A data-driven failure mode and effects analysis method

Abstract

This invention discloses a data-driven failure mode and effects analysis (FEM) method. First, historical data related to failure modes and their corresponding risk parameters are collected. The collected risk data is preprocessed to train a data-driven model, and the model with the highest accuracy is selected as the prediction model. Next, the trained model is used to predict the probability of occurrence of failure modes. Finally, an intuitionistic fuzzy method is used to evaluate the failure modes, and a distance operator is employed to measure the risk of the failure modes. This method fully utilizes the large amount of data generated throughout the product lifecycle to extract useful risk information, enabling data-driven assessment of product failure modes. This saves time and economic costs associated with uncertain design, avoids waste of human and material resources, improves product reliability, supports maintenance planning and operations, and ultimately achieves the identification, prediction, and risk assessment of product failure modes.

Description

Technical Field

[0001] This invention belongs to the field of reliability technology, specifically relating to a data-driven failure mode and effects analysis method. Technical Background

[0002] Product reliability is crucial for enterprises to maintain their core competitiveness. The failure of modern industrial products throughout their entire lifecycle exhibits complexity and uncertainty, primarily manifested in: complex product structures, high levels of knowledge and technology integration, cross-functional and multi-disciplinary product design, intricate and precise manufacturing processes, high manufacturing costs, and variable operating environments. Against this backdrop, effectively identifying, systematically managing, and scientifically assessing the failure mode risks of modern industrial products, improving product reliability, and providing enterprises with product risk information is an urgent problem to be solved.

[0003] Failure Mode and Effect Analysis (FMEA) is a risk assessment method designed to identify known or potential failure modes and their causes in a system, as well as the impact of these failure modes on the product. It assesses and determines the potential risks of these failure modes and takes timely measures to address serious risks, thereby improving the reliability and safety of product design or operation. As a classic engineering risk assessment technique, FMEA is widely used throughout the entire lifecycle of industrial products to define, identify, and eliminate known or potential product failures, malfunctions, and errors. The results of its analysis help enterprise risk analysts to promptly correct failure modes that adversely affect the product, thereby improving product reliability.

[0004] Traditional FMEA techniques, when analyzing a specific product or system, often require a cross-functional team of experts to conduct risk analysis, ultimately recording key information from the FMEA process in the form of a product FMEA analysis table. The Risk Priority Number (RPN) is considered a crucial basis for quantifying failure mode risk assessment, primarily supported by three risk factors: the probability of occurrence (OP), the severity (S), and the detectability (D).

[0005] In recent years, with the development of digitalization and intelligent manufacturing technologies, a large amount of data has been generated throughout the product lifecycle. This data includes not only extensive manufacturing, operation, and environmental data related to product equipment, but also a significant amount of qualitative data such as subjective evaluations. Traditional FMEA methods are primarily subjective assessments, and due to the complexity of products, most evaluations often suffer from significant information gaps and cognitive uncertainties. On the one hand, establishing a scientific, data-driven, semi-quantitative FMEA methodology that integrates quantitative and qualitative assessments based on large datasets is a challenge. On the other hand, reducing the subjectivity of assessment data and quantifying the cognitive uncertainties of experts is another challenge. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes a data-driven failure mode and effects analysis method.

[0007] The technical solution adopted in this invention is: a data-driven failure mode and effects analysis method, the specific steps of which are as follows:

[0008] S1. Data Acquisition and Preprocessing;

[0009] S2. For multiple failure modes of the product, build multiple data-driven models, and select the model with the highest diagnostic accuracy for the occurrence of the failure mode for the assessment of the probability of occurrence of the failure mode.

[0010] S3. Scientifically assess the severity and detectability of failure modes based on fuzzy evaluation.

[0011] S4. Using the assessment results of the three risk factors obtained in steps S2 and S3, the distance operator is used to aggregate information to measure the risk of different failure modes of the product and obtain the risk ranking of each failure mode.

