Intelligent risk optimization method and system based on FMEA5 seven-step method

Abstract

The invention discloses an intelligent risk optimization method and system based on an FMEA5 seven-step method, and relates to the field of product design, and the method comprises the steps: carrying out the vectorization processing of product function description through a BERT model, storing historical failure data and improvement measure data, obtaining a dynamic knowledge base, carrying out the retrieval and matching of an input target product function in the dynamic knowledge base, and carrying out the retrieval and matching of the target product function. Generating an initial failure chain, wherein the initial failure chain comprises a failure reason, a failure mode and a failure influence which are associated with one another; determining a severity S value according to a failure influence in the initial failure chain, determining an occurrence degree O value according to a failure reason, determining a detection degree D value according to a detection scheme, inputting the severity S value, the occurrence degree O value and the detection degree D value into a three-dimensional logic matrix, and outputting an action priority AP value by adopting a conditional branch judgment mode; and judging the risk level of the target product according to the action priority AP value, and generating a risk assessment result. By implementing the method, the accuracy of failure risk assessment in the product design stage can be improved.

Description

Technical Field

[0001] This application relates to the field of product design, and in particular to an intelligent risk optimization method and system based on the FMEA5 seven-step method. Background Technology

[0002] As the complexity of manufactured products continues to increase, the requirements for their quality, reliability, and safety are becoming increasingly stringent. Failure Mode and Effects Analysis (FMEA), as a systematic risk analysis method, is widely used in quality and failure management during product development. This method analyzes the potential failure modes, causes, and effects of products, and then takes preventative measures to improve product quality and reliability.

[0003] In existing technologies, product development teams typically use spreadsheets to record the FMEA analysis process. Analysts first decompose the product functions, then identify the potential failure modes for each function based on experience, and calculate the Risk Priority Number (RPN) by multiplying the three scores of Severity (S), Occurrence (O), and Detection (D) to rank and manage failure risks.

[0004] However, this risk assessment method based on RPN multiplication suffers from numerical repeatability issues. For example, when a certain failure mode has S=9, O=3, D=4, the RPN is 108, while when another failure mode has S=5, O=8, D=3, the RPN is 120. This causes the former, with its higher risk level, to be masked by the latter, with its lower risk level, making it impossible to identify and address some serious failure risks in a timely manner. Summary of the Invention

[0005] This application provides an intelligent risk optimization method and system based on the FMEA5 seven-step method to improve the accuracy of failure risk assessment in the product design stage.

[0006] Firstly, this application provides an intelligent risk optimization method based on the FMEA5 seven-step method, applied to an intelligent risk optimization system. The method includes: vectorizing product function descriptions using a BERT model and storing historical failure data and improvement measure data to obtain a dynamic knowledge base. The historical failure data includes failure causes, failure modes, and failure effects, while the improvement measure data includes error-proofing design schemes and detection schemes. The input target product function is searched and matched in the dynamic knowledge base to generate an initial failure chain, which includes interrelated failure causes, failure modes, and failure effects. The severity (S) value is determined based on the failure effects in the initial failure chain, the occurrence (O) value is determined based on the failure causes, and the detectability (D) value is determined based on the detection schemes. The severity (S), occurrence (O), and detectability (D) values ​​are input into a three-dimensional logic matrix, and the action priority (AP) value is output using a conditional branching decision method. The risk level of the target product is determined based on the action priority (AP) value, generating a risk assessment result.

[0007] In the above embodiments, the BERT model is vectorized to describe product functions, constructing a dynamic knowledge base containing failure data and improvement measures, enabling intelligent retrieval and matching of target product functions. Based on a three-dimensional logical matrix, conditional branching is performed on the severity (S), occurrence (O), and detectability (D) values, outputting the action priority (AP) value, thereby eliminating the numerical repetition problem caused by traditional RPN multiplication operations. When a failure mode has S=9 and O=3, even if its RPN value is low, it will be accurately identified as a high-risk item, effectively avoiding the problem of severe failure risks being masked by low-risk items.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the input target product function is retrieved and matched in a dynamic knowledge base to generate an initial failure chain. The initial failure chain includes steps of interrelated failure causes, failure modes, and failure effects, specifically including: vectorizing the input target product function description and historical failure data in the dynamic knowledge base using the BERT model; calculating the cosine similarity between the description vector of the target product function and the historical failure data vector to obtain the historical failure data with the highest similarity; and generating the initial failure chain based on the historical failure data with the highest similarity.

[0009] In the above embodiments, the BERT model is used to vectorize the target product function description and historical failure data, respectively. The most similar historical failure data is obtained by calculating cosine similarity. An initial failure chain is generated based on the historical failure data with the highest similarity, establishing a complete failure causal relationship network. During failure analysis, the system can automatically identify historical failure cases related to the current function and quickly construct a complete failure chain including failure causes, failure modes, and failure effects, making the failure analysis process more systematic and intelligent.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the severity S value is determined based on the failure impact in the initial failure chain, the occurrence O value is determined based on the failure cause, and the detectability D value is determined based on the detection plan. The S value, O value, and D value are input into a three-dimensional logic matrix, and the action priority AP value is output using a conditional branch decision method. Specifically, this includes: determining the severity S value based on the failure impact data using a first conditional branch decision method; determining the occurrence O value based on the failure cause data using a second conditional branch decision method; determining the detectability D value based on the detection plan data using a third conditional branch decision method; and determining the action priority AP value based on the correspondence between the S value, O value, and D value in the three-dimensional logic matrix.

[0011] In the above embodiments, three independent conditional branching methods are used to determine the severity (S) value, occurrence (O) value, and detectability (D) value, respectively. These scores are then input into a three-dimensional logic matrix for comprehensive analysis. Based on the pre-defined correspondence between the S, O, and D values ​​in the matrix, the action priority (AP) value is determined, forming a complete risk assessment system. This multi-dimensional combined judgment method avoids risk judgment biases that may result from simple numerical multiplication, making the risk assessment results more accurate and reliable.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, after vectorizing the product function description using the BERT model and storing historical failure data and improvement measure data to obtain a dynamic knowledge base, the method further includes: performing hierarchical modeling on the historical failure data to obtain a multi-level functional tree structure of the product function, wherein each functional node is set with corresponding failure feature data; mapping the target product function to the multi-level functional tree structure to obtain the corresponding nodes of the target product function at each level; performing constraint analysis on the failure feature data of the lower-level functional nodes based on the failure feature data of the upper-level functional nodes according to the hierarchical order of the multi-level functional tree structure to obtain the associated failure modes of functional nodes at each level; performing supplementary analysis on the failure feature data of the upper-level functional nodes based on the failure feature data of the lower-level functional nodes according to the hierarchical order of the multi-level functional tree structure to obtain the combined failure effects of functional nodes at each level; and determining the initial failure chain based on the associated failure modes and combined failure effects.

