An improved FMEA method for enhancing objectivity and improving accuracy

Abstract

The application provides an improved FMEA method which enhances objectivity and improves accuracy, improves the traditional FMEA method based on triangular fuzzy numbers, grey correlation degree and Gaussian mixture model clustering analysis. The method first converts the very system into triangular fuzzy numbers by using fuzzy language set to score the fault mode, solves the disadvantage that the risk factor is difficult to be evaluated. Secondly, the risk factor weight is calculated by combining the analytic hierarchy process and entropy weight method, solves the problem that the traditional FMEA method is subjective and does not consider the relative importance of risk factors. Thirdly, the grey correlation degree of each fault mode is calculated to sort it, avoiding the defect that the RPN calculation method has low accuracy. Finally, according to the RPN value and the grey correlation degree of the fault mode, the clustering method based on the Gaussian mixture model is used for clustering analysis of the fault mode, overcoming the shortcomings of the K-means algorithm, realizing adaptive clustering by using the expectation maximization algorithm, improving the effectiveness and efficiency of clustering, so as to improve the efficiency of subsequent prevention and monitoring of the fault mode.

Description

Technical Field

[0001] This invention belongs to the field of reliability analysis methods and relates to an improved FMEA method that enhances objectivity and improves accuracy. Specifically, it uses triangular fuzzy numbers, grey relational degree and Gaussian mixture model to improve the traditional FMEA. Background Technology

[0002] Failure Mode and Effects Analysis (FMEA), initially developed as a formal design methodology in the aerospace industry, has proven to be a powerful tool for assessing potential failure modes and preventing their occurrence. As an analytical technique, FMEA can define, identify, and eliminate failures from systems, designs, processes, and services before they reach the customer. The main objectives of FMEA are to identify potential failure modes, assess the causes and effects of different failure modes, and determine the likelihood of eliminating or reducing these failure modes. The results of FMEA analysis can help personnel identify and correct failure modes that adversely affect the system, and improve system performance during the design and production phases.

[0003] FMEA, an important technique for identifying and eliminating known or potential failures to enhance the reliability and safety of complex systems, aims to inform risk management decisions. When implementing FMEA, a cross-functional team should first be established, and then methods such as brainstorming should be used to identify all possible potential failure modes of the system. Next, these failure modes are analyzed, considering three risk factors: occurrence (O), severity (S), and detectability (D). The ultimate goal of FMEA is to prioritize the failure modes of a product or system so that limited resources can be allocated to the most critical risk items. Its principle is as follows: Figure 4 As shown.

[0004] However, the traditional FMEA method of obtaining RPN by multiplying S, O, and D does not take into account the relative importance of S, O, and D in real-world cases. At the same time, these three risk factors are difficult to assess accurately, and the formula for obtaining RPN by simply multiplying the three risk factors has low accuracy.

[0005] To address the problem that many concepts and variables involved in practical problems cannot be precisely measured (such as beauty or ugliness, height, etc.), the concept of fuzzy sets was first proposed in 1965. Fuzzy sets use mathematical methods to represent these fuzzy concepts, thereby enabling solutions that are more consistent with reality.

[0006] To address the problem of grey systems where some internal information is known but others is unknown, Professor Deng Julong proposed Grey Theory (GT), providing a new approach to solving related problems. Within Grey Theory, a novel multi-factor analysis method—grey relational analysis—emerged. Grey relational analysis describes the strength of relationships between objects by comparing the degree of correlation (grey relational degree) between sample data and reference data. It is a method for quantitative analysis and comparison of systems. Furthermore, this method possesses characteristics such as comprehensiveness, asymmetry, non-uniqueness, and dynamism, and is now widely used in various systems analyses.