[0012] Furthermore, in step S1, a risk parameter database consisting of failure causes (such as temperature, humidity, vibration, impact, wind speed, dust, etc.) is collected and established, and the collected data is preprocessed as follows:

[0013] S101. Identify the causes of product failure modes and confirm the input parameters of the data-driven model;

[0014] The FMEA analysis table identifies the various risk parameters (rp) that lead to product failure modes. Typically, rp includes three types of parameters:

[0015] (1) Product design and control parameters (e.g., mechanical properties, material properties);

[0016] (2) Product status parameters (e.g., power requirements, function duration);

[0017] (3) External factors (e.g., environmental conditions).

[0018] For different failure modes, risk parameters with strong correlations are identified for subsequent failure mode risk assessment. Finally, a set rf consisting of risk parameters with strong correlations to the i-th failure mode is identified. i for:

[0019]

[0020] Where i = 1, 2, ..., I represents a total of I failure modes. It represents the n subsets of a set consisting of n risk parameters that are strongly correlated with the i-th failure mode.

[0021] S102. Collect and establish a data sample library with risk parameters as input parameters and whether the failure mode occurs as output parameters;

[0022] The specific input parameter types are mainly obtained in step S101, and the output parameters can be regarded as a classification set fm. i ={0,1}, meaning that the risk parameter is 0 when the corresponding failure mode occurs, and 1 when the corresponding failure mode occurs.

[0023] In summary, the final collected dataset is as follows:

[0024] T i ={(rf i ,fm i )1,(rf i ,fm i )2,,(rf i ,fm i ) m}

[0025] Among them, T i T represents the data sample set for the i-th failure mode. i ={(rf i ,fm i )1,(rf i ,fm i )2,...,(rf i ,fm i ) m} indicates that a total of m samples were collected.

[0026] S103. Perform data preprocessing on the data sample library;

[0027] Data preprocessing includes four main steps:

[0028] (1) Fill in the missing data with the mean of the data samples;

[0029] (2) Normalize the input parameters using the minimum-maximum method to unify the units;

[0030] (3) Divide the processed dataset into training set, test set and validation set in a ratio of 7:2:1;

[0031] (4) Use data augmentation techniques to expand the data samples of small datasets.

[0032] Furthermore, step S2 specifically includes the following:

[0033] S201. Construct and train two classification data-driven models, including support vector machines and artificial neural networks, and complete the parameter tuning of the models;

[0034] In support vector machines, the parameter tuning target is the kernel function, including Gaussian kernel function, linear kernel function and polynomial kernel function; in artificial neural networks, the parameter tuning target is the number of neurons in the hidden layer.

[0035] Network parameter tuning is evaluated based on the accuracy of the sample test set. For different failure modes, the parameter tuning model with the highest accuracy on the sample test set is selected as the subsequent data-driven model for that failure mode. i .

[0036] S202, Iteration of risk data and failure prediction;

[0037] Since risk parameters are either pre-collected or follow a certain distribution, if the temperature at a certain moment is known in advance, the risk parameter is collected using S as the step size for value prediction. The collected risk parameters are then substituted into the data-driven model trained on the failure modes. i The model makes predictions based on newly collected risk parameters, determining whether each failure mode will occur. As time progresses, new data is fed into the model for updates and training, resulting in a new data-driven model. Implement model iteration.

