A quality infrastructure evaluation model construction method and system

Abstract

The application discloses a quality infrastructure evaluation model construction method and system, comprising collecting metrological calibration data and authentication process data of a preset quality infrastructure, and preprocessing the metrological calibration data and the authentication data; performing digital node self-checking on the metrological calibration data and the authentication process data according to authentication process digital twinning, obtaining node checking data, performing quantitative traceability analysis on the node checking data by adopting process calibration linkage, and obtaining linkage analysis data; performing service quality analysis on the node checking data to obtain quality data, constructing a metrological calibration authentication coefficient according to the linkage analysis data and the quality data; constructing a quality infrastructure evaluation model based on a double-cycle index system according to the metrological calibration authentication coefficient, optimizing the quality infrastructure evaluation model by adopting a recall risk coefficient, and outputting a target model.

Description

Technical Field

[0001] This invention relates to the field of automotive component evaluation, and more particularly to a method and system for constructing a quality infrastructure evaluation model. Background Technology

[0002] In the automotive parts manufacturing sector, quality infrastructure is the core support for ensuring product consistency and safety. It encompasses key aspects such as metrology and calibration, and certification processes, directly impacting production efficiency and product compliance. Currently, the evaluation of automotive parts quality infrastructure suffers from significant technical shortcomings: On the one hand, evaluations often rely on single-dimensional data (such as focusing only on calibration results or certification cycles), failing to achieve integrated analysis of metrology and calibration processes and certification processes. This results in the inability to identify quality risks arising from the combined effects of these two processes—for example, product recall risks caused by calibration deviations and process breakpoints. On the other hand, data processing lacks sophisticated verification mechanisms; systemic biases and environmental fluctuations easily lead to data distortion, and differentiated verification strategies are not developed for nodes of varying importance, resulting in wasted resources or missed detection of critical risks.

[0003] Meanwhile, existing evaluation models lack dynamic optimization capabilities: most models use fixed indicator weights, making it impossible to adjust the evaluation logic based on actual quality results such as recall risks, thus failing to adapt to the dynamic needs of the automotive industry for quality control; furthermore, the evaluation indicators are not sufficiently aligned with industry standards such as the International Automotive Task Force (IATF) quality management system, and compliance gaps are not clearly quantified, resulting in limited guiding value of evaluation results for actual production improvements. These problems lead to low accuracy and delayed response in quality infrastructure evaluation, failing to meet the stringent requirements of high quality and low risk in automotive component manufacturing. Therefore, there is an urgent need to construct an evaluation model that integrates multi-dimensional data and possesses dynamic optimization capabilities. Summary of the Invention

[0004] The purpose of this invention is to provide a method for constructing a quality infrastructure evaluation model.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution:

[0006] This invention includes the following steps:

[0007] Collect metrological calibration data and certification process data from a pre-defined quality infrastructure, and preprocess the metrological calibration data and certification data; the metrological calibration data includes calibration data, calibration cost data, and instrument lifecycle data; the certification process data includes process data and certification data.

[0008] Based on the digital twin of the certification process, the metrological calibration data and the certification process data are digitally self-verified at each node to obtain node verification data. Then, the node verification data is quantitatively traced and analyzed using process calibration linkage to obtain linkage analysis data. The linkage analysis data includes calibration input-output ratio, measurement reliability, calibration cost ratio, traceability coverage, and problem tracing accuracy.

[0009] Service quality analysis is performed on the node verification data to obtain quality data. Based on the linkage analysis data and the quality data, a metrological calibration certification coefficient is constructed. The quality data includes certification cycle ratio, cross-departmental collaboration time consumption, scientific nature of process nodes, and quality loss ratio.

[0010] A quality infrastructure evaluation model based on a dual-cycle indicator system is constructed based on the metrological calibration and certification coefficients. The quality infrastructure evaluation model is optimized using the recall risk coefficient, and the target model is output.

[0011] Furthermore, the method for performing digital node self-verification of the metrological calibration data and the certification process data includes:

[0012] Locate nodes according to the process, classify nodes according to the type of deviation, and treat node type as an additional attribute;

[0013] The failure risk coefficient of nodes is calculated by failure mode and impact analysis to obtain the quality impact, the stability of historical data is measured by the coefficient of variation to obtain the data volatility, and the correlation strength between nodes and upstream and downstream processes is analyzed to obtain the process coupling degree.

[0014] The importance score is obtained by objectively weighting the impact on quality, data volatility, and process coupling. When the importance score is greater than or equal to 8, the current node is marked as a critical node; when the importance score is less than 8 but greater than or equal to 5, the current node is marked as an important node; when the importance score is less than 5, the current node is marked as a general node.

[0015] Based on the self-verification triggering strategy, a real-time verification mode is used to verify critical nodes, an adaptive periodic verification mode is used to verify important nodes, and a batch verification mode is used to verify general nodes; the self-verification triggering strategy includes automatic triggering and passive triggering.

[0016] To address system bias, an exponential smoothing method is used to predict the trend of node parameters. When the predicted value deviates from the threshold by more than or equal to 3σ, a pre-calibration command is triggered. A bias cumulative index is introduced, and deep verification is initiated when the bias cumulative index exceeds the historical 95th percentile value.

[0017] To address random bias, an isolated forest model is trained to identify outliers, calculate the isolated path length of node data, distinguish between real anomalies and random fluctuations based on the isolated path length, automatically adjust the anomaly detection threshold based on the data distribution, and set a dynamic threshold.

[0018] To address structural deviations, a graph neural network is used to identify clusters of strongly correlated nodes, with nodes as vertices and data interaction frequency as edge weights. For nodes with a correlation degree greater than the correlation degree threshold, cosine similarity is used to verify the logical consistency of the data. When there is an inconsistency, the source node is automatically traced, and node self-verification is performed based on the source node.

[0019] To address environmental deviations, a multivariate linear regression model is established with calibration data as the dependent variable and environmental parameters and supply chain timeliness as independent variables. The measured values ​​are then corrected in real time for environmentally sensitive nodes using regression coefficients.