[0007] Cluster analysis refers to the process of grouping objects with similar characteristics into groups according to certain rules. Since its inception, cluster analysis has undergone decades of development and is now widely used in fields such as data mining, speech recognition, and machine vision. Currently, the most basic and widely used clustering algorithm is the K-means clustering algorithm. This algorithm first defines K cluster centers, then finds the locations of these K cluster centers such that the sum of the squared distances between each data point and its nearest cluster center is minimized. However, this method has limitations, such as being only applicable to datasets with convex clustering results. Summary of the Invention

[0008] To address the technical problems of traditional FMEA methods, such as high subjectivity, low accuracy, failure to consider the relative importance of risk factors, and difficulty in accurately assessing risk factors, this invention provides an improved FMEA method based on triangular fuzzy numbers, grey relational analysis, and Gaussian mixture model clustering analysis. The technical solution adopted in this invention is as follows:

[0009] This invention provides an improved FMEA method that enhances objectivity and accuracy, characterized by the following steps: Step 1, assembling an FMEA team based on analysis requirements and obtaining the operational mode of the analysis object; Step 2, determining n potential failure modes of the analysis object based on its operational mode: FM1, FM2, ..., FM n Step 3: Each member of the FMEA team evaluates and scores the three risk factors (S, O, D) for each failure mode based on a predetermined scoring standard. The average score of the team members is then taken to obtain the score of the three risk factors for the i-th failure mode, which are S... i O i D i Step 4: Calculate the risk priority number (RPN) for each failure mode, where the risk priority number for the i-th failure mode is r. i1 =S i ×O i ×D iStep 5: Establish a 7-granularity fuzzy language set, and correlate the evaluation language, member scores, and triangular fuzzy numbers to obtain the fault mode rating matrix P = (p ij ) n×3 Step 6: Determine the subjective and objective weights of the risk factors using the analytic hierarchy process (AHP) and entropy weight method, respectively. Calculate the comprehensive weight w of the risk factors based on the subjective and objective weights. j The overall weight vector is w = (w1, w2, w3); Step 7, let the reference matrix... Thus, the grey relational coefficient is obtained. The grey relational degree of the i-th fault mode is obtained as follows Step 8: The magnitude of grey relational degree is positively correlated with the degree of influence of failure modes on the analysis object. Therefore, the failure modes are sorted according to the grey relational degree to obtain the corresponding sorting results; Step 9: r i1 With r i2 As two feature values ​​describing the i-th fault mode, we obtain the description matrix R = (r it ) n×2 , representing n two-dimensional data points, and using a clustering method based on Gaussian mixture model to cluster these two-dimensional data points into K classes to obtain clustering results; Step 10, analyze the fault modes based on the sorting results and clustering results, and formulate corresponding prevention and response measures for different fault modes according to the size of the category and the gray relational degree.

[0010] The improved FMEA method provided by this invention, which enhances objectivity and improves accuracy, may also have the following technical features: In step 3, risk factors S, O, and D represent severity, occurrence, and detectability, respectively. When scoring the assessment, 1 point indicates that the failure mode has the least severe impact on the analysis object, and 10 points indicates that the failure mode has the most severe impact on the analysis object.

[0011] The improved FMEA method provided by this invention, which enhances objectivity and accuracy, also has the following technical features: the process of obtaining the fault mode scoring matrix is ​​as follows: the triangular fuzzy numbers A = (a1, a2, a3) are defuzzified using the gradient average synthesis method to obtain... The FMEA team consists of l members, and the weight of the k-th member is α. k The j-th risk factor corresponding to the i-th failure mode is u. ijk Then the team's comprehensive score for the j-th risk factor of the i-th failure mode. in Thus, the failure mode rating matrix P = (p ij ) n×3 .

[0012] The improved FMEA method provided by this invention, which enhances objectivity and accuracy, also has the following technical features, wherein step 6 includes the following sub-steps: Step 6-1, using the analytic hierarchy process (AHP) to determine the subjective weights of risk factors, i.e., obtaining the judgment matrix Z = (z... jj′ ) 3×3 The largest eigenvalue of matrix Z is λ. max Continuously adjust Z until The consistency test is then passed, thus obtaining the subjective weights of the three risk factors. Step 6-2: Determine the objective weights of risk factors using the entropy weight method, i.e., calculate the proportion of the i-th failure mode score in the j-th risk factor to the total weight of that risk factor. Thus, the information entropy redundancy of the j-th risk factor is obtained as follows: Its objective weight is Step 6-3: Adjust the subjective weights of the risk factors according to their information entropy redundancy to obtain the comprehensive weight of the j-th risk factor. The overall weight vector is w = (w1, w2, w3).