[0038] S203, Calculate the probability of failure mode occurrence;

[0039] Maximum likelihood estimation is used to estimate the probability P(fm) of each failure mode. i =1):

[0040]

[0041] Where S represents the risk parameter acquisition step size, fm i ∈fm i ,

[0042] Furthermore, in step S3, the assessment of failure mode severity and detectability relies on the product's design logic and is primarily conducted by an expert team based on their knowledge and experience, as detailed below:

[0043] S301. Use intuitive fuzzy methods to collect expert team assessments of the severity and detectability of product failure modes;

[0044] Let X be an evaluation set. Intuitive fuzzy logic requires an expert team to determine the membership degree μ(x) and non-membership degree υ(x) of the evaluation object x∈X, and to map them onto the number field using continuous functions. The membership function μ(x): X→[0,1], the non-membership function υ(x): X→[0,1], and satisfies... Furthermore, π(x) = 1 - μ(x) - υ(x) represents the degree of hesitation in mapping element x to the intuitionistic fuzzy set, and 0 ≤ π(x) ≤ 1. The smaller the value of π(x), the more comprehensive the knowledge of element x and the more precise the value of element x; conversely, the larger the value of π(x), the more uncertain the knowledge of element x and the more fuzzy the value of element x.

[0045] S302. Integrate intuitionistic fuzzy information and define intuitionistic fuzzy numbers and their properties;

[0046] For an evaluation result, a set of (μ(x),υ(x)) is called an intuitionistic fuzzy number α=(μ α ,υ α ), where μ is an intuitive fuzzy number α. α ∈[0,1] represents the membership degree of the evaluated object, υ α ∈[0,1] represents the non-membership degree of the evaluated object, and μ α +υ α ≤1.

[0047] For an intuitive fuzzy number, μ α The higher the value, the better, υ α The smaller the value, the better, therefore α + = (1,0) and α - = (0,1) represent the maximum and minimum intuitionistic fuzzy numbers, respectively. Furthermore, define S(α) = μ α -υ α And H(α)=μ α +υ α These represent the value and precision of an intuitive fuzzy number, respectively.

[0048] S303. Define the basic calculation rules, distance operator, and weighted average operator for intuitionistic fuzzy numbers;

[0049] For any three intuitive fuzzy numbers λ > 0 represents a real number that satisfies the following calculation rules:

[0050]

[0051] Based on the values ​​and precision of intuitionistic fuzzy numbers, a comparison is made between any two intuitionistic fuzzy numbers α1 and α2:

[0052]

[0053] To measure the deviation between any two intuitionistic fuzzy numbers α1 and α2, the intuitionistic fuzzy distance operator d is introduced. IFD Describe:

[0054]

[0055] Based on the calculation rules of intuitionistic fuzzy numbers, the weighted average operator of intuitionistic fuzzy numbers is derived as shown in equation (5), defined as follows: Given n intuitively fuzzy numbers, their weighted average operator IFWA:Ω n →Ω is:

[0056]

[0057] Where w = (w1, w2, ..., w n ) T α i The corresponding weight vector, and T denotes the transpose of a matrix.

[0058] S304. The subjective opinions of the FMEA team members are summarized by using the IFWA operator (Equation (5)), which mainly includes the risk assessment values ​​of the severity and detectability of the failure mode, as well as the subjective weights of the risk factors of the two.

[0059]

[0060] Where, λ k This represents the weights of different expert members, and This represents the intuitive fuzzy assessment result of the k-th team member regarding the i-th failure mode and the j-th risk factor. There are a total of K team members, and α... ij This represents the team's risk assessment value for the severity and detectability of the failure mode after aggregating information. This represents the subjective, intuitive, fuzzy assessment result of the k-th team member for the j-th risk factor. There are K team members in total. j This represents the team's subjective weighting of the two risk factors after aggregating the information.

[0061] Furthermore, step S4 specifically includes the following:

[0062] S401, Subjective weights w for the two risk factors in step S304 j =(μ j ,υ j The risk factors are then defuzzified to obtain their subjective weights.

[0063]

[0064] Where, π j =1-μ j -υ j Indicates the degree of hesitation, and

[0065] S402. Establish a reference series of risk factors;

[0066] The reference series of risk factors represents the optimal level of all risk factors for failure modes in FMEA, employing... As a reference sequence value.

[0067] S403. Combining the failure mode occurrence probability results obtained in step S203, calculate the risk value of each failure mode using the calculation rule of equation (2) in step S303 and equation (4):

[0068]

[0069] Among them, R i R represents the risk value of each failure mode. i The larger the value, the greater the overall risk value, and the higher the risk priority.