[0013] In the above embodiments, historical failure data is hierarchically modeled to construct a multi-level functional tree structure, with each functional node having corresponding failure characteristic data. Constraint and supplementary analyses are performed between upper and lower level functional nodes according to the hierarchical order, enabling bidirectional transmission of failure characteristic data. Failure of an upper-level function creates a constraint effect on lower-level functions, while failure of a lower-level function has a supplementary impact on upper-level functions, thus establishing a complete function-failure mapping network. This allows the system to comprehensively identify and analyze the associated failure modes between functions at each level.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, after determining the risk level of the target product based on the action priority AP value and generating a risk assessment result, the method further includes: establishing a multidimensional correlation matrix between improvement measures and failure modes, wherein the row vectors of the multidimensional correlation matrix represent the influence path of the improvement measures, and the column vectors represent the characteristic attributes of the failure modes; performing similarity matching on high-risk items based on the multidimensional correlation matrix, and retrieving historical improvement schemes from a dynamic knowledge base; quantitatively evaluating the implementation effect of historical improvement schemes to obtain an effectiveness index of the improvement measures; optimizing and adjusting the current improvement measures based on the effectiveness index to generate an optimal improvement scheme; collecting implementation data of the optimal improvement scheme, updating the severity S value, occurrence O value, and detectability D value to obtain updated scoring data; and storing the updated scoring data and the corresponding improvement measure data into the dynamic knowledge base.

[0015] In the above embodiments, a multidimensional correlation matrix between improvement measures and failure modes is established. Row vectors represent the impact paths of improvement measures, and column vectors represent the characteristic attributes of failure modes. The system performs similarity matching on high-risk items based on the multidimensional correlation matrix, obtains historical improvement schemes, and performs quantitative evaluation. Improvement measures are optimized and adjusted according to the effectiveness index, and implementation data is collected to update the scoring, forming a complete closed-loop management mechanism for improvement measures. This makes the selection and optimization of improvement schemes more accurate and efficient.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after determining the risk level of the target product based on the action priority AP value and generating a risk assessment result, the method further includes: generating a data packet containing the initial failure chain, severity S value, occurrence O value, detectability D value, action priority AP value and risk assessment result.

[0017] In the above embodiments, In conjunction with some embodiments of the first aspect, in some embodiments, In the above embodiments, based on the initial failure chain, the correlation strength between failure modes is calculated. The correlation strength represents the trigger probability between different failure modes. Failure mode combinations with a correlation strength greater than a preset threshold are marked as combined risk items. Scenario simulation analysis is performed on the combined risk items. Based on the combined changes in severity (S), occurrence (O), and detectability (D) values, a risk propagation path diagram is generated. The risk propagation path diagram shows the diffusion pattern of failure risks among different functional nodes. According to the risk propagation path diagram, corresponding error-proofing design schemes and detection schemes are extracted from the dynamic knowledge base to generate a combined protection strategy. Data on the prevention and control effect of the combined protection strategy on the combined risk items is collected. The prevention and control effect data includes the changes in severity (S), occurrence (O), and detectability (D) values, as well as the attenuation value of the correlation strength. The combined protection strategy is optimized based on the prevention and control effect data, and the optimized combined protection strategy and the corresponding prevention and control effect data are stored in the dynamic knowledge base.

[0018] In a second aspect, embodiments of this application provide an intelligent risk optimization system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the intelligent risk optimization system to perform the method described in the first aspect and any possible implementation thereof.

[0019] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on an intelligent risk optimization system, cause the intelligent risk optimization system to perform the method described in the first aspect and any possible implementation thereof.

[0020] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on an intelligent risk optimization system, cause the intelligent risk optimization system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Understandably, the intelligent risk optimization system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0022] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. This application vectorizes product function descriptions using the BERT model, constructing a dynamic knowledge base containing failure data and improvement measures, thus achieving intelligent retrieval and matching of target product functions. Based on a three-dimensional logical matrix, conditional branching is performed on severity (S), occurrence (O), and detectability (D) values, outputting action priority (AP) values, thereby eliminating the numerical repetition problem caused by traditional RPN multiplication operations. When a failure mode has S=9 and O=3, even if its RPN value is low, it will be accurately identified as a high-risk item, effectively avoiding the problem of severe failure risks being masked by low-risk items.

[0023] 2. This application employs three independent conditional branching methods to determine the severity (S) value, occurrence (O) value, and detectability (D) value, respectively. These scores are then input into a three-dimensional logic matrix for comprehensive analysis. Based on the pre-defined correspondence between the S, O, and D values ​​in the matrix, the action priority (AP) value is determined, forming a complete risk assessment system. This multi-dimensional combined judgment method avoids the risk judgment bias that may result from simple numerical multiplication, making the risk assessment results more accurate and reliable.

[0024] 3. This application constructs a multi-level functional tree structure by hierarchically modeling historical failure data, with each functional node having corresponding failure characteristic data. Constraint and supplementary analyses are performed between upper and lower level functional nodes according to the hierarchical order, achieving bidirectional transmission of failure characteristic data. Failure of an upper-level function creates a constraint effect on lower-level functions, while failure of a lower-level function has a supplementary impact on upper-level functions, thus establishing a complete function-failure mapping network. This enables the system to comprehensively identify and analyze the associated failure modes between functions at each level. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating an intelligent risk optimization method based on the FMEA5 seven-step method in an embodiment of this application. Figure 2 This is another flowchart illustrating the intelligent risk optimization method based on the FMEA5 seven-step method in this application embodiment; Figure 3 This is a schematic diagram of the physical device structure of the intelligent risk optimization system in the embodiments of this application. Detailed Implementation

[0026] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0027] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0028] To facilitate understanding, the application scenarios of the embodiments of this application are described below.

[0029] In the development of a new energy vehicle power battery system, the technical team needs to conduct FMEA analysis on a battery pack containing thousands of cells. The system involves the interaction of multiple subsystems, including cell management, thermal management, and high-voltage control. Each subsystem has complex functional requirements and potential failure modes. For example, the cell management system needs to ensure a voltage sampling accuracy of ±0.1V, an equalization response time of ≤50ms, and an operating temperature range of -40℃ to 85℃. In traditional manual analysis, the team needs to identify failure modes for each functional point, often taking 3-4 weeks to complete the initial FMEA. More importantly, due to the lack of systematic reuse of historical failure data, similar design flaws often recur in new projects. After a certain mass production run, it was discovered that the cell equalization circuit had a drift risk when operating at low temperatures. This problem had also occurred in other previous projects, but it was not identified in time because the analysis records were scattered in different documents.

[0030] The technical team used spreadsheets to record the FMEA analysis process. First, the battery pack functions were broken down into modules such as voltage sampling, temperature monitoring, and equalization control. Then, based on experience, potential failure modes for each function were identified. For example, the voltage sampling function had failure modes including "excessive sampling value deviation," with a severity rating of S=8 (potentially leading to incorrect SOC estimation), an occurrence rating of O=6 (the sampling circuit is susceptible to temperature effects), and a detectability rating of D=5 (based on comparison with redundant sampling channels), resulting in an RPN of 240. Another failure mode, "equalization MOSFET short circuit," had ratings of S=9, O=3, D=4, and an RPN of 108. Due to the limitations of traditional RPN calculation methods, the higher-risk MOSFET short circuit was masked by the lower-scoring sampling deviation. Furthermore, improvement measures often remained at the textual description level, such as "optimizing sampling circuit temperature drift," lacking specific technical guidance and implementation effectiveness verification mechanisms.

[0031] After adopting the intelligent FMEA system of this invention, the battery pack functional description is first input into the BERT model for vectorization. The system automatically retrieves historical failure data of similar functions from the knowledge base and quickly generates an initial failure chain. For example, after inputting the "voltage sampling" function, the system matches relevant failure cases from past projects based on cosine similarity and automatically associates failure chains such as "reference source temperature drift → sampling value deviation → SOC error". For each failure chain, the system determines the AP value through a three-dimensional logic matrix, no longer using simple RPN multiplication. When a high-risk failure mode with S≥9 (such as MOS short circuit) is detected, it will be automatically marked as high priority even if the O value is low. The system will also recommend the technical solution with the highest matching degree from the improvement measure library, such as "dual reference source differential sampling circuit + temperature compensation algorithm", and track the implementation effect of the solution through a closed-loop verification mechanism, automatically updating the SOD score. In this way, the analysis efficiency is improved by more than 60%, the high-risk failure mode identification rate reaches 95%, and the effectiveness of improvement measures is also significantly improved.