[0013] The improved FMEA method provided by this invention, which enhances objectivity and improves accuracy, may also have the following technical features, wherein step 9 includes the following sub-steps: Step 9-1, using n two-dimensional data points x i (r i1 r i2 Describing n failure modes, assuming that each cluster of data points follows a Gaussian distribution, the probability density function of the multivariate Gaussian distribution is: In the formula, μ is the mean vector, ∑ is the covariance matrix, and D is the feature dimension describing the data; Step 9-2, let K be the number of clusters, β s Let be the weights, and represent the probability of the s-th Gaussian distribution being drawn. Then the probability density function of the Gaussian mixture model is: Step 9-3, let θ be the parameter vector, then the likelihood function of the Gaussian mixture model is: Taking its logarithm, we get Step 9-4, Initialize parameters: Randomly select K data points from the dataset as cluster centers, calculate the distance from each data point to each cluster center, and complete the first clustering; Let θ0 = (β k0 μ s0 , ∑ s0 ), s = 1, 2, ..., K, the s-th Gaussian distribution containing u data points has a weight β. s0 =u / n, mean vector is μ s0 The covariance matrix is ​​∑ s0 Step 9-5: Calculate the probability of each data point: the probability of the i-th data point xi The probability obtained through the s-th Gaussian distribution is Step 9-6: Compare the probabilities of the same data point under different Gaussian distributions. If the i-th data point x i If the probability obtained through the s-th Gaussian distribution is the highest, then it belongs to the s-th Gaussian distribution cluster; Step 9-7, obtain new parameters based on the clustering results: Step 9-8: Repeat steps 9-5 to 9-7 until |ln(L) is reached. (N-1) -ln(L) (N) When the model converges to |<ε, the final clustering result is obtained.

[0014] Invention Function and Effect

[0015] This invention presents an improved FMEA method to enhance objectivity and accuracy. It improves upon traditional FMEA methods based on triangular fuzzy numbers, grey relational analysis, and Gaussian mixture model clustering analysis. First, the method uses a fuzzy language set to convert a ten-point scale into triangular fuzzy numbers for scoring fault modes, thus addressing the difficulty in assessing risk factors. Second, it combines the analytic hierarchy process (AHP) to determine subjective weights with the entropy weight method to determine objective weights, calculating the comprehensive weight of risk factors. This addresses the issues of strong subjectivity and failure to consider the relative importance of risk factors in traditional FMEA methods. Third, it calculates the grey relational degree of each fault mode and ranks the fault modes, avoiding the low accuracy of the RPN calculation method. Finally, it uses a Gaussian mixture model-based clustering method to perform cluster analysis on the fault modes based on their RPN values ​​and grey relational degrees, thereby improving the efficiency of subsequent fault mode prevention and monitoring.

[0016] Furthermore, to overcome the shortcomings of the K-means algorithm, this invention also employs a clustering method based on the Gaussian Mixture Model (GMM) and applies the expectation-maximization algorithm to achieve adaptive clustering, thereby improving the effectiveness and efficiency of clustering. Attached Figure Description

[0017] Figure 1 This is a structural block diagram of the improved FMEA method in this embodiment of the invention, which enhances objectivity and improves accuracy;

[0018] Figure 2 This is a flowchart of clustering based on Gaussian mixture model in an embodiment of the present invention;

[0019] Figure 3 This is a schematic diagram illustrating the results of Gaussian mixture model clustering of failure modes of a manufacturing execution system in an embodiment of the present invention; and

[0020] Figure 4This is the flowchart and principle diagram of the traditional FMEA method. Detailed Implementation

[0021] To make the technical means, creative features, objectives and effects of this invention easier to understand, the following describes in detail an improved FMEA method that enhances objectivity and improves accuracy, in conjunction with embodiments and accompanying drawings.

[0022] <Example>

[0023] This embodiment employs an improved FMEA method that enhances objectivity and increases accuracy to analyze and evaluate the Manufacturing Execution System (MES).

[0024] Figure 1 This is a structural block diagram of the improved FMEA method in this embodiment of the invention, which enhances objectivity and improves accuracy.

[0025] like Figure 1 As shown, this method first uses a fuzzy language set to convert the ten-point system into triangular fuzzy numbers to score the failure modes; secondly, it uses the analytic hierarchy process (AHP) to determine subjective weights and the entropy weight method to determine objective weights to calculate the comprehensive weight of risk factors; thirdly, it calculates the grey relational degree of each failure mode and ranks the failure modes; finally, it uses a clustering method based on Gaussian mixture model to perform cluster analysis on the failure modes based on the failure mode RPN value and grey relational degree.