[0070] The beneficial effects of this invention are as follows: The method of this invention first collects historical data related to failure modes and their corresponding risk parameters. The collected risk data is preprocessed to train a data-driven model, and the model with the highest accuracy is selected as the prediction model. Next, the trained model is used to predict the probability of occurrence of failure modes. Finally, an intuitionistic fuzzy method is used to evaluate the failure modes, and a distance operator is employed to measure the risk of the failure modes. This method can fully utilize the large amount of data generated throughout the product lifecycle to extract useful risk information, achieving data-driven assessment of product failure modes. It saves time and economic costs associated with uncertain design, avoids waste of human and material resources, improves product reliability, supports the planning and operation of maintenance enterprises, and ultimately achieves the identification, prediction, and risk assessment of product failure modes. Attached Figure Description

[0071] Figure 1 This is a flowchart illustrating the overall process of a data-driven failure mode and effects analysis method according to the present invention.

[0072] Figure 2 This is a schematic diagram of the failure modes of soldered electronic components in an embodiment of the present invention.

[0073] Figure 3 This is a flowchart of data-driven model training and failure mode probability prediction in an embodiment of the present invention.

[0074] Figure 4 This is a training accuracy diagram of each failure mode model in the embodiments of the present invention.

[0075] Figure 5 This is a risk measurement diagram for each failure mode in the embodiments of the present invention. Detailed Implementation

[0076] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0077] This invention uses soldering failures in electronic products as an example of data-driven risk assessment, such as... Figure 1 The flowchart shown is a data-driven failure mode and effects analysis method of the present invention. The specific steps are as follows:

[0078] S1. Collect historical data related to failure modes and their corresponding risk parameters, and perform data preprocessing for model training;

[0079] S2. Train the data-driven models corresponding to each failure mode, and select the data-driven model with the highest accuracy to predict the occurrence frequency of failure modes.

[0080] S3. Use intuitionistic fuzzy methods to collect the expert team's assessment results on the severity and detectability of failure modes, and complete the group assessment information processing.

[0081] S4. Using distance operators combined with fuzzy domain calculation rules, the risk information from steps S2 and S3 is aggregated, and the risk value of each failure mode is finally measured to provide a basis for enterprise decision-making.

[0082] The following uses four typical failures of electronic component soldering—misalignment, cold solder joint, polarity reversal, and insufficient solder—as examples to explain the four steps mentioned above in detail:

[0083] First, we introduce the four typical failure modes applied in this invention, such as... Figure 2 As shown, images of four typical failure scenarios and failure scenarios are presented. Secondly, to complete this FMEA analysis, a cross-functional expert team of five experts was assembled. Based on the experts' experience, qualifications, and the relevance of their professional knowledge, the five experts were assigned weights λ. k={0.1,0.2,0.2,0.3,0.2}, any parameter setting that satisfies the corresponding objective facts and physical principles is applicable to this invention.

[0084] In this embodiment, in step S1, the risk parameters that lead to various failure modes are identified and screened, relevant historical failure data are collected and data preprocessing is completed to obtain data-driven data samples and divide the sample set, as follows:

[0085] a. After accumulating knowledge based on the product FMEA analysis table in Table 1, the risk parameters corresponding to the four failure modes are confirmed from the company's failure knowledge base and the experience of experts, as shown in Table 2.

[0086] Table 1

[0087]

[0088] Table 2

[0089]

[0090] b. Collect 200 failure data samples for each failure mode. The input of the data is the risk parameter, and the output of the data is whether the failure mode has occurred. The failure mode offset data is shown in Table 3.

[0091] c. Missing data were filled with the sample mean, and the units of risk parameters were unified using the min-max method to complete the data preprocessing. The failure mode offset data are shown in Table 3.