[0032] Currently, improvements to FMEA analysis methods mainly include the following technical solutions: The RPN weighted iterative method introduces weighting coefficients into the traditional RPN (Severity S × Occurrence O × Detectability D) calculation, such as using RPN = S² × O × D, or adding a failure cost factor. However, this improved method only alleviates the numerical repeatability problem to a certain extent and cannot fundamentally solve the problem of high-risk missed detections. For example, high-risk failures with S=9 and O=3 may still be masked by medium-risk failures with S=5 and O=8. Furthermore, this method still relies on manual scoring and cannot automatically associate and utilize historical failure knowledge; its risk identification coverage is typically below 75%.

[0033] The QFD-FMEA integrated model attempts to enhance the completeness of functional analysis by systematically transforming customer requirements into input parameters for FMEA analysis through the introduction of the Quality Function Deployment (QFD) methodology. However, this approach significantly increases the workload of initial modeling, extending the analysis cycle by more than 30%. Furthermore, this approach still relies on traditional RPN calculations or manual AP determination during the risk assessment phase, lacking intelligent hierarchical engine support, making it difficult to achieve accurate risk identification and rapid assessment.

[0034] To facilitate understanding, the method provided in this implementation will be described in detail below, using the above scenario as an example. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating an intelligent risk optimization method based on the FMEA5 seven-step method in an embodiment of this application.

[0035] S101. Vectorize the product function description using the BERT model and store historical failure data and improvement measure data to obtain a dynamic knowledge base. The historical failure data includes failure causes, failure modes and failure effects, and the improvement measure data includes error-proofing design schemes and detection schemes.

[0036] Among them, the product function description represents a textual description of the functions that the product is expected to achieve, including information such as function name, performance indicators, and operating conditions; the BERT model refers to the bidirectional encoder-representer model, which is used to convert natural language text into numerical vectors; historical failure data represents information on recorded product failure cases, including three key elements: failure cause, failure mode, and failure impact; improvement measure data refers to the error-proofing design schemes and detection schemes adopted in response to historical failures; and the dynamic knowledge base represents a continuously updated data storage system used to manage the vectorized function descriptions, failure data, and improvement measure data.

[0037] This step is performed during the early stages of product development when FMEA analysis is required. Specifically, first, textual descriptions of product functions are collected, including functional specifications and performance requirements. Then, a pre-trained BERT model is used to process these textual descriptions, converting them into fixed-dimensional numerical vectors for subsequent similarity calculations. Simultaneously, the system imports failure data accumulated from historical projects, including identified failure causes (such as material fatigue, parameter deviations), failure modes (such as functional degradation, complete failure), and failure impacts (such as performance degradation, safety hazards). Furthermore, corresponding improvement measures data are collected, including error-proofing design schemes (such as redundant design, fault protection) and detection schemes (such as online monitoring, periodic inspections). Finally, this data is stored in a dynamic knowledge base for unified management.

[0038] In some embodiments, the vectorization of functional descriptions and the construction of a knowledge base can be achieved in the following ways: Optionally, the functional description text is first segmented and cleaned to remove stop words and special characters; then, the processed text is input into a pre-trained BERT model to obtain the average value of the last hidden state as a text vector; finally, the vector is associated with and stored with the corresponding failure data and measure data. Optionally, a structured template for the functional description is first constructed, including fields such as functional name, performance parameters, and environmental conditions; then, functional information is filled in based on the template, and each field is vectorized using the BERT model; finally, the field vectors are concatenated to form a complete functional vector representation. It is understood that other deep learning models or text representation methods can also be used to achieve the vectorization of functional descriptions, and this is not limited here.

[0039] S102. The target product function is searched and matched in the dynamic knowledge base to generate an initial failure chain, which includes interrelated failure causes, failure modes and failure effects.

[0040] Among them, the target product function refers to the specific product function that needs to be analyzed by FMEA; retrieval matching means searching for similar historical function cases in the dynamic knowledge base; the initial failure chain refers to a complete description of a failure scenario, including interrelated failure causes, failure modes and failure effects; interrelated means that there is a causal relationship between the elements in the failure chain.

[0041] This step is performed after the functional description vectorization and knowledge base construction are completed. Specifically, first, the input target product function is vectorized using the same BERT model; then, the cosine similarity between this vector and all historical function vectors in the knowledge base is calculated, and the highest number of historical records with the highest similarity are obtained; next, based on the failure data in these similar cases, the corresponding failure causes, failure modes, and failure effects are extracted; finally, these elements are organized into an initial failure chain according to causal relationships for subsequent risk assessment.

[0042] In some embodiments, similar function retrieval and failure chain generation can be achieved in the following ways: Optionally, firstly, a similarity threshold is set to filter historical cases with similarity exceeding the threshold; then, cluster analysis is performed on the failure data in these cases to identify the main failure mode types; finally, an initial failure chain is generated based on the frequency and impact of the failure modes. Optionally, firstly, an association rule base between failure elements is constructed; then, reasoning is performed on the retrieved historical failure data based on the rules to supplement potential failure scenarios; finally, all identified failure scenarios are integrated into a complete failure chain network. It is understood that other similarity calculation methods or failure chain construction strategies can also be used, and are not limited here.

[0043] In some embodiments, this step may further include: The BERT model is used to vectorize the input target product function description and historical failure data in the dynamic knowledge base. The cosine similarity between the target product function description vector and the historical failure data vector is calculated to obtain the historical failure data with the highest similarity. An initial failure chain is generated based on the historical failure data with the highest similarity.

[0044] S103. Determine the severity S value based on the failure impact in the initial failure chain, determine the occurrence O value based on the failure cause, and determine the detectability D value based on the detection plan. Input the severity S value, occurrence O value, and detectability D value into the three-dimensional logic matrix, and output the action priority AP value using a conditional branch decision method.

[0045] Among them, the severity S value represents the degree of harm that the failure affects to users or the system, with a value range of 1-10; the occurrence O value represents the probability of the failure cause occurring, with a value range of 1-10; the detectability D value represents the ease with which the failure can be detected through the detection plan, with a value range of 1-10; the three-dimensional logic matrix refers to a set of preset judgment rules containing the three dimensions of S, O, and D; the conditional branch judgment method represents the decision based on the combination of multiple logical conditions; and the action priority AP value refers to the quantitative assessment result of the failure risk level, which is divided into three levels: high, medium, and low.

[0046] This step is performed after the initial failure chain is generated. Specifically, first, the nature and scope of the failure impact are analyzed, and the severity S-value is determined according to a preset scoring standard. For example, the S-value for failure modes affecting safety is usually greater than 8. Then, the probability of occurrence of the failure cause is assessed, and the occurrence O-value is determined by considering factors such as technology maturity and environmental factors. Next, the detectability D-value is determined based on the reliability and timeliness of the existing detection scheme. Finally, these three values ​​are input into a pre-established three-dimensional logic matrix, and the final AP value is determined through a series of conditional judgments (such as directly determining high priority when S≥9 and O≥7), avoiding the risk of misjudgment that may be caused by traditional RPN multiplication.