[0026] The specific process is as follows:

[0027] Step 1: Assemble an FMEA team based on the analysis requirements and obtain information on how the object of analysis operates.

[0028] In this embodiment, the FMEA team consists of MES information project senior manager A, MES implementation consultant B, MES information consultant C, and university MES reliability researcher D.

[0029] Step 2: The FMEA team identifies n potential failure modes for the analyzed object based on its operational behavior: FM1, FM2, ..., FM n .

[0030] After team brainstorming, the FMEA failure modes were finally categorized into 4 types and 13 variations based on four aspects: technical failures, requirement failures, management and personnel failures, and unexpected failures, as shown in Table 1 below. For ease of description, these failure modes are numbered FM. i (i = 1, 2, ..., 13) represents the following table 2.

[0031]

[0032]

[0033] Table 1 FMEA Table

[0034]

[0035] Table 2 Fault Mode Correspondence Numbers

[0036] Step 3: Each member of the FMEA team evaluates and scores the above 13 failure modes according to the scoring criteria shown in Table 3 below, based on the three risk factors of S, O, and D. The average score of each member's score is then taken to obtain the score for each failure mode. The final scoring results are shown in Table 4 below.

[0037]

[0038] Table 3 FMEA Scoring Criteria

[0039]

[0040] Table 4 FMEA Scoring Results

[0041] Step 4: Calculate the risk priority number (RPN) for each failure mode, where the risk priority number for the i-th failure mode is r. i1 =S i ×O i ×D i .

[0042] Step 5: Establish a 7-granularity fuzzy language set, and correlate the evaluation language and member scores with triangular fuzzy numbers. Defuzzify the triangular fuzzy numbers A = (a1, a2, a3) using the gradient average synthesis method to obtain... The team has l members, and the weight of the k-th member is α. k Its risk factor for the j-th failure mode is u. ijk Then the team's overall score for this item. in Thus, the failure mode rating matrix P = (p ij ) n×3 .

[0043] In this embodiment, the fuzzy evaluation languages ​​and triangular fuzzy numbers corresponding to different scores are shown in Table 5 below, and the fuzzy evaluation results corresponding to the fault modes are shown in Table 6 below:

[0044]

[0045]

[0046] Table 5. Evaluation Language Corresponding Scores and Triangular Fuzzy Numbers

[0047]

[0048] Table 6. Expert Evaluation Results of Improved FMEA

[0049] For the four experts—Senior Manager A of MES Information Project, MES Implementation Consultant B, MES Information Consultant C, and MES Reliability Researcher D from a university—A and C have relatively less practical experience in MES applications compared to B, and their theoretical knowledge is not as comprehensive and in-depth as that of D. After team discussion, the weight vector for the four experts was determined to be α = (0.2, 0.3, 0.2, 0.3).

[0050] The results obtained after defuzzifying the evaluation are shown in Table 7 below. The evaluation matrix P of the experts is obtained according to the weight vector α of the experts. The comprehensive scores of each failure mode are shown in Table 8 below.

[0051]

[0052]

[0053] Table 7 Deblurring Results

[0054]

[0055] Table 8 Comprehensive Failure Mode Score

[0056] Step 6: Determine the subjective and objective weights of the risk factors using the analytic hierarchy process (AHP) and entropy weight method, respectively. Calculate the comprehensive weight w of the risk factors based on the subjective and objective weights. j The overall weight vector is w = (w1, w2, w3).

[0057] The specific process for this step is as follows:

[0058] Step 6-1: Use the analytic hierarchy process (AHP) to determine the subjective weights of the risk factors, i.e., obtain the judgment matrix Z = (z... jj′ ) 3×3 The matrix scale is shown in Table 9 below, and the largest eigenvalue of matrix Z is λ. max Continuously adjust Z until The consistency test was passed, where RI is the average random consistency index. According to Table 10 below, RI = 0.58, thus yielding the subjective weights of the three risk factors.