[0020] For complex nodes, a weighted voting mechanism is used to integrate system deviation prediction, isolated forest anomaly detection, and graph neural network association analysis to comprehensively determine the node status. The node status is then self-verified, and the verified data is output as node verification data.

[0021] Furthermore, a method for quantitatively tracing and analyzing the node verification data using process calibration linkage includes:

[0022] A three-dimensional few-selection model of impact, correlation, and anomaly is constructed to calculate the quality impact, process correlation, and data anomaly of node verification data; where the quality impact is the probability of the problem occurring multiplied by the severity, the process correlation is the coupling strength of several points, and the data anomaly is the verification deviation multiple.

[0023] The quality impact, process relevance, and data anomaly of the node verification data are objectively weighted to obtain a screening score. When the screening score is greater than or equal to 7, the corresponding node is given high priority; when the screening score is less than 7 but greater than or equal to 4, the corresponding node is given medium priority; when the screening score is less than 4, the corresponding node is given low priority.

[0024] When the abnormality of node verification data exceeds 1.5, real-time filtering is automatically initiated; mid-priority nodes are batch filtered daily, and a sliding window algorithm is used to identify persistently abnormal nodes; when process changes or quality incidents occur, the selection range is automatically expanded.

[0025] Nodes are categorized according to their type to obtain metrology calibration nodes and certification process nodes. For metrology calibration nodes, Monte Carlo simulation is used to calculate calibration deviation. By taking advantage of the amplification effect of the measurement process on product size, a calibration drift coefficient is introduced. When the calibration drift coefficient is greater than the calibration drift threshold, the instrument is triggered for deep calibration. The correlation between calibration data and process parameters is traced to identify the superimposed effects of calibration deviation and improper process parameters.

[0026] For certification process nodes, the impact of node delays on the total certification cycle is calculated using the critical path method. The differences between abnormal nodes and the quality management system of the International Automotive Task Force are quantified to identify compliance gaps, and the fishbone diagram analysis method is used to locate process breakpoints.

[0027] For metrology and calibration nodes, the standard value is automatically increased when the calibration standard is too low after tracing. For certification process nodes, the process is restructured when the cross-department approval time accounts for more than 0.4% after tracing. The deviation is calculated based on the node verification data and standard reference data before and after the adjustment, and output as linkage analysis data.

[0028] Furthermore, the method for obtaining quality data by performing service quality analysis on the node verification data includes:

[0029] Based on the time sequence records in the node verification data, the material cycle ratio is obtained by comparing the material standard certification cycle with the actual verification data showing the time consumed, and the process cycle ratio is obtained by comparing the actual process time with the process certification standard cycle.

[0030] Based on the compliance markers in the node verification data, the node pass rate is statistically analyzed in layers according to key nodes, important nodes, and general nodes.

[0031] Based on the verification frequency in the node verification data and linked with financial data, the unit calibration cost of high-precision equipment and conventional equipment is calculated; the cost-benefit ratio is calculated by combining the reduction in quality problems in the node verification data with the calibration cost.

[0032] The calibration data accuracy is calculated based on the deviation value in the node verification data to obtain the system deviation accuracy and random deviation accuracy; the average time difference from alarm to status termination for each step is calculated based on the warning log timestamp of the node verification data to obtain the time consumed by the abnormal response step.

[0033] Based on the departmental interaction records in the node verification data, the cross-departmental collaboration waiting index is calculated; the cosine similarity algorithm is used to calculate the coupling degree of process nodes.

[0034] The cycle ratio, node pass rate, calibration unit cost, cost-benefit ratio, calibration data accuracy, time spent in abnormal response, collaborative waiting index, and process node coupling degree are output as quality data.

[0035] Furthermore, the method for constructing metrological calibration certification coefficients based on the linkage analysis data and the quality data includes:

[0036] Obtain standard reference data, standard response time threshold, systematic deviation accuracy, and random deviation accuracy, and construct the metrological calibration and certification coefficient, expressed as follows:

[0037] ;

[0038] ;

[0039] ;

[0040] ;

[0041] ;

[0042] in For metrological calibration certification coefficient, From an efficiency perspective, For quality dimensions, From the perspective of efficiency, For consistency dimension, This is the comprehensive adjustment coefficient. This is the efficiency adjustment coefficient. This is the quality adjustment coefficient. This is the benefit adjustment coefficient. This is the consistency adjustment coefficient. For material cycle ratio, This refers to the proportion of the process cycle. For abnormal response time, The standard response time threshold, For the coordinated waiting index, For the pass rate of key nodes, For the pass rate of important nodes, This represents the pass rate of a typical node. For systematic bias accuracy, For random bias accuracy, For cost-benefit ratio, For actual calibration costs, To calibrate costs for the budget, For the coupling degree of process nodes, To adjust for deviation, To adjust for previous deviations, For standard reference data, For material cycle weighting coefficients, This is the process cycle weighting coefficient. As the weight of key nodes, Weights for important nodes This represents the weight of a typical node.

[0043] Furthermore, the method for constructing a quality infrastructure evaluation model based on a dual-cycle indicator system includes:

[0044] The internal circulation indicators include the metrological calibration dimension and the certification process dimension. The metrological calibration dimension includes the standard input-output ratio, measurement reliability, and calibration cost ratio; the certification process dimension includes the certification cycle ratio, cross-departmental collaboration time, and the scientific nature of process nodes.

[0045] External circulation metrics include quality outcome dimensions and risk control dimensions. Quality outcome dimensions include quality loss ratio, cost-benefit ratio, and node pass rate; risk control dimensions include traceability coverage and problem tracing accuracy.

[0046] The internal and external circulation indicators are standardized. The internal circulation index is obtained by objectively weighting the metrological calibration dimension and the certification process dimension and multiplying it by the metrological calibration and certification coefficient. The external circulation index is obtained based on the quality result dimension and the risk control dimension.