[0047] In some embodiments, risk scoring and prioritization can be implemented in the following ways: Optionally, firstly, a scoring standard library is established, defining specific rules for determining S, O, and D values ​​for different industries and product types; then, a machine learning model is trained based on historical case data to achieve intelligent scoring; finally, an AP value is determined by combining multiple conditions using a decision tree algorithm. Optionally, firstly, a fuzzy rule library based on expert knowledge is constructed; then, fuzzy inference methods are used to comprehensively evaluate the S, O, and D values; finally, a clear AP value output is obtained through defuzzification. It is understood that other risk scoring methods or prioritization strategies can also be used, and are not limited here.

[0048] In some embodiments, this step may further include the following steps: The severity S value is determined based on the failure impact data using the first conditional branch decision method; the occurrence O value is determined based on the failure cause data using the second conditional branch decision method; the detectability D value is determined based on the detection plan data using the third conditional branch decision method; and the action priority AP value is determined based on the correspondence between the S value, O value, and D value in the three-dimensional logic matrix.

[0049] S104. Determine the risk level of the target product based on the action priority AP value and generate risk assessment results.

[0050] The risk level indicates the degree of harm that product failure may cause; the risk assessment result refers to an analysis report that includes the risk level, key failure modes, and recommended measures.

[0051] This step is executed after obtaining the action priority (AP) value. Specifically, first, the corresponding risk level is determined based on the high, medium, and low AP values, where high priority corresponds to high risk, medium priority to medium risk, and low priority to low risk. Then, for each risk level's failure mode, the system automatically generates a corresponding risk assessment report, including a failure mode description, the basis for risk level determination, an assessment of the effectiveness of current control measures, and improvement suggestions. Finally, the assessment results are fed back to the relevant teams to guide subsequent improvement activities.

[0052] In some embodiments, risk level determination and assessment report generation can be achieved through the following methods: Optionally, firstly, threshold ranges for AP values ​​are set for different risk levels; then, failure modes of the same risk level are categorized and summarized to identify common problems; finally, standardized assessment documents are automatically generated based on a report template. Optionally, firstly, a risk assessment knowledge graph is established, containing information such as risk types and impact paths; then, the propagation characteristics and key nodes of the current risk are analyzed based on the graph; finally, a visualized risk analysis report is generated. It is understood that other risk level classification methods or report generation methods can also be used, and are not limited here.

[0053] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 2 This is another flowchart illustrating the intelligent risk optimization method based on the FMEA5 seven-step method in this application embodiment.

[0054] S201. Perform hierarchical modeling on historical failure data to obtain a multi-level functional tree structure of product functions, where each functional node is set with corresponding failure characteristic data.

[0055] Hierarchical modeling refers to a modeling method that divides complex systems into different levels according to their functional hierarchy. A multi-level functional tree structure represents a hierarchical tree structure, decomposing the system functions from the top level down to the specific functions at the bottom level. Functional nodes are the basic units in the functional tree, containing functional descriptions and performance indicators. Failure characteristic data contains information such as the possible failure modes, failure mechanisms, and failure effects of the function. For example, in an electric vehicle powertrain system, the top-level function is "providing power output," which is divided into sub-functions such as motor control and battery management. Each sub-function node contains corresponding failure characteristic data.

[0056] This step begins by collecting and organizing historical failure data, establishing a multi-level functional tree based on the principle of functional decomposition. Starting with the overall system function, it is progressively decomposed into subsystem functions, component functions, and finally the most basic functional unit. For each functional node, its failure characteristic data is extracted and a corresponding relationship is established. Taking electric vehicles as an example, the powertrain function is decomposed into motor control function and battery management function, with the motor control function further decomposed into sub-functions such as speed control and torque control. Each functional node is associated with corresponding failure data; for example, the speed control function node includes failure characteristics such as "signal interference causing speed fluctuations" and "sensor failure causing reduced control accuracy." Through this hierarchical modeling method, a complete function-failure mapping relationship is established, providing basic data support for subsequent analysis.

[0057] S202. Map the target product functions to a multi-level function tree structure to obtain the corresponding nodes of the target product functions at each level.

[0058] Target product function mapping refers to matching the functions of the product to be analyzed with an established function tree; the corresponding node represents the matching function node found at each level of the function tree for the target function. For example, when analyzing the speed control function of a new type of motor controller, it is necessary to find the corresponding speed control function node and its related upper and lower level function nodes in the function tree.

[0059] This step uses functional similarity analysis to match the target product's functions with nodes in the function tree. First, the target functions are hierarchically decomposed to clarify the specific content and performance indicators of each level. Then, similar functional nodes are searched in the function tree, and mapping is performed based on the semantic similarity of the functional descriptions and the matching degree of the performance indicators. For each target function, a matching node at the corresponding level in the function tree needs to be found. Taking a motor controller as an example, its speed control function will be mapped to the speed control node in the function tree, and it will also be associated with the upper-level motor control function node and the lower-level signal acquisition, algorithm processing, and other function nodes, forming a complete functional mapping relationship.

[0060] S203. Following the hierarchical order of the multi-level functional tree structure, constrain the failure characteristic data of the lower-level functional nodes based on the failure characteristic data of the upper-level functional nodes to obtain the associated failure modes of each level of functional nodes.

[0061] Constraint analysis is an analytical method for studying the failure propagation relationship between upper and lower functional nodes; associated failure modes represent cascading failures caused by the mutual influence between functions. For example, a failure in speed signal acquisition in a motor controller can affect the implementation of the upper-level speed control function, thereby affecting the performance of the entire power system.

[0062] This step analyzes the failure impact relationships between functional nodes at each level based on the hierarchical structure of the functional tree. It analyzes the constraint effect of upper-level functional failures on lower-level functions from top to bottom, and the impact of lower-level functional failures on upper-level functions from bottom to top. Through the transmission analysis of failure characteristic data, it identifies failure association patterns between functions. Taking a motor control system as an example, the accuracy requirements of the upper-level speed control function impose sampling accuracy constraints on the lower-level signal acquisition function. When the upper-level function requires speed control accuracy ≤1%, the sampling error of the lower-level signal acquisition function must be controlled within 0.1%. Simultaneously, failures in the lower-level signal acquisition function (such as sensor drift) will cause the upper-level speed control function to deviate from its target value. Through this constraint analysis, a complete functional failure association network is established.

[0063] S204. Following the hierarchical order of the multi-level functional tree structure, the failure characteristic data of the lower-level functional nodes are used to supplement the failure characteristic data of the upper-level functional nodes to obtain the combined failure impact of functional nodes at each level.

[0064] Supplementary analysis refers to the process of analyzing the impact of lower-level functional failures on upper-level functions; combined failure impact refers to the system-level impact caused by the combination of failures of multiple functional nodes. For example, in a battery management system, the combination of temperature sensor failure and voltage sampling failure can lead to more serious system-level faults, such as battery overcharging or over-discharging.

[0065] This step first identifies the failure characteristic data of each lower-level functional node and analyzes its impact on upper-level functions in terms of mode and extent. The impact of lower-level functional failures is accumulated to upper-level functional nodes through a hierarchical upward transmission method. For functional nodes with interactive effects, their combined failure effects also need to be analyzed. Taking the battery management system as an example, when the voltage sampling function has a 2% sampling error and the temperature detection function has a 5°C measurement deviation, the combination of the two can lead to a state of charge (SOC) estimation error exceeding 10%, triggering abnormal battery pack charge and discharge control. Through this supplementary analysis, a complete failure impact transmission chain is formed, providing a basis for subsequent risk assessment.

