Good product production real-time monitoring system for automotive device

Abstract

The application discloses a good product production real-time monitoring system of vehicle gauge equipment, and belongs to the technical field of quality control, and comprises a sensing module, a reinforcement test module and a decision module; the sensing module is used for collecting physical parameters and performance test physical data of vehicle gauge product production processes in real time, constructing a batch deviation matrix, calculating a quality deviation degree, generating an edge identification table, and screening key control parameters affecting physical performance; the reinforcement test module is used for performing virtual testing on a target edge piece, generating an equipment control scheme by using a parameter coupling matrix, dynamically adjusting parameters, comparing virtual and real curves to calibrate a twin body, optimizing a digital twin body, and reissuing parameters; and the decision module is used for constructing a traceability file of the target edge piece, defining decision parameters, calculating a comprehensive decision score through a multi-dimensional judgment matrix, executing differentiated decisions in different scenes, generating equipment control instructions, and accurately identifying edge pieces with performance close to a qualified threshold.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of quality control, and relates to a good product production real-time monitoring system for automotive-grade equipment. BACKGROUND

[0002] With the continuous progress of artificial intelligence technology, in the field of industrial manufacturing, especially in the manufacturing of intelligent automobiles, there are strict requirements for quality control, data monitoring and result analysis of each link of automotive-grade semiconductor devices, and the production process needs to be managed through systematic means to ensure that the products meet the relevant standards of automotive-grade.

[0003] The existing Chinese patent with the publication number CN118226816A discloses a kind of system and method for controlling size quality of parts, the system includes detection module and data analysis module, the detection module is used to carry out real-time detection to the gear produced, obtains the detection data of each gear;The data analysis module is used for quality control analysis on the detection data of gear, determines the corresponding detection result according to the detection data of each gear, the detection result includes whether the detection is qualified and corresponding detailed result data;For detection unqualified, detailed result data includes unqualified position, deviation value;For detection qualified, detailed result data includes error position, error value;Gear is divided into several statistical departments, the detection result of each gear is acquired in real time, and the detection curve graph corresponding to each statistical department is set according to the obtained detection result;Based on each detection curve graph, it is judged whether gear production quality control is needed.

[0004] Although the prior art can give early warning to production quality problems, ensure production quality and reduce production loss, it does not consider the potential risks and differentiated control requirements of marginal quality state under extreme working conditions of automotive-grade, specifically, for parts in the qualified marginal state, only binary logic of qualified or unqualified is used to determine the quality of parts, and no special identification and differentiation mechanism is established for those parts with performance close to the qualified threshold. In actual application, automobiles are faced with various extreme working conditions such as high temperature, heavy load and long-term vibration, and marginal parts close to the qualified threshold may have performance degradation or even failure. If such marginal parts are not automatically identified and marked and no additional reliability test is performed, it is difficult to detect potential risks in advance, so that the parts that seem to be qualified in regular detection may expose quality problems under complex working conditions after being installed, and it is difficult to meet the strict requirements of automotive-grade for full-condition reliability. SUMMARY

[0005] In view of the deficiencies of the prior art, the purpose of the present application is to provide a good product production real-time monitoring system for automotive-grade equipment, which can accurately identify marginal parts by calculating the parameter and quality deviation, break the limitation of binary judgment, build a twin-body simulation extreme working condition for marginal parts, perform intensive physical testing, predict failure risks, and make multi-dimensional decision control to avoid marginal parts from being installed, so as to meet the requirements of automotive-grade full-condition reliability.

[0006] To achieve the above object, the present application provides the following technical solutions:

[0007] The good product production real-time monitoring system of the automotive-grade device comprises a perception module, a reinforcement test module and a decision module.

[0008] The perception module is used for collecting physical parameters and performance test physical data of the automotive-grade product production process in real time, constructing a batch deviation matrix, calculating a quality deviation degree in combination with an ideal solution and a negative ideal solution, generating an edge recognition table, and screening key control parameters affecting physical performance.