[0047] The objective function of the quality infrastructure evaluation model is constructed by objectively weighting the internal circulation index and the external circulation index. The quality infrastructure evaluation model includes grey relational analysis, clustering algorithm, and graph neural network algorithm. A three-level algorithm pipeline is established. Grey relational analysis is responsible for index selection and outputting a set of key evaluation indicators. The clustering algorithm groups the samples based on the key indicators and generates category labels. The graph neural network uses the clustering results as the initial features of the nodes and aggregates topological information through a message passing mechanism to finally output a comprehensive evaluation value. Among them, the graph neural network algorithm optimizes the mapping between node states and true evaluation values ​​by using the objective function based on the node features of the classified data, and outputs a comprehensive evaluation value of quality infrastructure based on topological relationships.

[0048] Furthermore, the method for optimizing the quality infrastructure evaluation model using the recall risk coefficient includes:

[0049] A multilayer perceptron neural network is used to construct a recall risk coefficient calculation model, and the input data is preprocessed to calculate the recall risk coefficient.

[0050]

[0051] in To determine the recall risk factor, Let be the total number of products recalled in the i-th recall within period t. The actual production volume of the i-th recall within period t. The severity weighting factor for the i-th recall is... Let i be the time interval from the occurrence of the i-th recall to the present. The attenuation coefficient is... Number of recalls;

[0052] Risk is categorized based on recall risk coefficient values, and high-risk cluster centers are encoded as particle position vectors.

[0053] Using the recall risk coefficient as the core moderating variable in the external circulation, the adjustment coefficient of the quality infrastructure evaluation model is adjusted as follows:

[0054]

[0055] in The j-th adjustment coefficient is the adjusted value. Let j be the adjustment coefficient. Risk sensitivity coefficient;

[0056] Introducing the recall risk coefficient, we obtain the risk-optimized objective function, expressed as:

[0057]

[0058] in To adjust the objective function of the quality infrastructure evaluation model, The objective function after risk optimization. This represents the risk impact coefficient.

[0059] The recall risk coefficient model is retrained and updated quarterly using new recall and quality data; when the actual recall rate deviates from the model prediction by more than 5%, the model parameters are backtracked for optimization; the RFC grading threshold is adjusted annually based on the industry average recall rate.

[0060] Secondly, a quality infrastructure evaluation model construction system includes:

[0061] Data acquisition and processing module: used to acquire metrological calibration data and certification process data of a preset quality infrastructure, and to preprocess the metrological calibration data and certification data; the metrological calibration data includes calibration data, calibration cost data, and instrument lifecycle; the certification process data includes process data and certification data;

[0062] Analysis and traceability verification module: used to perform digital node self-verification of the metrological calibration data and the certification process data based on the digital twin of the certification process, to obtain node verification data, and to perform quantitative traceability analysis on the node verification data using process calibration linkage to obtain linkage analysis data; the linkage analysis data includes calibration input-output ratio, measurement reliability, calibration cost ratio, traceability coverage, and problem tracing accuracy.

[0063] Function construction module: Performs service quality analysis on the node verification data to obtain quality data, and constructs metrological calibration certification coefficients based on the linkage analysis data and the quality data; the quality data includes certification cycle ratio, cross-departmental collaboration time consumption, scientific nature of process nodes, and quality loss ratio;

[0064] Modeling and optimization module: This module is used to construct a quality infrastructure evaluation model based on a dual-cycle indicator system according to the metrological calibration and certification coefficients, optimize the quality infrastructure evaluation model using the recall risk coefficient, and output the target model.

[0065] The beneficial effects of this invention are:

[0066] This invention provides a method and system for constructing a quality infrastructure evaluation model. Compared with existing technologies, this invention has the following technical advantages:

[0067] This invention integrates metrological calibration and certification process data through preprocessing, digital node self-verification, quantitative traceability analysis, service quality analysis, construction of metrological calibration and certification coefficients, model building, and model optimization. By linking digital node self-verification with process calibration analysis, it accurately identifies the combined quality risks of both processes, avoiding the limitations of single-dimensional evaluation. Differentiated verification based on node importance, combined with multiple algorithms to handle different deviations, improves data accuracy and reduces resource waste and missed risk assessments. A dual-cycle indicator system is constructed, introducing a dynamic optimization model for recall risk coefficients to adapt to the quality control needs of the automotive industry, align with industry standards, quantify compliance gaps, and enhance the guiding value of evaluation for production improvement. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating the steps of constructing a quality infrastructure evaluation model according to the present invention. Detailed Implementation

[0069] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0070] The present invention provides a method and system for constructing a quality infrastructure evaluation model, comprising the following steps:

[0071] like Figure 1 As shown, this embodiment includes the following steps:

[0072] Collect metrological calibration data and certification process data from a pre-defined quality infrastructure, and preprocess the metrological calibration data and certification data; the metrological calibration data includes calibration data, calibration cost data, and instrument lifecycle data; the certification process data includes process data and certification data.

[0073] In actual evaluation, the metrological calibration data includes: calibration data of the coordinate measuring machine (monthly calibration deviation value 0.002-0.005mm), calibration cost data (the cost of a single calibration of high-precision equipment is 8,000 yuan, and that of conventional equipment is 2,000 yuan), and instrument life cycle (the lifespan of the coordinate measuring machine is 8 years).

[0074] Certification process data: Crankshaft material certification process data (including 3 approval nodes, total standard cycle of 5 working days), process certification data (compliance mark pass rate initial value 92%).

[0075] During the preprocessing stage, outliers (such as data with calibration deviations > 0.01 mm due to equipment failure) are removed, and the data format is unified as timestamp-index value-associated node ID;

[0076] Based on the digital twin of the certification process, the metrological calibration data and the certification process data are digitally self-verified at each node to obtain node verification data. Then, the node verification data is quantitatively traced and analyzed using process calibration linkage to obtain linkage analysis data. The linkage analysis data includes calibration input-output ratio, measurement reliability, calibration cost ratio, traceability coverage, and problem tracing accuracy.

[0077] Service quality analysis is performed on the node verification data to obtain quality data. Based on the linkage analysis data and the quality data, a metrological calibration certification coefficient is constructed. The quality data includes certification cycle ratio, cross-departmental collaboration time consumption, scientific nature of process nodes, and quality loss ratio.