[0066] S205. Determine the initial failure chain based on associated failure modes and combined failure effects.

[0067] The initial failure chain refers to the complete causal path from the cause of failure to the final effect; associated failure modes and combined failure effects are two key inputs for determining the failure chain. For example, in the power system of new energy vehicles, a typical failure chain is "sensor drift → signal distortion → control deviation → system performance degradation".

[0068] This step, based on the previously analyzed associated failure modes and combined failure effects, constructs a complete failure causal chain. First, the basic failure modes of each functional node are identified. Then, according to the interrelationships between functions, these failure modes are connected according to their occurrence sequence and impact path. For each identified combined failure effect, its triggering conditions and propagation path are traced to form a corresponding failure chain. Taking electric vehicles as an example, when a system-level failure of "power output interruption" is discovered, analysis may trace multiple failure chains, such as "motor temperature sensor failure → overheat protection false triggering → motor controller shutdown → power output interruption," thus establishing a complete failure analysis framework.

[0069] S206. Establish a multidimensional correlation matrix between improvement measures and failure modes. The row vectors of the multidimensional correlation matrix represent the influence paths of the improvement measures, and the column vectors represent the characteristic attributes of the failure modes.

[0070] A multidimensional correlation matrix is ​​a data structure used to represent the relationship between improvement measures and failure modes. The influence path represents the specific way and process by which the improvement measures act on the system; the characteristic attributes represent the key characteristic parameters of the failure modes. For example, in a motor control system, the influence path of the improvement measure "dual sensor redundancy design" includes "improving signal reliability → reducing signal failure probability → improving control accuracy".

[0071] This step establishes a matrix structure for systematically managing the relationship between improvement measures and failure modes. The row vectors of the matrix describe the mechanism of action of each improvement measure, including the transmission paths of its direct and indirect effects. The column vectors describe the characteristics of the failure modes, such as failure mechanism, scope of influence, and occurrence conditions. Taking a motor controller as an example, a typical cell in the matrix might represent the improvement effect of the "oversampling filtering algorithm" (row) on "signal interference" (column), where the influence path is "increasing sampling points → improving signal averaging processing → reducing the impact of random interference," and the characteristic attributes of the failure mode include parameters such as "interference frequency range, amplitude distribution, and duration." Through this matrix-based management, a quantitative basis for selecting improvement measures is established.

[0072] S207. Based on the multidimensional association matrix, perform similarity matching on high-risk items and retrieve historical improvement schemes from the dynamic knowledge base.

[0073] High-risk items refer to failure modes that require priority handling after risk assessment; similarity matching refers to calculating the similarity between the current failure mode and historical cases; historical improvement solutions refer to solutions that have been validated in past projects. For example, in a battery management system, when a high-risk failure mode such as "cell temperature runaway" is detected, the system will search for solutions to similar thermal runaway problems in the past.

[0074] This step utilizes a multidimensional correlation matrix to perform feature matching on high-risk failure items. First, the feature vectors of high-risk items are extracted, including attributes such as failure mechanism, impact range, and occurrence conditions. Then, the cosine similarity between this feature vector and historical failure modes in the matrix is ​​calculated, identifying cases with similarity exceeding a preset threshold. Improvement solutions corresponding to these similar cases are then retrieved and ranked. Taking a power battery system as an example, when the risk of "cell temperature runaway" is identified, the system searches for similar cases based on its characteristics (such as thermal runaway trigger temperature, propagation rate, and thermal coupling degree) and returns validated improvement solutions, such as "multiple temperature monitoring + active heat dissipation control + thermal diffusion blocking" technologies.

[0075] S208. Quantitatively evaluate the implementation effect of historical improvement plans to obtain the effectiveness index of improvement measures.

[0076] Implementation effect refers to the degree of improvement achieved by the improvement plan in practical application; quantitative evaluation refers to measuring the improvement effect through specific indicators; effectiveness index is a numerical expression of the effect of improvement measures. For example, if a heat dissipation optimization plan reduces the system's maximum temperature by 15°C and slows down the heat diffusion rate by 50%, its thermal management improvement effectiveness index can be calculated.

[0077] This step involves quantitative analysis of implementation data from historical improvement schemes. First, an evaluation index system is established, including direct indicators (such as the degree of failure rate reduction) and indirect indicators (such as changes in maintenance costs). Then, comparative data before and after implementation are collected, and the improvement rate of each indicator is calculated. Finally, a weighted average is used to derive the overall effectiveness index. Taking a motor control system as an example, the evaluation indicators for a certain anti-interference optimization scheme include: a 30% improvement in signal stability, a 40% improvement in control accuracy, and a 25% improvement in system reliability. Through weight allocation (0.4, 0.4, 0.2), an effectiveness index of 0.85 is calculated. This quantitative evaluation provides an objective basis for scheme selection.

[0078] S209. Optimize and adjust the current improvement measures based on the effectiveness index to generate the optimal improvement plan.

[0079] Optimization and adjustment refer to refining an improvement plan based on the results of an effectiveness evaluation; the optimal improvement plan refers to the final plan determined after comprehensively considering effectiveness and feasibility. For example, in a motor control system, a "single filtering algorithm" may be optimized into an "adaptive multi-level filtering scheme" to achieve better anti-interference performance.

[0080] This step involves systematically optimizing the improvement scheme based on the effectiveness index. First, an optimization objective function is established, using the effectiveness index as the primary evaluation criterion, while also considering constraints such as implementation cost and technical difficulty. Then, existing schemes are optimized in terms of parameters or structural improvements, generating multiple alternative schemes. Finally, the scheme with the highest comprehensive score is selected as the final scheme through evaluation using the objective function. Taking a battery management system as an example, the optimization process of a temperature control scheme includes: adjusting the sensor placement density (from uniform distribution to concentrated hotspot placement), improving the control strategy (from fixed threshold to dynamic prediction), and introducing machine learning algorithms to improve temperature prediction accuracy, ultimately resulting in an optimized scheme with a 20% improvement in the effectiveness index.

[0081] S210. Collect implementation data of the optimal improvement plan, update the severity S value, occurrence O value and detectability D value, and obtain the updated score data.

[0082] Implementation data refers to the performance indicators and monitoring data generated during the actual application of the improved scheme; updated scoring data represents the S, O, and D values ​​re-evaluated after the improvements. For example, after implementing a dual-sensor redundancy design, the failure occurrence O value of a motor control system decreased from 7 to 3, and the detection D value increased from 6 to 2.

[0083] This step updates the risk score through data collection and analysis. First, data collection points are established to gather key parameters during the implementation of the improvement plan, including failure frequency, fault detection rate, and impact severity. Then, based on the measured data, the severity S-value is reassessed, a new occurrence O-value is determined through statistical analysis, and the detection D-value is updated based on the fault detection effect. Taking a battery management system as an example, data after implementing a temperature monitoring optimization plan shows that: a 15°C reduction in the system's maximum temperature causes the severity S-value to drop from 9 to 6; a 90% reduction in the probability of temperature runaway causes the occurrence O-value to drop from 8 to 3; and an increase in the accuracy of temperature anomaly warnings to 98% causes the detection D-value to drop from 7 to 2. This data-driven scoring update ensures the accuracy of the risk assessment.

[0084] S211. Store the updated scoring data and the corresponding improvement measures data into the dynamic knowledge base.

[0085] The scoring data refers to the updated S, O, D values ​​and their evolution; the improvement measure data includes information such as specific implementation plans, implementation effects, and resource input. For example, after the protection design of a motor controller is improved, the changes in its S, O, D values ​​and the corresponding technical solution details are recorded.