[0059]

[0060] Table 9 Determines the matrix scale

[0061]

[0062] Table 10 Average Random Consistency Index

[0063] Step 6-2: Determine the objective weights of risk factors using the entropy weight method, i.e., calculate the proportion of the i-th failure mode score in the j-th risk factor to the total weight of that risk factor. Thus, the information entropy redundancy of the j-th risk factor is obtained as follows: Its objective weight is

[0064] Step 6-3: Adjust the subjective weights of the risk factors according to their information entropy redundancy to obtain the comprehensive weight of the j-th risk factor. The overall weight vector is w = (w1, w2, w3).

[0065] Based on the above steps for determining the weights of the three fault elements, in this embodiment, after team discussion, the judgment matrices for S, O, and D are shown in Table 11 below:

[0066]

[0067] Table 11 S, O, D Judgment Matrix

[0068] After calculating λ max =3.0037, CI =0.0018, CR =0.0032 < 0.1, passing the consistency test. The subjective weight w is obtained. S = (0.6479, 0.2299, 0.1222). Next, the objective weight w is calculated using the entropy weight method. O = (0.1015, 0.2933, 0.6052), the redundancy is d = (0.0342, 0.0988, 0.2039), and the final comprehensive weight of S, O, and D is w = (0.3176, 0.3254, 0.3570).

[0069] Step 7, let the reference matrix Thus, the grey relational coefficient is obtained. The grey relational degree of the i-th fault mode is obtained as follows

[0070] Step 8: The magnitude of the grey relational degree is positively correlated with the degree of influence of the failure mode on the analysis object. Therefore, the failure modes are sorted according to the grey relational degree to obtain the corresponding sorting results.

[0071] In this embodiment, based on the scores of each risk factor of each failure mode, the grey relation coefficient calculation formula in step 7 above is substituted to obtain the grey relation coefficients of the three risk factors of different failure modes. Then, combined with the comprehensive weight of the risk factors, the grey relation degree of each failure mode is obtained. The modes are sorted according to the grey relation degree, and the sorting results are shown in Table 12 below.

[0072]

[0073] Table 12 Improved FMEA Sorting Results

[0074] Step 9, r i1 With r i2 As two feature values ​​describing the i-th fault mode, we obtain the description matrix R = (r it ) n×2 Let represent n two-dimensional data points, which are used to describe n fault modes. These two-dimensional data points are clustered into K classes using a clustering method based on Gaussian mixture model to obtain the clustering results.

[0075] Specifically: for n two-dimensional data points x i (r i1 r i2 Assuming that each cluster of data points follows a Gaussian distribution, the probability density function of the multivariate Gaussian distribution is:

[0076]

[0077] In the formula, μ is the mean vector, ∑ is the covariance matrix, and D is the feature dimension describing the data.

[0078] Let K be the number of clusters, β s Let be the weights, and represent the probability of the s-th Gaussian distribution being drawn. Then the probability density function of the Gaussian mixture model is: Let θ be a parameter vector, then the likelihood function of the Gaussian mixture model is: Taking its logarithm, we get

[0079] Figure 2 This is a flowchart of clustering based on Gaussian mixture model in an embodiment of the present invention.

[0080] like Figure 2 As shown, this embodiment completes cluster analysis based on Gaussian mixture model through the following steps:

[0081] (1) Initialize parameters: Randomly select K data points from the dataset as cluster centers, calculate the distance from each data point to each cluster center, and thus complete the first clustering; let θ0 = (β s0 μ s0 , ∑ s0), s = 1, 2, ..., K, the s-th Gaussian distribution containing u data points has a weight β. s0 =u / n, mean vector is μ s0 The covariance matrix is ​​∑ s0 .

[0082] (2) Calculate the probability of each data point: the probability of the i-th data point x i The probability obtained from the s-th Gaussian distribution

[0083] (3) Compare the probabilities of the same data point under different Gaussian distributions. If the i-th data point x i If the probability obtained through the s-th Gaussian distribution is the highest, then it belongs to the s-th Gaussian distribution cluster.

[0084] (4) Obtain new parameters based on the clustering results:

[0085] (5) Repeat steps (2) to (4) until |ln(L) is reached. (N-1) -ln(L) (N) If | < ε (ε is a sufficiently small threshold), the model converges, and the final clustering result is obtained.

[0086] In this embodiment, the description matrix of the failure mode is obtained based on the RPN value and gray relational degree of the failure mode.