[0009] The reinforcement test module is used for performing virtual testing on the target edge piece, performing multi-physical field simulation by using a parameter coupling matrix, generating a device control scheme, dynamically adjusting parameters, comparing virtual and real curves to calibrate the twin, and optimizing the digital twin and reissuing parameters.

[0010] The decision module is used for constructing a traceability file of the target edge piece, defining decision parameters, calculating a comprehensive decision score by using a multi-dimensional judgment matrix, and executing differentiated decisions in different scenarios to generate device control instructions.

[0011] Specifically, the perception module comprises a data acquisition unit, a quality deviation unit and a coupling unit.

[0012] The data acquisition unit is used for testing multiple performance parameters of the automotive-grade product and configuring a product identification code to obtain finished product performance data.

[0013] The quality deviation unit is configured with an edge recognition strategy, which is used for receiving finished product performance data, constructing a threshold matching table, calculating parameter deviation by using a measured difference and an upper limit difference, eliminating unqualified products, calculating the quality deviation of the automotive-grade product by using an entropy weight method, setting an edge state threshold, and identifying edge pieces.

[0014] The coupling unit is configured with a coupling recognition strategy, which is used for creating a digital twin, judging a coupling relationship by using a Pearson coefficient and mutual information, screening coupling parameter pairs, calculating coupling strength, and generating a coupling weight matrix.

[0015] Specifically, the edge recognition strategy comprises:

[0016] A classified threshold library is established according to parameter types, each parameter is assigned a unique parameter ID, a threshold matching table is constructed in combination with finished product performance data;

[0017] For each parameter in the threshold matching table, a threshold median value in the matching threshold range is calculated.

[0018] Obtain the measured difference between the measured value and the median threshold, and the upper limit difference between the upper limit threshold and the median threshold, and calculate the parameter deviation. ,once Mark the parameter as exceeding the threshold;

[0019] For automotive-grade products with out-of-threshold markings, they are marked as non-conforming products and directed to the scrap area, generating a single-parameter deviation table for each conforming product.

[0020] Specifically, the edge recognition strategy further includes:

[0021] Constructing the batch deviation matrix ,in, For the first The first automotive-grade product The deviation of the parameter. , , For the number of automotive-grade products, The number of parameter types;

[0022] Calculate the first The entropy value of the parameter is calculated, and the first parameter is calculated. The difference coefficients of the parameters are normalized to obtain the first parameter. The weights of the item parameters;

[0023] For the Parameters, obtain ideal solution and negative ideal solution, and calculate the first ideal solution. The ideal distance and negative ideal distance of each automotive-grade product are used to calculate the quality deviation. Generate a comprehensive deviation table;

[0024] Based on historical quality deviation data, the minimum quality deviation is obtained, a buffer coefficient is introduced, and an edge state threshold is set. ;

[0025] like Determine the first For each automotive-grade product as an edge component, an edge component marker is generated, and an edge recognition table is created; if Determine the first These automotive-grade products are standard qualified parts.

[0026] Specifically, the coupling identification strategy includes:

[0027] For edge components, construct a geometric model, assign material property data, deploy a data interaction interface, and construct a twin model;

[0028] The simulation deviation is calculated; if the simulation deviation is less than the simulation difference threshold, the verification is deemed successful; otherwise, the geometric modeling and attribute assignment are re-examined to generate a digital twin database.

[0029] A coupling analysis data set is constructed and preprocessed to generate a standardized coupling data set;

[0030] Based on the standardized coupling data set, a key parameter list is generated by screening key parameters using variance analysis;

[0031] For any two key parameters, linear coupling coefficients and nonlinear coupling coefficients are calculated, coupling parameter pairs are screened out, coupling strength is calculated based on partial correlation coefficients, and a coupling weight matrix is constructed.