[0078] A quality infrastructure evaluation model based on a dual-cycle indicator system is constructed based on the metrological calibration and certification coefficients. The quality infrastructure evaluation model is optimized using the recall risk coefficient, and the target model is output.

[0079] In this embodiment, the method for performing digital node self-verification on the metrological calibration data and the certification process data includes:

[0080] Locate nodes according to the process, classify nodes according to the type of deviation, and treat node type as an additional attribute;

[0081] The failure risk coefficient of nodes is calculated by failure mode and impact analysis to obtain the quality impact, the stability of historical data is measured by the coefficient of variation to obtain the data volatility, and the correlation strength between nodes and upstream and downstream processes is analyzed to obtain the process coupling degree.

[0082] The importance score is obtained by objectively weighting the impact on quality, data volatility, and process coupling. When the importance score is greater than or equal to 8, the current node is marked as a critical node; when the importance score is less than 8 but greater than or equal to 5, the current node is marked as an important node; when the importance score is less than 5, the current node is marked as a general node.

[0083] Based on the self-verification triggering strategy, a real-time verification mode is used to verify critical nodes, an adaptive periodic verification mode is used to verify important nodes, and a batch verification mode is used to verify general nodes; the self-verification triggering strategy includes automatic triggering and passive triggering.

[0084] To address system bias, an exponential smoothing method is used to predict the trend of node parameters. When the predicted value deviates from the threshold by more than or equal to 3σ, a pre-calibration command is triggered. A bias cumulative index is introduced, and deep verification is initiated when the bias cumulative index exceeds the historical 95th percentile value.

[0085] To address random bias, an isolated forest model is trained to identify outliers, calculate the isolated path length of node data, distinguish between real anomalies and random fluctuations based on the isolated path length, automatically adjust the anomaly detection threshold based on the data distribution, and set a dynamic threshold.

[0086] To address structural deviations, a graph neural network is used to identify clusters of strongly correlated nodes, with nodes as vertices and data interaction frequency as edge weights. For nodes with a correlation degree greater than the correlation degree threshold, cosine similarity is used to verify the logical consistency of the data. When there is an inconsistency, the source node is automatically traced, and node self-verification is performed based on the source node.

[0087] To address environmental deviations, a multivariate linear regression model is established with calibration data as the dependent variable and environmental parameters and supply chain timeliness as independent variables. The measured values ​​are then corrected in real time for environmentally sensitive nodes using regression coefficients.

[0088] For complex nodes, a weighted voting mechanism is used to integrate system bias prediction, isolated forest anomaly detection, and graph neural network correlation analysis to comprehensively determine the node status. The node status is then self-verified, and the verified data is output as node verification data.

[0089] In practical assessments, the types of deviations include systematic deviations, random deviations, structural deviations, and environmental deviations.

[0090] Automatic triggering: Critical nodes are automatically verified hourly, important nodes are triggered daily, and general nodes are batch verified weekly; Passive triggering: When upstream or downstream nodes experience anomalies, environmental parameters change abruptly, or production switches occur, the relevant nodes are automatically verified.

[0091] The correlation threshold was 0.812. The key threshold was determined through historical data statistical analysis and expert consultation. The correlation threshold of 0.812 corresponds to a significance level of P<0.05.

[0092] The process is divided into 12 nodes, among which coordinate measuring machine calibration and material compliance approval are key nodes (importance score 8.5), routine dimensional sampling inspection is an important node (score 6.2), and record archiving is a general node (score 3.8).

[0093] The exponential smoothing method is used to predict the deviation trend of the coordinate measuring machine (CMM) for systematic deviation. When the deviation between the predicted value and the threshold reaches 3σ (σ=0.001mm), pre-calibration is triggered. For environmental deviation, a regression model is established with calibration data as the dependent variable and temperature (20±2℃) and humidity (50±5% RH) as independent variables to correct the measured values ​​in real time. Real-time verification data of key nodes (such as real-time calibration deviation of the CMM of 0.003mm) and daily verification data of important nodes (pass rate of 96% for routine dimension sampling inspection) are used.

[0094] In this embodiment, the method for quantitative traceability analysis of the node verification data using process calibration linkage includes:

[0095] A three-dimensional few-selection model of impact, correlation, and anomaly is constructed to calculate the quality impact, process correlation, and data anomaly of node verification data; where the quality impact is the probability of the problem occurring multiplied by the severity, the process correlation is the coupling strength of several points, and the data anomaly is the verification deviation multiple.

[0096] The quality impact, process relevance, and data anomaly of the node verification data are objectively weighted to obtain a screening score. When the screening score is greater than or equal to 7, the corresponding node is given high priority; when the screening score is less than 7 but greater than or equal to 4, the corresponding node is given medium priority; when the screening score is less than 4, the corresponding node is given low priority.

[0097] When the abnormality of node verification data exceeds 1.5, real-time filtering is automatically initiated; mid-priority nodes are batch filtered daily, and a sliding window algorithm is used to identify persistently abnormal nodes; when process changes or quality incidents occur, the selection range is automatically expanded.

[0098] Nodes are categorized according to their type to obtain metrology calibration nodes and certification process nodes. For metrology calibration nodes, Monte Carlo simulation is used to calculate calibration deviation. By taking advantage of the amplification effect of the measurement process on product size, a calibration drift coefficient is introduced. When the calibration drift coefficient is greater than the calibration drift threshold, the instrument is triggered for deep calibration. The correlation between calibration data and process parameters is traced to identify the superimposed effects of calibration deviation and improper process parameters.

[0099] For certification process nodes, the impact of node delays on the total certification cycle is calculated using the critical path method. The differences between abnormal nodes and the quality management system of the International Automotive Task Force are quantified to identify compliance gaps, and the fishbone diagram analysis method is used to locate process breakpoints.

[0100] For metrology and calibration nodes, the standard value is automatically increased when the calibration standard is too low after tracing. For certification process nodes, the process is restructured when the cross-department approval time accounts for more than 0.4% after tracing. The deviation is calculated based on the node verification data and standard reference data before and after the adjustment, and output as linkage analysis data.