[0087] To facilitate subsequent analysis, the description matrix is ​​arranged as follows: Normalization is performed to ensure that the dimensions of the two sets of description vectors are consistent. After normalization, the description matrix R of the fault mode is as follows:

[0088]

[0089] Figure 3 This is a schematic diagram of the results of clustering the failure modes of the manufacturing execution system based on the Gaussian mixture model in an embodiment of the present invention.

[0090] Based on practical considerations, failure modes are generally categorized into three types: those with significant impact, those with moderate impact, and those with minor impact. Therefore, we set K=3. Using a Gaussian mixture model-based clustering method, the failure modes are clustered according to their description matrix R. The clustering results are as follows: Figure 3 As shown.

[0091] Depend on Figure 3 It can be seen that "Cluster3" includes FM6, FM2, and FM 13Three failure modes. The RPN values ​​and gray correlation coefficients for these three failure modes are all relatively high, indicating a strong impact on MES operation. Therefore, maximum resources should be invested in prevention and monitoring. "Cluster1" includes FM1, FM4, FM5, FM7, FM8, FM9, and FM... 12 There are seven failure modes. The RPN values ​​and grey relational degrees of these seven failure modes are all at a moderate level, or one value is larger than another, indicating that their impact on MES operation is not as strong as the previous failure mode cluster. Therefore, excessive resources need to be invested in their prevention and monitoring. "Cluster 2" includes FM3 and FM... 10 FM 11 The three failure modes all exhibited low RPN values ​​and low grey relational degrees, indicating a weak impact on MES operation. Therefore, appropriate resource allocation for prevention and monitoring is sufficient. After discussion with experts, this clustering result is also consistent with the actual situation.

[0092] In summary, "Cluster1" represents a fault mode cluster with moderate impact, "Cluster2" represents a fault mode cluster with minor impact, and "Cluster3" represents a fault mode cluster with significant impact. This clustering result is quite reasonable.

[0093] Step 10: Analyze the failure modes based on the sorting and clustering results, and formulate corresponding prevention and response measures for different failure modes according to the category and the magnitude of gray relational degree.

[0094] Functions and effects of the embodiments

[0095] This embodiment presents an improved FMEA method that enhances objectivity and accuracy. It improves upon traditional FMEA methods by using triangular fuzzy numbers, grey relational analysis, and Gaussian mixture model clustering analysis. First, it transforms the 10-point scale into triangular fuzzy numbers to score failure modes using a fuzzy language set, thus addressing the difficulty in assessing risk factors. Second, it combines subjective weights determined by the analytic hierarchy process (AHP) with objective weights determined by the entropy weight method to calculate the comprehensive weight of risk factors, overcoming the inherent subjectivity and lack of consideration for the relative importance of risk factors in traditional FMEA methods. Third, it calculates the grey relational degree of each failure mode and ranks them, avoiding the low accuracy of the RPN calculation method. Finally, it uses a Gaussian mixture model-based clustering method to cluster failure modes based on their RPN values ​​and grey relational degrees, improving the efficiency of subsequent failure mode prevention and monitoring. This embodiment uses this method to conduct reliability and risk assessment analysis on an MES system, and the results verify its effectiveness and applicability.

[0096] Furthermore, to overcome the shortcomings of the K-means algorithm, this embodiment adopts a clustering method based on Gaussian Mixture Model (GMM) and applies the expectation-maximization algorithm to achieve adaptive clustering, thereby improving the effectiveness and efficiency of clustering.

[0097] The above embodiments are only used to illustrate specific implementations of the present invention, and the present invention is not limited to the scope of the description of the above embodiments.