[0032] Specifically, the reinforcement test module is configured with a difference test strategy for virtual testing of the target edge part;

[0033] The difference test strategy includes:

[0034] Based on the edge identification table, the target edge part is obtained, the corresponding twin model is called, and the target edge test set is generated;

[0035] A parameterized virtual working condition load scheme is developed to generate simulation instructions;

[0036] The target edge part is subjected to multi-physics field coupling simulation, the coupling weight matrix is called, a virtual test report is formed, and the single parameter deviation table of the target edge part is combined to locate weak points;

[0037] A physical test content including basic test items and special test items is developed for the target edge part;

[0038] The quality deviation of the target edge part is obtained, a deviation risk threshold is set to divide the risk level of the target edge part, the test parameters are dynamically adjusted based on the risk level, and a physical test scheme is generated.

[0039] Specifically, the difference test strategy further includes:

[0040] The physical test scheme is executed, and real-time test data including test working condition data and test performance data are collected, and real-time simulation calculation is performed based on the test performance data;

[0041] The test performance data is preprocessed, and each test performance data is judged one by one by comparing with the preset result judgment standard in the physical test scheme, and a physical test report of the target edge part is generated;

[0042] The parameter evolution curve of virtual testing and the performance data curve of actual measurement are called to calculate the absolute deviation rate;

[0043] Once the absolute deviation rate exceeds a reasonable range, the corresponding parameters in the twin model are adjusted and optimized, and the virtual simulation is re-run after adjustment until the deviation rate of the virtual curve and the measured curve is reduced to a reasonable range.

[0044] Specifically, the decision module comprises a judgment unit and a feedback unit.

[0045] The judgment unit is configured to build a traceability file of the target edge piece, build a hierarchical edge library, simulate unqualified scenarios regularly, calculate a comprehensive decision score based on a multi-dimensional judgment matrix, divide scenarios, and execute differentiated decisions.

[0046] The feedback unit is configured to build a double-objective optimization function, obtain historical full-link data, output an optimal parameter set based on reinforcement learning, and perform autonomous optimization.

[0047] Specifically, the step of executing differentiated decisions comprises:

[0048] The traceability file of the target edge piece is built, and each file is assigned a unique traceability file ID.

[0049] The hierarchical edge library is built, including a basic layer, a test layer, and a core layer, and independent access permissions are set for each layer.

[0050] The unqualified scenarios of the target edge piece are simulated, and the retrieval algorithm of the database is optimized.

[0051] The decision parameters are defined, a coupling weight matrix and a quality deviation degree are combined, a multi-dimensional judgment matrix is built using the analytic hierarchy process, and a comprehensive decision score is calculated.

[0052] A two-level differentiated decision threshold is set, the comprehensive decision score is divided into scenarios, including a first scenario, a second scenario, and a third scenario, and differentiated decisions are executed.

[0053] Specifically, the step of autonomous optimization comprises:

[0054] Based on the target edge piece, a double-objective optimization function is built, including minimizing the misjudgment rate and minimizing the test cost.

[0055] A historical full-link data set is obtained, and cleaning and standardization processing are performed.

[0056] A triple consisting of state, action, and reward is defined, a reinforcement learning model based on deep Q network is designed, and model training is performed to output an optimal parameter combination.

[0057] Based on the optimal parameter combination, small-batch pilot testing and full-batch deployment are performed.

[0058] Establish a monthly optimization iteration mechanism, extract new data to supplement the historical data set, retrain the reinforcement learning model, and dynamically adjust the optimal parameter group.

[0059] The beneficial effects of the present application are:

[0060] Through a plurality of performance parameter tests, measured data are obtained, a parameter deviation degree is calculated in combination with a threshold median value, a quality deviation degree is calculated by an entropy weight method to objectively assign parameter weights and an approximation ideal solution sorting method, a traditional binary judgment is replaced, edge parts close to a qualified threshold can be accurately identified, and edge parts and ordinary qualified parts can be avoided from being mixed together, a digital twin is constructed for the edge parts, parameter coupling relationships are excavated, a plurality of physical field simulation analogs are simulated under extreme working conditions of a vehicle, and weak points of the edge parts are located in advance, special and basic items are customized for strengthened physical tests, virtual test results are combined to verify the performance of the edge parts under extreme working conditions, and the performance of the edge parts is not only detected conventionally, performance degradation risks can be perceived in advance, a full life cycle tracing archive and a hierarchical library are constructed, and differentiated management and control are combined with multi-dimensional decision execution, the edge parts are released after being verified, are monitored, or are intercepted, unverified edge parts are prevented from flowing into general assembly, identification and test parameters are continuously optimized through reinforcement learning, management and control precision is improved, costs are reduced, a full process from identification to management and control is covered, failure risks under extreme working conditions are avoided in advance, parts with regular qualification but actual hidden dangers are prevented from being installed on vehicles, and vehicle-level full-working-condition reliability requirements are met. BRIEF DESCRIPTION OF DRAWINGS

[0061] Fig. 1 A structure diagram of a good product production real-time monitoring system for vehicle-level equipment is shown in the figure.

[0062] Fig. 2 A flowchart of an edge identification strategy in the present application is shown in the figure.

[0063] Fig. 3 A flowchart of a coupling identification strategy in the present application is shown in the figure.

[0064] Fig. 4 A flowchart of a differential test strategy in the present application is shown in the figure. DETAILED DESCRIPTION

[0065] The technical scheme of the present application will be described in detail below with reference to the accompanying drawings and specific embodiments, and it should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical scheme of the present application, and are not limitations of the technical scheme of the present application, and the technical features in the embodiments and the embodiments can be combined with each other without conflict.

[0066] Embodiment 1

[0067] Reference Figs. 1 to 4 As shown in the figure, the embodiment introduces a good product production real-time monitoring system for vehicle-level equipment, which includes a perception module, a strengthened test module and a decision module.

[0068] The perception module tests the vehicle-grade product for multiple performance parameters through industrial control sensors such as temperature sensors, current sensors, vibration sensors, and insulation testers deployed in the vehicle-grade product production line and preliminary test stations, collects physical parameters and performance test physical data of the vehicle-grade product in the production process such as injection molding temperature, welding current, vibration displacement, and insulation resistance, obtains product performance data, combines the device control threshold range and threshold median value of each physical parameter, calculates the parameter deviation, and screens out unqualified products to trigger the interception instruction of the sorting device to remove them, thereby generating a single-parameter deviation table of qualified products, constructing a batch deviation matrix, calculating the parameter weight using the entropy weight method, combining the ideal solution and negative ideal solution, calculating the quality deviation of the vehicle-grade product, setting an edge state threshold to generate an edge recognition table, constructing a digital twin for the target edge piece, passing the static precision verification to meet the standard and realizing virtual-real precise mapping, and simultaneously constructing a coupling analysis data set, using variance analysis to screen key control parameters such as temperature and pressure that affect physical performance to reduce computing power, combining the Pearson coefficient and mutual information to determine the coupling strength, screening coupling parameter pairs, and mining hidden coupling relationships between different performance parameters to generate a coupling weight matrix, taking into account efficiency and cost.

[0069] The reinforcement test module is used to call the corresponding twin model for the identified edge piece in combination with vehicle-grade test equipment, perform virtual testing, drive parameter dynamic adjustment of the physical testing process through virtual simulation results, use the parameter coupling matrix to perform multi-physical field simulation, locate weak points, realize precise risk focusing, generate a test equipment control scheme according to the risk level based on the physical scheme containing basic and special test items specified in the vehicle, dynamically adjust parameters, pass the technical verification meeting the vehicle-grade standards, and then send the test equipment control scheme to the equipment controller of the test station through the industrial bus to automatically adjust the test equipment parameters, compare the virtual-real curve to calibrate the twin, optimize the digital twin, and then reissue the test parameters to the equipment, reducing resource waste and human error throughout the process. The module is configured with a difference test strategy for virtual testing of the target edge piece and comparison of the virtual test results with the actual test results.