[0101] In actual evaluation, the calibration drift coefficient is determined based on a combination of instrument accuracy level and process requirements, and the calibration drift threshold is 0.05 / month;

[0102] The critical node was selected based on the following weighted factors: node quality impact (probability of occurrence 0.05 × severity 8 = 0.4), process correlation (coupling strength 0.9), and data anomaly (verification deviation multiple 1.2). The objective weighted score was 7.8, and the node was prioritized as a high-level node.

[0103] For metrology calibration nodes (coordinate measuring machine), Monte Carlo simulation was used to calculate the calibration drift coefficient as 0.045 / month (<0.05 / month, no deep calibration required); for certification process nodes (material approval), the critical path method was used to find that cross-departmental approval accounted for 0.42% of the time (>0.4), and the process was restructured.

[0104] The calibration input-output ratio is 1:3.2, the measurement reliability is 99.8%, and the traceability coverage is 95%.

[0105] In this embodiment, the method for obtaining quality data by performing service quality analysis on the node verification data includes:

[0106] Based on the time sequence records in the node verification data, the material cycle ratio is obtained by comparing the material standard certification cycle with the actual verification data showing the time consumed, and the process cycle ratio is obtained by comparing the actual process time with the process certification standard cycle.

[0107] Based on the compliance markers in the node verification data, the node pass rate is statistically analyzed in layers according to key nodes, important nodes, and general nodes.

[0108] Based on the verification frequency in the node verification data and linked with financial data, the unit calibration cost of high-precision equipment and conventional equipment is calculated; the cost-benefit ratio is calculated by combining the reduction in quality problems in the node verification data with the calibration cost.

[0109] The calibration data accuracy is calculated based on the deviation value in the node verification data to obtain the system deviation accuracy and random deviation accuracy; the average time difference from alarm to status termination for each step is calculated based on the warning log timestamp of the node verification data to obtain the time consumed by the abnormal response step.

[0110] Based on the departmental interaction records in the node verification data, the cross-departmental collaboration waiting index is calculated; the cosine similarity algorithm is used to calculate the coupling degree of process nodes.

[0111] The cycle ratio, node pass rate, calibration unit cost, cost-benefit ratio, calibration data accuracy, abnormal response time, collaborative waiting index, and process node coupling degree are output as quality data.

[0112] In actual evaluation, for critical nodes: a pass rate of ≥99% indicates process stability; if three consecutive batches are below 95%, a process optimization warning is triggered. For important nodes: a pass rate of ≥97% indicates process stability. For general nodes: a pass rate of ≥95% indicates process stability. The pass rates for critical nodes, important nodes, and general nodes are obtained based on the number of times the process is stable and the total number of tests.

[0113] Cost-benefit ratio = (Defective product loss before improvement - Defective product loss after improvement) / Total calibration cost;

[0114] Systematic bias accuracy = 1 - |mean deviation| / standard allowable deviation; Random bias accuracy = 1 - |maximum single deviation| / 3σ;

[0115] Cross-departmental collaboration waiting index = Σ (inter-departmental waiting time) / total process time;

[0116] Process node coupling degree = cos(θ) = (A·B) / (||A||·||B||), where A and B are the data vectors of the associated nodes;

[0117] Material cycle ratio (actual time 6 days / standard 5 days = 1.2), key node pass rate 98%, calibration unit cost (high-precision equipment 8000 yuan / time, conventional equipment 2000 yuan / time), cost-benefit ratio 25%.

[0118] In this embodiment, the method for constructing metrological calibration certification coefficients based on the linkage analysis data and the quality data includes:

[0119] Obtain standard reference data, standard response time threshold, systematic deviation accuracy, and random deviation accuracy, and construct the metrological calibration and certification coefficient, expressed as follows:

[0120] ;

[0121] ;

[0122] ;

[0123] ;

[0124] ;

[0125] in For metrological calibration certification coefficient, From an efficiency perspective, For quality dimensions, From the perspective of efficiency, For consistency dimension, This is the comprehensive adjustment coefficient. This is the efficiency adjustment coefficient. This is the quality adjustment coefficient. This is the benefit adjustment coefficient. This is the consistency adjustment coefficient. For material cycle ratio, This refers to the proportion of the process cycle. For abnormal response time, The standard response time threshold, For the coordinated waiting index, For the pass rate of key nodes, For the pass rate of important nodes, This represents the pass rate of a typical node. For systematic bias accuracy, For random bias accuracy, For cost-benefit ratio, For actual calibration costs, To calibrate costs for the budget, For the coupling degree of process nodes, To adjust for deviation, To adjust for previous deviations, For standard reference data, For material cycle weighting coefficients, This is the process cycle weighting coefficient. As the weight of key nodes, Weights for important nodes These are the weights of a typical node;

[0126] In the actual evaluation, μ=0.8, α=0.3, β=0.4, γ=0.2, ε=0.1 are taken. Substituting these values ​​into the formula, we get b1=0.85, b2=0.92, b3=0.88, b4=0.9. The final metrological calibration certification coefficient is:

[0127] .

[0128] In this embodiment, the method for constructing a quality infrastructure evaluation model based on a dual-cycle indicator system includes:

[0129] The internal circulation indicators include the metrological calibration dimension and the certification process dimension. The metrological calibration dimension includes the standard input-output ratio, measurement reliability, and calibration cost ratio; the certification process dimension includes the certification cycle ratio, cross-departmental collaboration time, and the scientific nature of process nodes.

[0130] External circulation metrics include quality outcome dimensions and risk control dimensions. Quality outcome dimensions include quality loss ratio, cost-benefit ratio, and node pass rate; risk control dimensions include traceability coverage and problem tracing accuracy.

[0131] The internal and external circulation indicators are standardized. The internal circulation index is obtained by objectively weighting the metrological calibration dimension and the certification process dimension and multiplying it by the metrological calibration and certification coefficient. The external circulation index is obtained based on the quality result dimension and the risk control dimension.