[0086] This step stores the improved scoring data and measures information into a knowledge base. First, the scoring data is organized, including a comparison of S, O, and D values ​​before and after the improvement, and an analysis of changing trends. Then, detailed information on the improvement measures is collected, including technical principles, implementation steps, cost input, and maintenance requirements. Finally, this data is stored in a dynamic knowledge base according to a unified format, establishing relationships between the data. Taking a battery management system as an example, the stored data includes: process data of the temperature control optimization scheme reducing the S value from 9 to 6, technical details of the multi-sensor deployment scheme, specific parameters of the prediction algorithm, and complete information such as an implementation cost of 12,000 yuan. This data will be used for analysis and decision-making on similar issues in the future.

[0087] S212. Generate a data packet containing the initial failure chain, severity S value, occurrence O value, detectability D value, action priority AP value, and risk assessment results.

[0088] A data packet refers to a collection of multiple related data points organized according to a predetermined format. The initial failure chain comprises a causal link of three elements: cause, mode, and impact. The severity (S) value indicates the degree of harm caused by the failure, graded from 1 to 10. The occurrence (O) value indicates the frequency of the failure, graded from 1 to 10. The detectability (D) value indicates the ease with which the failure is detected, graded from 1 to 10. The action priority (AP) value is a risk level indicator calculated using a three-dimensional logical matrix. The risk assessment results include a risk level determination and specific improvement recommendations.

[0089] This step first constructs a data packet template in JSON format, containing six key fields: failureChain, severityValue, occurrenceValue, detectionValue, actionPriority, and riskResult. Initial failure chain data is stored in the failureChain field, using a tree structure to record the hierarchical relationship of failure causes, patterns, and impacts. Severity (S), occurrence (O), and detection (D) values ​​are stored in their respective fields. Action priority (AP) values ​​are stored in the actionPriority field, with high, medium, and low risks represented by H, M, and L, respectively. The risk assessment results are stored in the riskResult field, containing both the risk level and improvement recommendations. Finally, all fields are packaged to generate a complete JSON data packet.

[0090] S213. Calculate the hash value of the data packet, write the hash value and timestamp into the blockchain distributed ledger, and obtain the blockchain evidence data of the risk assessment process.

[0091] A hash value maps data of arbitrary length to a fixed-length string using a hash algorithm. A timestamp is a digital identifier that records when data was generated. A blockchain distributed ledger is a decentralized data storage structure that uses cryptographic methods to ensure data immutability. Blockchain-based evidence refers to legally valid electronic evidence recorded on the blockchain.

[0092] This step first uses the SHA256 hash algorithm to calculate a 256-bit hash value for the data packet. The current UTC time is obtained as the timestamp, accurate to the millisecond. A transaction data structure is constructed, including fields such as the hash value, timestamp, and data packet ID. A suitable blockchain network (such as Hyperledger Fabric) is selected, and a smart contract is invoked to write the transaction data into a block. The blockchain network automatically completes consensus verification, permanently recording the data in the distributed ledger. The data packet's provenance record can be queried through a block explorer, including information such as the transaction hash, timestamp, and block height. Ultimately, a complete chain of evidence for the risk assessment process is formed, ensuring the authenticity and integrity of the assessment data.

[0093] In some embodiments, after determining the risk level of the target product based on the action priority AP value and generating a risk assessment result, the following steps may also be included: Based on the initial failure chain, the correlation strength between failure modes is calculated. The correlation strength represents the trigger probability between different failure modes. Failure mode combinations with a correlation strength greater than a preset threshold are marked as combined risk items.

[0094] Specifically, the initial failure chain refers to a complete causal chain that includes the cause of failure, the failure mode, and the impact of failure. Association strength refers to the probability that one failure mode triggering another failure mode, ranging from 0 to 1. Trigger probability represents the likelihood that one event will cause another event under specific conditions. Combined risk refers to a combination of risks where there is a significant correlation between multiple failure modes. The preset threshold is a probability critical value determined based on historical data and expert experience.

[0095] This step first constructs a failure mode correlation matrix R[i][j], with an n×n dimension, where n is the total number of failure modes. By analyzing historical failure data, the number of times failure mode i occurs alone (count(i)) and the number of times it occurs together with failure mode j (count(i,j)) are counted. The correlation strength is calculated using the conditional probability formula: R[i][j] = count(i,j) / count(i). For example, in a battery system, if a temperature sensor fails 100 times, and 30 of these failures cause voltage sampling failures, the correlation strength is 0.3. A correlation strength threshold is set (e.g., 0.5), and the matrix R is iterated through, marking failure mode pairs (i,j) that satisfy R[i][j] > 0.5 as combined risk items. Finally, a list of combined risk items containing all highly correlated failure mode pairs is output.

[0096] Scenario simulation analysis was conducted on the combined risk items. Based on the combined changes of severity (S), occurrence (O), and detectability (D) values, a risk propagation path diagram was generated, which showed the diffusion pattern of failure risk among different functional nodes.

[0097] Specifically, scenario simulation analysis refers to simulating the propagation and diffusion of failure risks within a system through numerical simulation. Combined changes refer to the dynamic changes of multiple scoring indicators over time. A risk propagation path diagram is a directed graph structure used to illustrate the transmission relationship of risk between functional nodes. Diffusion patterns represent the temporal characteristics and impact of risk propagation from one functional node to related nodes.

[0098] This step employs a Markov chain model for risk propagation simulation. The risk state vector of each functional node is defined as [S, O, D]. A state transition matrix P is constructed based on the association strength matrix R. For each time step t, the state of node i at time t+1 is calculated: State[i, t+1] = P × State[i, t]. For example, if a node's initial state is [8, 3, 4], after the transition it becomes [9, 5, 4], indicating an increased risk level. A directed graph G(V, E) is used to represent the propagation path, where the vertex set V represents the functional nodes, and the edge set E represents the risk propagation relationships between nodes. The edge weights are equal to the association strength, and the node size is proportional to the risk level. Using a breadth-first search algorithm, starting from the initial failed node, the entire graph is traversed according to the edge weights, ultimately generating a complete risk propagation path graph.

[0099] Based on the risk propagation path diagram, relevant error prevention design schemes and detection schemes are extracted from the dynamic knowledge base to generate a combined protection strategy.

[0100] Specifically, error-proofing design schemes refer to technical solutions that prevent or reduce failures through optimized design. Detection schemes refer to technical solutions that detect failures promptly through monitoring and diagnosis. Combined protection strategies refer to overall schemes that coordinate multiple protection schemes to achieve systematic prevention and control.

[0101] This step first converts the risk propagation path graph into an adjacency matrix. The in-degree (number of affected edges) and out-degree (number of edges affecting other nodes) of each node are calculated. Based on the node's degree characteristics, matching protection schemes are retrieved from a dynamic knowledge base. For nodes with high in-degree (those susceptible to other failures), monitoring and diagnostic schemes are deployed; for nodes with high out-degree (those prone to triggering other failures), prevention and control schemes are deployed. For example, in a battery management system, the temperature sensor node has a high out-degree, so a dual-sensor redundancy design and a temperature anomaly early warning scheme are deployed. The voltage sampling node has a high in-degree, so an online calibration and data verification scheme is deployed. Finally, the protection schemes for each node are integrated into a complete combined protection strategy, clarifying the implementation sequence and configuration requirements of each scheme.