[0092] Table 3

[0093]

[0094] like Figure 3 As shown, in this embodiment, in step S2, the collected data is used to train a data-driven model and adjust its parameters. For different failure modes, the most suitable data-driven model is selected based on high accuracy to predict the probability of failure mode occurrence, thus obtaining the probability of failure mode occurrence. Specifically, as follows:

[0095] a. Construct two data-driven models, support vector machine and artificial neural network, using the collected data and tune their parameters.

[0096] In this study, the parameter tuning target for Support Vector Machines (SVMs) is the kernel function, including Gaussian, linear, and polynomial kernel functions; the parameter tuning target for artificial neurons is the number of neurons in the hidden layer. After parameter tuning, the accuracy results of the data-driven models for the four failure modes on the test set are as follows: Figure 4 As shown in Table 4, the models trained for the four failure modes were finally confirmed.

[0097] Table 4

[0098] Offset Support Vector Machine Polynomial kernel function cold solder joint Artificial Neural Networks Number of neurons 20 Crossing legs Artificial Neural Networks Number of neurons 10 Tin deficiency Support Vector Machine Polynomial kernel function

[0099] b. Using the distribution properties and the Latin hypercube sampling method, 1000 sets of new risk parameter data were collected and substituted into the data-driven models trained for each failure mode to predict the probability of the failure mode occurring. The prediction results and the probability of occurrence of the four failure modes are shown in Table 5.

[0100] Table 5

[0101] Number of failures 11 6 2 4 Predict the probability of occurrence 0.011 0.006 0.002 0.004

[0102] In this embodiment, in step S3, the subjective assessments of the severity and detectability of the failure mode by the expert team are collected using intuitionistic fuzzy collection, and the group assessment information is processed to obtain the group assessment results of the severity and detectability of the failure mode, as follows:

[0103] a. Select the triangular membership function to map the evaluation semantics to the fuzzy domain. To facilitate expert operation, the evaluation experts can refer to the indicators in Table 6 to perform semantic evaluation of the evaluation object. The evaluation content mainly includes the severity and detectability of each failure mode and its corresponding subjective weight. The final evaluation results are shown in Table 7.

[0104] Table 6

[0105]

[0106] Table 7

[0107]

[0108] b. Using the processing method and calculation rules of intuitive fuzzy numbers in steps S302-S303, combined with the group processing method in step S304, the group information of each evaluation result is finally obtained, as shown in Table 8.

[0109] Table 8

[0110]

[0111] In this embodiment, in step S4, a failure mode risk measurement method is established, all risk factor information is aggregated, the risk values ​​of failure modes are obtained and sorted, as follows:

[0112] a. Using the fuzzy group processing information of the subjective weights of risk factors obtained above, the fuzziness is eliminated and normalized to obtain accurate risk factor weight values.

[0113] b. Using a distance metric operator, the aforementioned failure mode occurrence probability, severity, and detectability results are aggregated to finally obtain the risk value for each failure mode, such as... Figure 5 As shown.

[0114] exist Figure 5 Among the failure modes, lead lifting has the highest risk value, followed by misalignment, cold solder joints, and insufficient solder. Further investigation reveals that lead lifting causes open circuits in components, significantly impacting system failure. Furthermore, the microscopic nature of lead lifting makes it difficult to detect; therefore, lead lifting requires focused improvement. Additionally, misalignment occurs frequently, necessitating improved soldering precision to reduce its frequency.

[0115] In summary, the method of this invention can identify and measure the failure mode risk of products based on product lifecycle data, improving product reliability to support maintenance planning and operations. This invention constructs a data-driven FMEA-based product failure mode risk assessment model. On one hand, the method provides step-by-step guidance from data acquisition and preprocessing, over-analysis and algorithm parameter optimization to final failure prediction, accurately predicting the probability of product failure modes using product lifecycle data. On the other hand, the method provides step-by-step guidance and data processing methods for expert subjective assessment, effectively reducing the subjectivity of assessment data and quantifying the uncertainty of expert perception. It also details the risk measurement methods for failure modes, ultimately achieving the identification, prediction, and risk assessment of product failure modes.

