNN evaluation matrix is the SVTraNN corresponding to the potential risk of artificial intelligence security, is an expert is the first column of the corresponding SVTraNN evaluation matrix is the SVTraNN corresponding to the potential risk of artificial intelligence security, , , , corresponds to the true membership degree, , , , corresponds to the uncertain membership degree, , , , corresponds to the false membership degree, Risk factor S under expert Corresponding confidence level The calculation expression is: ， In the formula is the total number of experts, Risk factor S under expert Corresponding expert weight The calculation expression is: 。 2. The improved FMEA device considering the correlation between uncertainty and failure modes according to claim 1, wherein: wherein expert Corresponding SVTraNN evaluation matrix The expression is: ， in the formula, columns 1, 2, and 3 correspond to the scores of risk factors S, O, and D, respectively, the expression of the SVTraNN decision matrix is: ， ， ， ， In the formula , is the total number of potential risk patterns of artificial intelligence security, , and are the scores of the first expert in the SVTraNN evaluation matrix under the risk factors S, O and D respectively, , , and are the scores of the first expert under the risk factors S, O and D respectively.

3. The improved FMEA device considering uncertainty and failure mode interrelationship of claim 1, wherein: the subjective weight calculation module comprises: ​ An SVTraNN transformation unit is configured to transform the risk factor importance score data according to the SVTraNN language variable transformation table to obtain SVTraNN data corresponding to each expert; An objective weight solving unit is configured to solve the SVTraN-FUCOM planning model according to the SVTraNN data corresponding to each expert, respectively, to obtain the subjective weight of each risk factor for each expert. An ultimate subjective weight calculating unit is configured to calculate the average of the subjective weights of all experts for each risk factor, and take the calculation result as the ultimate subjective weight of the risk factor.

4. The improved FMEA device considering the correlation between uncertainty and failure modes according to claim 3, characterized in that: wherein To the experts When solving the subjective weight, the expression of the SVTraN-FUCOM planning model is as follows: ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， ， In the formula For the first The expert on the first SVTraNN data corresponding to the importance scores of each risk factor. For the first The expert on the first SVTraNN data corresponding to the subjective weights of each risk factor. For the first The expert on the first Subjective weights of each risk factor.

5. The improved FMEA device considering uncertainty and failure mode interrelationship of claim 1, characterized in that: wherein, the objective weight calculating module comprises: A score matrix generating unit is configured to perform SVTraNN score calculation and normalization on the SVTraNN decision matrix to obtain an SVTraNN normalized score matrix; A parameter calculating unit is configured to calculate the standard deviation corresponding to each risk factor and the Pearson correlation coefficient between two risk factors according to the SVTraNN normalized score matrix; An objective weight solving unit is configured to solve the difference construction objective programming model according to the standard deviation and the Pearson correlation coefficient to obtain the objective weight corresponding to each risk factor.

6. The improved FMEA device considering the correlation between uncertainty and failure modes according to claim 5, characterized in that: wherein, the calculation expression of the SVTraNN score calculation is: ， wherein is the SVTraNN element the corresponding expression, is the element the result of the SVTraNN score calculation, the expression of the difference construction objective programming model is: ， ， ， ， ， ， ， ， ， wherein is the objective weight of the th risk factor, is the standard deviation of all normalized scores of the th risk factor, is the Pearson correlation coefficient between all normalized scores of the th risk factor and the th risk factor.

7. The improved FMEA device considering uncertainty and failure mode interrelationship of claim 1, characterized in that: wherein, the comprehensive score calculating module comprises: A normalization unit is configured to normalize the SVTraNN decision matrix by the normalization method of RAFSI to obtain an SVTraNN normalized decision matrix; A weighting unit is configured to obtain a weighted normalized matrix according to the comprehensive weight and the SVTraNN normalized decision matrix; A negative ideal solution unit is configured to obtain the negative ideal solution of each risk factor according to the SVTraNN normalized decision matrix and the corresponding SVTraNN score; A distance calculating unit is configured to calculate the Euclidean distance and the Manhattan distance corresponding to each artificial intelligence security potential risk mode according to the SVTraNN normalized decision matrix and the negative ideal solution; A correlation evaluation matrix constructing unit is configured to construct a correlation evaluation matrix according to all the Euclidean distances and the Manhattan distances; A comprehensive score calculating unit is configured to calculate the comprehensive score of each artificial intelligence security potential risk mode according to the correlation evaluation matrix.

8. The improved FMEA device considering the correlation between uncertainty and failure modes according to claim 7, characterized in that: wherein the expression of the weighted normalized matrix is: ， ， In the formula For a weighted normalized matrix, For weighted normalized matrix The Middle Line number Column elements, For the first The comprehensive weights corresponding to each risk factor The SVTraNN normalized decision matrix is ​​the th Line number Column elements, The first Euclidean distance corresponding to the potential risk pattern of personal artificial intelligence security The calculation expression is: ， In the formula and For the first Under the risk factor, the first True membership degree corresponding to each potential AI security risk pattern and For the first Under the risk factor, the first Uncertain membership degree corresponding to each potential AI security risk pattern and For the first Under the risk factor, the first The pseudo-membership degree corresponding to the potential security risk patterns of artificial intelligence and For the first The true membership degree corresponding to the negative ideal solution of each risk factor and For the first The uncertain membership degree corresponding to the negative ideal solution of each risk factor. and For the first The pseudo-membership degree corresponding to the negative ideal solution of each risk factor No. Manhattan distance corresponding to a potential AI security risk pattern The calculation expression is: ， the expression of the correlation evaluation matrix is: ， ， ， In the formula is a preset value, The first The overall score corresponding to the potential risk pattern of personal artificial intelligence security The calculation expression is: ， In the formula is the total number of potential risk patterns for artificial intelligence safety.

Images