[0102] Data on the prevention and control effects of combined protection strategies on combined risk items are collected. The prevention and control effect data includes the changes in severity (S value), occurrence (O value), detectability (D value), and the attenuation value of association strength.

[0103] Specifically, prevention and control effectiveness data refers to the changes in various evaluation indicators after the implementation of protective strategies. The change represents the difference in indicator values ​​before and after protection. The attenuation value indicates the degree to which the correlation strength decreases after the implementation of protective measures.

[0104] This step involves collecting data on the effectiveness of prevention and control measures through online monitoring and regular evaluation. For each combined risk item, the S, O, and D values ​​before and after the implementation of the protection strategy are recorded, and the changes ΔS, ΔO, and ΔD are calculated. For example, after implementing a dual-sensor redundancy design for a certain failure mode, the O value drops from 8 to 3, so ΔO = -5. Simultaneously, the change in correlation strength is measured; for example, the probability of a temperature sensor failure triggering voltage sampling failure drops from 0.3 to 0.1, with an attenuation value of -0.2. A data collection table is established, including fields such as risk item ID, protection scheme ID, score change, and correlation strength attenuation value. A complete dataset of prevention and control effectiveness is formed through a combination of automatic data collection and manual entry.

[0105] The combined protection strategy is optimized based on the prevention and control effect data, and the optimized combined protection strategy and the corresponding prevention and control effect data are stored in the dynamic knowledge base.

[0106] Specifically, optimization refers to adjusting and improving protection strategies based on actual prevention and control results. A dynamic knowledge base is a continuously updated data storage system used to accumulate and manage protection experience.

[0107] This step first calculates the comprehensive effectiveness index of the protection strategy: E = w1×ΔS + w2×ΔO + w3×ΔD + w4×attenuation value, where w1-w4 are weighting coefficients. When the E value is lower than the expected target, the strategy needs to be optimized. Optimization methods include increasing the coverage of protection measures, increasing the detection frequency, and optimizing control parameters. For example, if the detectability improvement of a certain protection scheme is not significant (ΔD is small), the number of detection points should be increased or the detection algorithm should be updated. After optimization, the specific parameter configuration, implementation requirements, maintenance specifications, and other information of the protection strategy, along with the corresponding prevention and control effect data, are stored in a dynamic knowledge base. A key-value pair storage structure is used to ensure rapid data retrieval and updates. In this way, protection experience is continuously accumulated and optimized, providing a reference for the prevention and control of similar risks in the future.

[0108] To verify the practical application effect of this invention, a DFMEA analysis project of the power battery pack of a new energy vehicle company was selected for comparative verification. The traditional FMEA method and the method of this invention were used for analysis, and the specific results are compared below: When using the traditional FMEA analysis process, a complete analysis required 32 people and identified 11 critical failures. However, post-analysis revealed that 3 potential failures were not identified in a timely manner. Regarding the effectiveness of the improvement measures, 6 out of the 11 measures achieved their intended objectives, resulting in an effectiveness rate of 54.5%. This reflects that there is still room for improvement in the efficiency and accuracy of the traditional method.

[0109] After adopting the intelligent FMEA analysis process of this invention, the system operates as follows: First, in the structural analysis stage, the system automatically decomposes the battery pack into functional units such as cell modules, BMS control unit, and external casing, establishing a structural hierarchy. In the functional analysis stage, when the core functional requirement "voltage regulation" is input, the system correlates it with the specific technical indicator "±0.1V accuracy". Entering the failure chain generation stage, a failure chain is generated based on the knowledge base: "BMS chip overheating → voltage sampling drift → vehicle power failure protection triggering".

[0110] During the AP intelligent classification phase, the system assesses a critical failure mode with a severity S=10, occurrence O=6, and detectability D=4. Traditional RPN calculations yield a score of 240, classifying it as medium risk, while the AP classification engine of this invention identifies it as a high-risk item requiring priority handling. Subsequently, in the action recommendation phase, the system matches and pushes an improvement plan from the knowledge base: "Adopt a dual-redundant temperature sensor design, combined with a dynamic current limiting algorithm." In the closed-loop verification phase, after implementing the improvement measures, the detectability D value increases to 2, and the system accordingly adjusts the risk level to medium.

[0111] After adopting the method of this invention, the analysis time was reduced to 12 person-hours, improving analysis efficiency; the identification of key failures was more comprehensive, and the implementation effect of improvement measures was significantly improved, with a significant reduction in the frequency of related failures during the 3-month follow-up period. This case verifies the technical advantages of this invention in practical applications and provides a new solution for reliability analysis of products such as new energy vehicles.

[0112] Based on the above embodiments, the technical features and innovations of the present invention are further elaborated. The present invention, by establishing an intelligent FMEA analysis system, realizes the transformation from traditional static analysis to dynamic intelligent decision-making. Its specific innovations are reflected in the following key aspects: First, this invention proposes a dynamic knowledge base-driven failure closed-loop management system. In the structural analysis (step 2) and functional analysis (step 3) stages, it pioneers a function-requirement-failure dynamic mapping model based on a parametric diagram (P-Diagram). This model automatically generates failure chains (failure cause → mode → impact) by associating with a historical failure database, replacing reliance on manual experience. This innovative approach makes the failure analysis process more systematic and intelligent, significantly improving analysis efficiency.

[0113] Secondly, this invention develops an innovative AP intelligent grading engine. In the risk analysis (step 5) stage, a multi-dimensional combined judgment algorithm (non-RPN multiplication) is used to match the combination of severity (S), occurrence (O), and detection (D) with a preset logical matrix in real time, outputting high / medium / low action priority (AP), eliminating the numerical repetition blind spot of traditional RPN. This innovative risk assessment method effectively avoids the risk judgment bias caused by simple numerical multiplication in traditional FMEA.

[0114] Third, this invention establishes a cross-functional collaborative optimization mechanism. For high AP risk (step 6), the system automatically pushes improvement measure knowledge bases (such as error-proofing design libraries and detection scheme libraries), tracks the effectiveness of the measures in real time, and updates S / O / D scores synchronously, forming an "analysis-decision-verification" closed loop. Through this intelligent collaborative mechanism, the implementation effect of improvement measures is significantly improved.

[0115] Fourth, this invention innovatively introduces a blockchain-enabled documented traceability system. In the results documentation stage (step 7), distributed ledger technology is used to solidify the analysis process and measures, ensuring that the data is tamper-proof and auditable. This innovative documented management method provides a reliable traceability foundation for failure analysis.

[0116] Through the aforementioned technological innovations, this invention transforms the FMEA five-step method from a static document into an intelligent decision-making system, improving risk identification efficiency by over 40% and increasing high-hazard failure coverage from 60% to 90%. Specifically, this invention has the following significant advantages: This invention replaces RPN multiplication with an AP logic matrix, eliminating numerical repetitive distortion problems (such as misjudgments of S9-O3-D4=108 and S5-O8-D3=120), ensuring 100% identification of high-risk failures without omission. Simultaneously, through dynamic failure chain automatic generation technology, it utilizes a historical knowledge base to automatically link requirements and failures, replacing manual experience-based analysis and reducing analysis time by 50%–70%. Furthermore, the closed-loop management mechanism of the intelligent measures library can automatically match error-proofing designs and detection schemes and verify their effectiveness, increasing the compliance rate of measures to over 80%.