Claims

1. A data-driven failure mode and effects analysis method, the specific steps of which are as follows: S1. Data Acquisition and Preprocessing; Collect and establish a risk parameter database consisting of failure causes, and preprocess the collected data. First, clarify the causes of product failure modes and confirm the input parameters of the data-driven model. Then, collect and establish a data sample library with risk parameters as input parameters and whether the failure mode occurs as output parameters. Finally, preprocess the data sample library. S2. For multiple failure modes of the product, build multiple data-driven models, and select the model with the highest diagnostic accuracy for the occurrence of the failure mode for the assessment of the probability of occurrence of the failure mode. S201. Construct and train two classification data-driven models, including support vector machines and artificial neural networks, and complete the parameter tuning of the models; in, The parameter tuning object for support vector machines is the kernel function, including: Gaussian kernel function, linear kernel function and polynomial kernel function; the parameter tuning object for artificial neural networks is the number of neurons in the hidden layer. Network parameter tuning is evaluated based on the accuracy of the sample test set. For different failure modes, the parameter tuning model with the highest accuracy on the sample test set is selected as the subsequent data-driven model for that failure mode. ; S202, Iteration of risk data and failure prediction; If the temperature at a certain moment is known in advance, then Predict the value of the risk parameter acquisition step size, and then substitute the acquired risk parameters into the data-driven model trained on the failure modes. The model makes predictions based on newly collected risk parameters, determining whether each failure mode will occur. As time progresses, new data is fed into the model for updates and training, resulting in a new data-driven model. To achieve model iteration; S203, Calculate the probability of failure mode occurrence; Maximum likelihood estimation is used to estimate the probability of occurrence of each failure mode. : (1)； in, Indicates the step size for risk parameter collection. , ; S3. Scientifically assess the severity and detectability of failure modes based on fuzzy evaluation. The assessment of failure mode severity and detectability relies on the product's design logic and is primarily conducted by an expert team based on their knowledge and experience. First, the expert team's assessment results on the product's failure mode severity and detectability are collected using an intuitionistic fuzzy logic method. Then, by integrating this intuitionistic fuzzy information, the intuitionistic fuzzy number and its properties are defined, along with the basic calculation rules, distance operator, and weighted average (IFWA) operator. Finally, the IFWA operator is used to summarize the subjective opinions of the FMEA team members, mainly including the risk assessment values ​​for failure mode severity and detectability, as well as the subjective weights of the risk factors for both. S4. Using the assessment results of the three risk factors obtained in steps S2 and S3, the distance operator is used to aggregate information to measure the risk of different failure modes of the product and obtain the risk ranking of each failure mode.

2. The data-driven failure mode and effects analysis method according to claim 1, characterized in that, In step S1, a risk parameter database consisting of failure causes is collected and established, and the collected data is preprocessed as follows: S101. Identify the causes of product failure modes and confirm the input parameters of the data-driven model; The FMEA analysis table was used to identify the various risk parameters that could lead to product failure modes. ,generally It includes three types of parameters: 1) Product design and control parameters; 2) Product status parameters; 3) External factors; For different failure modes, risk parameters with strong correlations are identified for subsequent failure mode risk assessments, ultimately confirming the correlation with the first failure mode. A set of risk parameters that are highly correlated with failure modes for: ; in, Indicates shared ownership Failure mode of item, Indicates shared ownership Item and the A set of risk parameters that are highly correlated with failure modes A subset; S102. Collect and establish a data sample library with risk parameters as input parameters and whether the failure mode occurs as output parameters; The specific input parameter types are mainly obtained in step S101, and the output parameters can be regarded as a classification set. That is, the risk parameter is 0 when the corresponding failure mode occurs, and 1 when the corresponding failure mode occurs. In summary, the final collected dataset is as follows: ; in, Indicates the first Data sample set of failure modes, Indicates shared ownership Item samples were collected; S103. Perform data preprocessing on the data sample library; Data preprocessing includes four main steps: 1) Fill in missing data using the mean of the data samples; 2) Normalize the input parameters using the minimum-maximum method to unify the units; 3) Divide the processed dataset into training, testing, and validation sets in a 7:2:1 ratio; 4) Use data augmentation techniques to expand the data samples in small datasets.