[0132] The objective function of the quality infrastructure evaluation model is constructed by objectively weighting the internal circulation index and the external circulation index. The quality infrastructure evaluation model includes grey relational analysis, clustering algorithm, and graph neural network algorithm. A three-level algorithm pipeline is established. Grey relational analysis is responsible for index selection and outputting a set of key evaluation indicators. The clustering algorithm groups the samples based on the key indicators and generates category labels. The graph neural network uses the clustering results as the initial features of the nodes and aggregates topological information through a message passing mechanism to finally output a comprehensive evaluation value. Among them, the graph neural network algorithm optimizes the mapping between node states and true evaluation values ​​by using the objective function based on the node features of the classified data, and outputs a comprehensive evaluation value of quality infrastructure based on topological relationships.

[0133] Grey relational analysis calculates the correlation between each evaluation indicator and the reference sequence, sets a threshold, and selects indicators with a correlation higher than the threshold as key evaluation indicators.

[0134] In the actual evaluation, the internal circulation index (weighted score of metrological calibration dimension 85×0.823 + weighted score of certification process dimension 80×0.823≈134.8) and the external circulation index (score of quality results dimension 88 + score of risk control dimension 85≈173) were used to construct the objective function S=0.4×134.8+0.6×173≈157.7.

[0135] In this embodiment, the method for optimizing the quality infrastructure evaluation model using the recall risk coefficient includes:

[0136] A multilayer perceptron neural network is used to construct a recall risk coefficient calculation model, and the input data is preprocessed to calculate the recall risk coefficient.

[0137]

[0138] in To determine the recall risk factor, Let be the total number of products recalled in the i-th recall within period t. The actual production volume of the i-th recall within period t. The severity weighting factor for the i-th recall is... Let i be the time interval from the occurrence of the i-th recall to the present. The attenuation coefficient is... Number of recalls;

[0139] Risk is categorized based on recall risk coefficient values, and high-risk cluster centers are encoded as particle position vectors.

[0140] Using the recall risk coefficient as the core moderating variable in the external circulation, the adjustment coefficient of the quality infrastructure evaluation model is adjusted as follows:

[0141]

[0142] in The j-th adjustment coefficient is the adjusted value. Let j be the adjustment coefficient. Risk sensitivity coefficient;

[0143] Introducing the recall risk coefficient, we obtain the risk-optimized objective function, expressed as:

[0144]

[0145] in To adjust the objective function of the quality infrastructure evaluation model, The objective function after risk optimization. This represents the risk impact coefficient.

[0146] The recall risk coefficient model is retrained and updated quarterly using new recall and quality data; when the actual recall rate deviates from the model prediction by more than 5%, the model parameters are backtracked for optimization; the recall risk coefficient grading threshold is adjusted annually based on the industry average recall rate.

[0147] In actual evaluation, the recall risk coefficient calculation model is as follows: 8 neurons in the input layer; 2 hidden layers (16 neurons in the first layer and 8 neurons in the second layer), with the linear rectified function as the activation function; 1 neuron in the output layer (recall risk coefficient value, ranging from [0,1], the higher the value, the greater the risk); cross-entropy loss is used as the loss function, Adam is selected as the optimizer, and iterative training is performed for 500 rounds.

[0148] Attenuation coefficient Determined based on product lifecycle and failure modes;

[0149] According to the ISO 26262 functional safety standard for the automotive industry and the NHTSA recall classification, recalls are divided into three levels and assigned weights: Level 1 is serious, which may lead to safety accidents or personal injury, with a weight of 3; Level 2 is minor, which affects performance but does not directly cause safety risks, with a weight of 1.5; and Level 3 is minor, which is an appearance or packaging problem that does not affect the use of the product, with a weight of 0.5.

[0150] When the recall risk coefficient value is greater than or equal to 0 and less than 0.2, the risk level is extremely low; when the recall risk coefficient value is greater than or equal to 0.2 and less than 0.4, the risk level is low; when the recall risk coefficient value is greater than or equal to 0.4 and less than 0.6, the risk level is medium; when the recall risk coefficient value is greater than or equal to 0.6 and less than 0.8, the risk level is high; when the recall risk coefficient value is greater than or equal to 0.8 and less than 1, the risk level is extremely high.

[0151] The risk sensitivity coefficient is set based on the degree of impact of the indicator on the recall; the risk impact coefficient is adjusted according to the company's risk appetite.

[0152] Recall data in the past 3 months (1 Level 1 recall), 500 pieces For 10,000 pieces, Given 3, Δt is 2 months, z=0.2), calculate b=(500×3×exp(-0.2×2)) / 10000×100%≈13.2%, after adjustment The optimized objective function is:

[0153] Output the target model.

[0154] Secondly, a quality infrastructure evaluation model construction system includes:

[0155] Data acquisition and processing module: used to acquire metrological calibration data and certification process data of a preset quality infrastructure, and to preprocess the metrological calibration data and certification data; the metrological calibration data includes calibration data, calibration cost data, and instrument lifecycle; the certification process data includes process data and certification data;

[0156] Analysis and traceability verification module: used to perform digital node self-verification of the metrological calibration data and the certification process data based on the digital twin of the certification process, to obtain node verification data, and to perform quantitative traceability analysis on the node verification data using process calibration linkage to obtain linkage analysis data; the linkage analysis data includes calibration input-output ratio, measurement reliability, calibration cost ratio, traceability coverage, and problem tracing accuracy.

[0157] Function construction module: Performs service quality analysis on the node verification data to obtain quality data, and constructs metrological calibration certification coefficients based on the linkage analysis data and the quality data; the quality data includes certification cycle ratio, cross-departmental collaboration time consumption, scientific nature of process nodes, and quality loss ratio;

[0158] Modeling and optimization module: This module is used to construct a quality infrastructure evaluation model based on a dual-cycle indicator system according to the metrological calibration and certification coefficients, optimize the quality infrastructure evaluation model using the recall risk coefficient, and output the target model.