Claims

1. An improved FMEA method that enhances objectivity and increases accuracy, characterized in that, Includes the following steps: Step 1: Assemble an FMEA team based on the analysis requirements and obtain information on how the object of analysis operates; Step 2: Determine the analysis object based on its operating mode. Potential failure modes: ; Step 3: Each member of the FMEA team evaluates and scores the three risk factors (S, O, D) for each failure mode based on a predetermined scoring standard. The average score of all team members is then calculated to obtain the result. The scores of the three risk factors for each failure mode are as follows: ; Step 4, calculate the Risk Priority Number (RPN) for each failure mode. The risk priority of each failure mode is: ; Step 5: Establish a 7-granularity fuzzy language set, and correlate the evaluation language, member scores, and triangular fuzzy numbers to obtain the fault mode rating matrix. ; Step 6: Determine the subjective and objective weights of the risk factors using the analytic hierarchy process (AHP) and entropy weight method, respectively; and calculate the comprehensive weight of the risk factors based on the subjective and objective weights. The comprehensive weight vector is ; Step 7, let the reference matrix Thus, the grey relational coefficient is obtained. , obtained the The grey relational degree of each failure mode is ; Step 8: The magnitude of the grey relational degree is positively correlated with the degree of influence of the fault mode on the analysis object. Therefore, the fault modes are sorted according to the grey relational degree to obtain the corresponding sorting results. Step 9, As a description of the Two feature values ​​of each fault mode are used to obtain the description matrix. , express Two-dimensional data points are used to cluster these two-dimensional data points into K classes using a clustering method based on Gaussian mixture model to obtain the clustering results; Step 10: Analyze the failure modes based on the sorting results and clustering results, and formulate corresponding prevention and response measures for different failure modes according to the size of the category and the gray relational degree.

2. The improved FMEA method according to claim 1, which enhances objectivity and improves accuracy, is characterized in that: in, In step 3, risk factors S, O, and D represent severity, occurrence, and detectability, respectively. When scoring the assessment, 1 point indicates that the failure mode has the least severe impact on the analyzed object, and 10 points indicates that the failure mode has the most severe impact on the analyzed object.

3. The improved FMEA method according to claim 2, which enhances objectivity and improves accuracy, is characterized in that: in, The process of obtaining the fault mode scoring matrix is ​​as follows: Triangular blur number Defuzzification is performed using the gradient average synthesis method to obtain... ; The FMEA team has a total of Member, No. The weight of each member is Its corresponding number The first failure mode The risk factors are Then the team for the first Comprehensive score of each failure mode ,in Thus, the fault mode scoring matrix is ​​obtained. .

4. The improved FMEA method according to claim 3, which enhances objectivity and improves accuracy, is characterized in that: in, Step 6 includes the following sub-steps: Step 6-1: Use the analytic hierarchy process (AHP) to determine the subjective weights of risk factors, i.e., obtain the judgment matrix based on the comparison of the importance of different risk factors. ,matrix The largest eigenvalue is Continuously adjust Until The consistency test is then passed, thus obtaining the subjective weights of the three risk factors. ; Step 6-2: Determine the objective weights of risk factors using the entropy weight method, i.e., calculate the first... Among the risk factors, the first one is... The proportion of each failure mode score to this risk factor is: Thus, the first The information entropy redundancy of each risk factor is: Its objective weight is ; Step 6-3: Adjust the subjective weights of risk factors according to their information entropy redundancy to obtain the... The combined weight of each risk factor is The comprehensive weight vector is .

5. An improved FMEA method to enhance objectivity and increase accuracy according to claim 4. Its features are: in, Step 9 includes the following sub-steps: Step 9-1, using Two-dimensional data points describe Given several failure modes, assuming that each cluster of data points conforms to a Gaussian distribution, the probability density function of the multivariate Gaussian distribution is: , In the formula, It is the mean vector. Let covariance matrix be the variance matrix. To describe the feature dimensions of the data; Step 9-2, let The number of clusters in the clustering. Let be the weight, representing the th If the probability of each Gaussian distribution being drawn is given, then the probability density function of the Gaussian mixture model is: ; Step 9-3, let Let be the parameter vector, then the likelihood function of the Gaussian mixture model is: Taking its logarithm, we get ; Step 9-4, Initialize parameters: Randomly select from the dataset Using each data point as a cluster center, calculate the distance from each data point to each cluster center to complete the first clustering. make ,Include The first data point The weights of the Gaussian distribution are: The mean vector is The covariance matrix is ; Step 9-5, calculate the probability of each data point: Data points Through the first The probability obtained from a Gaussian distribution is ; Step 9-6: Compare the probabilities of the same data point under different Gaussian distributions. If the... Data points Through the first If the probability obtained from the Gaussian distribution is the largest, then it belongs to the th . A Gaussian distribution cluster; Step 9-7: Obtain new parameters based on the clustering results. , , , ; Step 9-8, repeat steps 9-5 to 9-7, until... The model converges, yielding the final clustering result.

Images