3. The data-driven failure mode and effects analysis method according to claim 1, characterized in that, In step S3, the assessment of failure mode severity and detectability relies on the product's design logic and is mainly conducted by an expert team based on their knowledge and experience, as detailed below: S301. Use intuitive fuzzy methods to collect expert team assessments of the severity and detectability of product failure modes; set up For an evaluation set, intuitionistic ambiguity requires a team of experts to evaluate the objects. membership degree Non-membership degree Make a judgment and map it to the number field using a continuous function; membership function Non-membership function And satisfy ; Represents element Mapping to the degree of hesitation in intuitionistic fuzzy sets, and ; S302. Integrate intuitionistic fuzzy information and define intuitionistic fuzzy numbers and their properties; For an evaluation result, a set Known as an intuitive fuzzy number Among them, an intuitive fuzzy number of Indicates the degree of membership of the evaluated object. This indicates the degree of non-membership of the evaluated object, and ; For an intuitive fuzzy number, The higher the value, the better. The smaller the value, the better, therefore and Let these represent the largest and smallest intuitionistic fuzzy numbers, respectively; define... as well as These represent the value and precision of an intuitive fuzzy number, respectively. S303. Define the basic calculation rules, distance operator, and weighted average operator for intuitionistic fuzzy numbers; For any three intuitive fuzzy numbers , Let represent a real number that satisfies the following rules of calculation: (2)； Based on the value and precision of intuitionistic fuzzy numbers, for any two intuitionistic fuzzy numbers Perform a size comparison: (3)； To measure any two intuitive fuzzy numbers To address the bias, an intuitionistic fuzzy distance operator is introduced. Describe: (4)； Based on the calculation rules of intuitionistic fuzzy numbers, the weighted average operator of intuitionistic fuzzy numbers is derived as shown in equation (5), defined as follows: express If an intuitive fuzzy number is given, then its weighted average operator... for: (5)； in, express The corresponding weight vector, and , Represents the transpose of a matrix; S304. By using the IFWA operator of Equation (5), the subjective opinions of the FMEA team members are summarized, mainly including the risk assessment values ​​of the severity and detectability of the failure mode, as well as the subjective weights of the risk factors of the two. (6)； (7)； in, This represents the weights of different expert members, and , Indicates the first Team members on the first type of failure mode, first The intuitive fuzzy assessment results of the risk factors totaled [number missing]. Team members, This represents the team's risk assessment value for the severity and detectability of the failure mode after aggregating information; Indicates the first Team member gave the first The subjective weight intuition fuzzy assessment results of various risk factors totaled [number missing]. Team members, This represents the team's subjective weighting of the two risk factors after aggregating the information.

4. The data-driven failure mode and effects analysis method according to claim 1, characterized in that, In step S4, the specific details are as follows: S401, Subjective weighting of the two risk factors in step S304 Defuzzification is performed to obtain the subjective weights of the risk factors. : (8)； in, Indicates the degree of hesitation, and ; S402. Establish a reference series of risk factors; The reference series of risk factors represents the optimal level of all risk factors for failure modes in FMEA, employing... As a reference sequence value; S403. Combining the failure mode occurrence probability results obtained in step S203, calculate the risk value of each failure mode using the calculation rule of equation (2) in step S303 and equation (4): (9)； in, This indicates the risk value of each failure mode. The larger the value, the greater the overall risk value, and the higher the risk priority.

Images