[0117] The AP engine of this invention can capture 100% of fatal failures (S≥9 and O≥7), increasing the early identification rate of high-hazard failures from 65% to 92%. Regarding resource allocation, the system focuses only on 10%~15% of AP=H level critical risks (traditional RPN>100 includes 40% redundancy), reducing engineering resource waste by 60%. For example, after applying this system, a car manufacturer reduced its stamping scrap rate from 3.1% to 0.4%. In terms of knowledge asset management, the knowledge graph constructed by the system improves the efficiency of FMEA construction for new projects by 90%, while closed-loop data feedback enables the system to self-learn, automatically reassessing O / D values ​​when measures are ineffective.

[0118] This invention possesses significant technical scalability. By real-time access to IoT data (e.g., automatic update of O values ​​for sensor anomalies), it achieves second-level response capabilities for production line risks. Simultaneously, the use of NLP models to automatically generate potential failure descriptions improves analysis completeness by 38% (already verified in an electronics factory). These scalable technologies provide a solid foundation for the continued development of this invention in the field of smart manufacturing.

[0119] The intelligent risk optimization system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference]. Figure 3 This is a schematic diagram of the physical device structure of the intelligent risk optimization system in the embodiments of this application.

[0120] It should be noted that, Figure 3 The structure of the intelligent risk optimization system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0121] like Figure 3As shown, the intelligent risk optimization system includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 302 or programs loaded from storage section 308 into Random Access Memory (RAM) 303, such as performing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.

[0122] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0123] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the various functions defined in the present invention.

[0124] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0125] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0126] Specifically, the intelligent risk optimization system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the intelligent risk optimization method based on the FMEA5 seven-step method provided in the above embodiment.

[0127] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the intelligent risk optimization system described in the above embodiments; or it may exist independently and not assembled into the intelligent risk optimization system. The storage medium carries one or more computer programs, which, when executed by a processor of the intelligent risk optimization system, cause the intelligent risk optimization system to implement the intelligent risk optimization method based on the FMEA5 seven-step method provided in the above embodiments.

[0128] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0129] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0130] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A smart risk optimization method based on the FMEA5 seven-step method, characterized in that, The method, applied to an intelligent risk optimization system, includes: The product function description is vectorized using the BERT model, and historical failure data and improvement measure data are stored to obtain a dynamic knowledge base. The historical failure data includes failure causes, failure modes and failure effects, and the improvement measure data includes error-proofing design schemes and detection schemes. The input target product function is searched and matched in the dynamic knowledge base to generate an initial failure chain, which includes interrelated failure causes, failure modes and failure effects. The severity S value is determined based on the failure impact in the initial failure chain, the occurrence O value is determined based on the failure cause, and the detectability D value is determined based on the detection scheme. The severity S value, the occurrence O value, and the detectability D value are input into a three-dimensional logic matrix, and the action priority AP value is output using a conditional branch decision method. The risk level of the target product is determined based on the action priority AP value, and a risk assessment result is generated.

2. The method according to claim 1, characterized in that, The input target product function is retrieved and matched in the dynamic knowledge base to generate an initial failure chain. The initial failure chain includes interrelated failure causes, failure modes, and failure effects, specifically including: The BERT model is used to vectorize the input target product function description and the historical failure data in the dynamic knowledge base, respectively. Calculate the cosine similarity between the description vector of the target product function and the historical failure data vector to obtain the historical failure data with the highest similarity; The initial failure chain is generated based on the historical failure data with the highest similarity.

3. The method according to claim 1, characterized in that, The process involves determining a severity S value based on the failure impact in the initial failure chain, an occurrence O value based on the failure cause, and a detectability D value based on the detection scheme. The S, O, and D values ​​are then input into a three-dimensional logic matrix, and an action priority AP value is output using a conditional branching decision method. Specifically, this includes: The severity S value is determined based on the failure impact data using a first conditional branch decision method. The occurrence degree O value is determined based on the failure cause data using a second conditional branch decision method. The detectivity D value is determined based on the detection scheme data using a third conditional branch decision method. The action priority AP value is determined based on the correspondence between the S, O, and D values ​​in the three-dimensional logic matrix.

4. The method according to claim 1, characterized in that, After the steps of vectorizing product function descriptions using the BERT model and storing historical failure data and improvement measure data to obtain a dynamic knowledge base, the method further includes: The historical failure data is hierarchically modeled to obtain a multi-level functional tree structure of the product function, wherein each functional node is set with corresponding failure feature data. The target product function is mapped to the multi-level function tree structure to obtain the corresponding node of the target product function at each level; According to the hierarchical order of the multi-level functional tree structure, the failure characteristic data of the lower-level functional nodes are constrained based on the failure characteristic data of the upper-level functional nodes to obtain the associated failure modes of each level of functional nodes. According to the hierarchical order of the multi-level functional tree structure, the failure characteristic data of the lower-level functional nodes are used to supplement the failure characteristic data of the upper-level functional nodes to obtain the combined failure impact of functional nodes at each level. The initial failure chain is determined based on the associated failure modes and the combined failure effects.

5. The method according to claim 1, characterized in that, After the step of determining the risk level of the target product based on the action priority AP value and generating a risk assessment result, the method further includes: Establish a multidimensional correlation matrix between improvement measures and failure modes. The row vectors of the multidimensional correlation matrix represent the influence paths of the improvement measures, and the column vectors represent the characteristic attributes of the failure modes. Based on the multidimensional association matrix, similarity matching is performed on high-risk items, and historical improvement schemes are retrieved from the dynamic knowledge base; The effectiveness of the historical improvement plans was quantitatively evaluated to obtain an effectiveness index of the improvement measures; The current improvement measures are optimized and adjusted based on the effectiveness index to generate the optimal improvement plan; Collect implementation data of the optimal improvement scheme, update the severity S value, the occurrence O value and the detectability D value to obtain the updated scoring data; The updated rating data and the corresponding improvement measures data are stored in the dynamic knowledge base.

6. The method according to claim 1, characterized in that, After the step of determining the risk level of the target product based on the action priority AP value and generating a risk assessment result, the method further includes: Generate a data packet containing the initial failure chain, the severity S value, the occurrence O value, the detectability D value, the action priority AP value, and the risk assessment result; The hash value of the data packet is calculated, and the hash value and timestamp are written into the blockchain distributed ledger to obtain the blockchain evidence data for the risk assessment process.

7. The method according to claim 1, characterized in that, After the step of determining the risk level of the target product based on the action priority AP value and generating a risk assessment result, the method further includes: Based on the initial failure chain, the correlation strength between the failure modes is calculated. The correlation strength represents the trigger probability between different failure modes. Failure mode combinations with a correlation strength greater than a preset threshold are marked as combined risk items. A scenario simulation analysis is performed on the combined risk items. Based on the combined changes of the severity S value, the occurrence O value, and the detectability D value, a risk propagation path diagram is generated. The risk propagation path diagram shows the diffusion pattern of failure risk among different functional nodes. Based on the risk propagation path diagram, corresponding error prevention design schemes and detection schemes are extracted from the dynamic knowledge base to generate a combined protection strategy. Collect data on the prevention and control effect of the combined protection strategy on the combined risk item, wherein the prevention and control effect data includes the change in the severity S value, the occurrence O value, the detectability D value, and the attenuation value of the association strength; The combined protection strategy is optimized based on the prevention and control effect data, and the optimized combined protection strategy and the corresponding prevention and control effect data are stored in the dynamic knowledge base.

8. An intelligent risk optimization system, characterized in that, The intelligent risk optimization system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the intelligent risk optimization system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the intelligent risk optimization system, the intelligent risk optimization system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the intelligent risk optimization system, the intelligent risk optimization system performs the method as described in any one of claims 1-7.

Images