[0159] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing a quality infrastructure evaluation model, characterized in that, Includes the following steps: A method for constructing a quality infrastructure evaluation model is applied to the evaluation of automotive components. This method involves collecting metrological calibration data and certification process data from a pre-defined quality infrastructure, and preprocessing the data. The metrological calibration data includes calibration data, calibration cost data, and instrument lifecycle data for a coordinate measuring machine. The certification process data includes crankshaft material certification process data and process certification data. Based on the digital twin of the certification process, the metrological calibration data and the certification process data are digitally self-verified at each node to obtain node verification data. Then, the node verification data is quantitatively traced and analyzed using process calibration linkage to obtain linkage analysis data. The linkage analysis data includes calibration input-output ratio, measurement reliability, calibration cost ratio, traceability coverage, and problem tracing accuracy. Service quality analysis is performed on the node verification data to obtain quality data. Based on the linkage analysis data and the quality data, a metrological calibration certification coefficient is constructed. The quality data includes certification cycle ratio, cross-departmental collaboration time consumption, scientific nature of process nodes, and quality loss ratio. A quality infrastructure evaluation model based on a dual-cycle indicator system is constructed using the metrological calibration and certification coefficients. The quality infrastructure evaluation model is then optimized using a recall risk coefficient, and a target model is output. This includes: The internal circulation indicators include the metrological calibration dimension and the certification process dimension. The metrological calibration dimension includes the standard input-output ratio, measurement reliability, and calibration cost ratio; the certification process dimension includes the certification cycle ratio, cross-departmental collaboration time, and the scientific nature of process nodes. External circulation metrics include quality outcome dimensions and risk control dimensions. Quality outcome dimensions include quality loss ratio, cost-benefit ratio, and node pass rate; risk control dimensions include traceability coverage and problem tracing accuracy. The internal and external circulation indicators are standardized. The internal circulation index is obtained by objectively weighting the metrological calibration dimension and the certification process dimension and multiplying it by the metrological calibration and certification coefficient. The external circulation index is obtained based on the quality result dimension and the risk control dimension. The objective function of the quality infrastructure evaluation model is constructed by objectively weighting the internal circulation index and the external circulation index. The quality infrastructure evaluation model includes grey relational analysis, clustering algorithm, and graph neural network algorithm. A three-level algorithm pipeline is established. Grey relational analysis is responsible for index selection and outputting a set of key evaluation indicators. The clustering algorithm groups the samples based on the key indicators and generates category labels. The graph neural network uses the clustering results as the initial features of the nodes and aggregates topological information through a message passing mechanism to finally output a comprehensive evaluation value. Among them, the graph neural network algorithm optimizes the mapping between node states and true evaluation values ​​by using the objective function based on the node features of the classified data, and outputs a comprehensive evaluation value of quality infrastructure based on topological relationships.

2. The method for constructing a quality infrastructure evaluation model according to claim 1, characterized in that, A method for performing digital node self-verification of the metrological calibration data and the certification process data includes: Locate nodes according to the process, classify nodes according to the type of deviation, and treat node type as an additional attribute; The failure risk coefficient of nodes is calculated by failure mode and impact analysis to obtain the quality impact, the stability of historical data is measured by the coefficient of variation to obtain the data volatility, and the correlation strength between nodes and upstream and downstream processes is analyzed to obtain the process coupling degree. The importance score is obtained by objectively weighting the impact on quality, data volatility, and process coupling. When the importance score is greater than or equal to 8, the current node is marked as a critical node; when the importance score is less than 8 but greater than or equal to 5, the current node is marked as an important node; when the importance score is less than 5, the current node is marked as a general node. Based on the self-verification triggering strategy, a real-time verification mode is used to verify critical nodes, an adaptive periodic verification mode is used to verify important nodes, and a batch verification mode is used to verify general nodes; the self-verification triggering strategy includes automatic triggering and passive triggering. To address system bias, an exponential smoothing method is used to predict the trend of node parameters. When the predicted value deviates from the threshold by more than or equal to 3σ, a pre-calibration command is triggered. A bias cumulative index is introduced, and deep verification is initiated when the bias cumulative index exceeds the historical 95th percentile value. To address random bias, an isolated forest model is trained to identify outliers, calculate the isolated path length of node data, distinguish between real anomalies and random fluctuations based on the isolated path length, automatically adjust the anomaly detection threshold based on the data distribution, and set a dynamic threshold. To address structural deviations, a graph neural network is used to identify clusters of strongly correlated nodes, with nodes as vertices and data interaction frequency as edge weights. For nodes with a correlation degree greater than the correlation degree threshold, cosine similarity is used to verify the logical consistency of the data. When there is an inconsistency, the source node is automatically traced, and node self-verification is performed based on the source node. To address environmental deviations, a multivariate linear regression model is established with calibration data as the dependent variable and environmental parameters and supply chain timeliness as independent variables. The measured values ​​are then corrected in real time for environmentally sensitive nodes using regression coefficients. For complex nodes, a weighted voting mechanism is used to integrate system deviation prediction, isolated forest anomaly detection, and graph neural network association analysis to comprehensively determine the node status. The node status is then self-verified, and the verified data is output as node verification data.

3. The method for constructing a quality infrastructure evaluation model according to claim 1, characterized in that, A method for quantitative tracing and analysis of the node verification data using process calibration linkage includes: A three-dimensional few-selection model of impact, correlation, and anomaly is constructed to calculate the quality impact, process correlation, and data anomaly of node verification data; where the quality impact is the probability of problem occurrence multiplied by the severity, the process correlation is the node coupling strength, and the data anomaly is the verification deviation multiple. The quality impact, process relevance, and data anomaly of the node verification data are objectively weighted to obtain a screening score. When the screening score is greater than or equal to 7, the corresponding node is given high priority; when the screening score is less than 7 but greater than or equal to 4, the corresponding node is given medium priority; when the screening score is less than 4, the corresponding node is given low priority. When the abnormality of node verification data exceeds 1.5, real-time filtering is automatically initiated; mid-priority nodes are batch filtered daily, and a sliding window algorithm is used to identify persistently abnormal nodes; when process changes or quality incidents occur, the selection range is automatically expanded. Nodes are categorized according to their type to obtain metrology calibration nodes and certification process nodes. For metrology calibration nodes, Monte Carlo simulation is used to calculate calibration deviation. By taking advantage of the amplification effect of the measurement process on product size, a calibration drift coefficient is introduced. When the calibration drift coefficient is greater than the calibration drift threshold, the instrument is triggered for deep calibration. The correlation between calibration data and process parameters is traced to identify the superimposed effects of calibration deviation and improper process parameters. For certification process nodes, the impact of node delays on the total certification cycle is calculated using the critical path method. The differences between abnormal nodes and the quality management system of the International Automotive Task Force are quantified to identify compliance gaps, and the fishbone diagram analysis method is used to locate process breakpoints. For metrology and calibration nodes, the standard value is automatically increased when the calibration standard is too low after tracing. For certification process nodes, the process is restructured when the cross-department approval time accounts for more than 0.4% after tracing. The deviation is calculated based on the node verification data and standard reference data before and after the adjustment, and output as linkage analysis data.

4. The method for constructing a quality infrastructure evaluation model according to claim 1, characterized in that, A method for obtaining quality data by performing service quality analysis on the node verification data includes: Based on the time sequence records in the node verification data, the material cycle ratio is obtained by comparing the material standard certification cycle with the actual verification data showing the time consumed, and the process cycle ratio is obtained by comparing the actual process time with the process certification standard cycle. Based on the compliance markers in the node verification data, the node pass rate is statistically analyzed in layers according to key nodes, important nodes, and general nodes. Based on the linkage between the calibration frequency in the node calibration data and financial data, the unit calibration cost of high-precision equipment and conventional equipment is calculated; the cost-benefit ratio is calculated by combining the reduction in quality problems in the node calibration data with the calibration cost. The calibration data accuracy is calculated based on the deviation value in the node verification data to obtain the system deviation accuracy and random deviation accuracy; the average time difference from alarm to state termination in each link is calculated based on the warning log timestamp of the node verification data to obtain the time consumption of the abnormal response link. Based on the departmental interaction records in the node verification data, the cross-departmental collaboration waiting index is calculated; the cosine similarity algorithm is used to calculate the coupling degree of process nodes. The cycle ratio, node pass rate, calibration unit cost, cost-benefit ratio, calibration data accuracy, time spent in abnormal response, collaborative waiting index, and process node coupling degree are output as quality data.

5. The method for constructing a quality infrastructure evaluation model according to claim 1, characterized in that, The method for constructing metrological calibration certification coefficients based on the linkage analysis data and the quality data includes: Obtain standard reference data, standard response time threshold, systematic deviation accuracy, and random deviation accuracy, and construct the metrological calibration and certification coefficient, expressed as follows: ； ； ； ； ； in For metrological calibration certification coefficient, From an efficiency perspective, For quality dimensions, From the perspective of efficiency, For consistency dimension, This is the comprehensive adjustment coefficient. This is the efficiency adjustment coefficient. This is the quality adjustment coefficient. This is the benefit adjustment coefficient. This is the consistency adjustment coefficient. For material cycle ratio, This refers to the proportion of the process cycle. For abnormal response time, The standard response time threshold, For the coordinated waiting index, For the pass rate of key nodes, For the pass rate of important nodes, This represents the pass rate of a typical node. For systematic bias accuracy, For random bias accuracy, For cost-benefit ratio, For actual calibration costs, To calibrate costs for the budget, For the coupling degree of process nodes, To adjust for deviation, To adjust for previous deviations, For standard reference data, For material cycle weighting coefficients, This is the process cycle weighting coefficient. As the weight of key nodes, Weights for important nodes This represents the weight of a typical node.

6. The method for constructing a quality infrastructure evaluation model according to claim 1, characterized in that, The method for optimizing the quality infrastructure evaluation model using the recall risk coefficient includes: A multilayer perceptron neural network is used to construct a recall risk coefficient calculation model, and the input data is preprocessed to calculate the recall risk coefficient. in To determine the recall risk factor, Let be the total number of products recalled in the i-th recall within period t. The actual production volume of the i-th recall within period t. The severity weighting factor for the i-th recall is... Let i be the time interval from the occurrence of the i-th recall to the present. The attenuation coefficient is... Number of recalls; Risk is classified according to the recall risk coefficient value, and high-risk cluster centers are encoded as particle position vectors; Using the recall risk coefficient as the core moderating variable in the external circulation, the adjustment coefficient of the quality infrastructure evaluation model is adjusted as follows: in The j-th adjustment coefficient is the adjusted value. Let j be the adjustment coefficient. Risk sensitivity coefficient; Introducing the recall risk coefficient, we obtain the risk-optimized objective function, expressed as: in To adjust the objective function of the quality infrastructure evaluation model, The objective function after risk optimization. This represents the risk impact coefficient. The recall risk coefficient model is retrained and updated every quarter with new recall and quality data; when the actual recall rate deviates from the model prediction by more than 5%, the model parameters are backtracked and optimized.

7. A quality infrastructure evaluation model construction system, used to perform the method according to any one of claims 1-6, characterized in that, include: Data acquisition and processing module: used to acquire metrological calibration data and certification process data of preset quality infrastructure, and to preprocess the metrological calibration data and certification process data; The metrological calibration data includes calibration data, calibration cost data, and instrument lifecycle data; The authentication process data includes process data and authentication data; Analysis and traceability verification module: used to perform digital node self-verification of the metrological calibration data and the certification process data based on the digital twin of the certification process, to obtain node verification data, and to perform quantitative traceability analysis on the node verification data using process calibration linkage to obtain linkage analysis data; the linkage analysis data includes calibration input-output ratio, measurement reliability, calibration cost ratio, traceability coverage, and problem tracing accuracy. Function construction module: Performs service quality analysis on the node verification data to obtain quality data, and constructs metrological calibration certification coefficients based on the linkage analysis data and the quality data; the quality data includes certification cycle ratio, cross-departmental collaboration time consumption, scientific nature of process nodes, and quality loss ratio; Modeling and optimization module: This module is used to construct a quality infrastructure evaluation model based on a dual-cycle indicator system according to the metrological calibration and certification coefficients, optimize the quality infrastructure evaluation model using the recall risk coefficient, and output the target model.

Images