[0070] The decision module is used to build a traceability file of the target edge part and build a hierarchical edge library, adopt distributed storage and block chain storage, guarantee data unforgeable, credible and available, balance the needs of production line efficiency, quality analysis and audit compliance, simulate the unqualified scene caused by abnormal equipment parameters regularly, such as insufficient strength caused by welding current fluctuation, optimize the control parameter retrieval logic, realize fast locking of the root cause after abnormality, take virtual risk, physical result and coupling influence as decision parameters, calculate the comprehensive decision score through multi-dimensional judgment matrix, execute differentiated decision according to scene, generate equipment control instruction, accurately control quality and effectively avoid risk diffusion, take double target optimization function, process historical data, design reinforcement learning model, deploy after pilot verification, iterate parameter adjustment according to month to adapt to production changes, and finally realize the goal of continuously improving the accuracy of quality control and gradually reducing the cost.

[0071] Specifically, the perception module includes a data acquisition unit, a quality deviation unit and a coupling unit.

[0072] The data acquisition unit is used for testing multiple performance parameters of each automotive-grade product, giving each automotive-grade product a unique product identification code during automotive-grade product processing, binding product performance data with the identification code, such as time series data, image data and environmental data, and uploading the data to the quality deviation unit in real time through an industrial network;

[0073] The quality deviation unit is configured with an edge recognition strategy, which is used to call preset performance parameter qualified threshold range, establish a classification threshold library, receive product performance data, build a threshold matching table and calculate the threshold median, calculate the parameter deviation degree through the measured difference and the upper limit difference for any parameter in each automotive-grade product, and eliminate automotive-grade products with parameters exceeding the threshold, thereby generating a single-parameter deviation table of qualified products, building a batch deviation matrix based on the single-parameter deviation table for qualified products of the same batch, calculating parameter weights to avoid subjective bias using entropy weight method, obtaining ideal solution and negative ideal solution, calculating the quality deviation degree of the automotive-grade product, setting an edge state threshold to identify edge parts, and generating an edge recognition table;

[0074] The coupling unit is configured with a coupling recognition strategy, which is used to create a digital twin for the edge part, only keep the basic parameter file for standard qualified parts to avoid waste of computing power and modeling cost, perform static precision verification on the twin, compare the measured value of the edge part with the model parameter to calculate the simulation deviation, store it in the digital twin database if it meets the standard, realize accurate mapping of physical components and virtual models, build a coupling analysis dataset based on historical data of the production line, screen key parameters after preprocessing to reduce computing power, and judge the coupling relationship through Pearson coefficient and mutual information, mine the hidden nonlinear coupling relationship between different performance parameters, screen out coupled parameter pairs, and calculate the coupling strength to generate a coupling weight matrix to quantify the coupling relationship.

[0075] Specifically, the specific steps of the edge recognition strategy include:

[0076] Import the vehicle-grade standard and the device control parameter qualified threshold customized by the vehicle manufacturer, establish a classified threshold library according to the parameter type, such as gear pitch, semiconductor package height, leakage current, insulation resistance, sealing, hardness, assign a unique parameter ID associated with the device control signal to each parameter, and bind the corresponding qualified threshold range , and mark the threshold corresponding to the vehicle-grade standard clause to ensure that the threshold compliance is traceable; wherein, , are the upper and lower limits of the threshold, respectively;

[0077] Receive the finished product performance data, extract each parameter name in the measured data, automatically match the corresponding qualified threshold range through the mapping relationship between the parameter name and the parameter ID, and generate a threshold matching table, including the product identification code, parameter ID, parameter name, measured value, and matching threshold range. If there is no matching threshold for a parameter, such as a newly added test item, a production device shutdown instruction is triggered and an audible and visual alarm is issued, suspending processing at the corresponding station. After manually supplementing the compliant threshold and passing the technical verification, the device is restarted for operation;

[0078] For each parameter in the threshold matching table, calculate the threshold median of the matching threshold range to reflect the ideal qualified state, and obtain the measured difference between the measured value and the threshold median, and the upper limit difference between the threshold upper limit and the threshold median. Through the absolute value of the measured difference and the upper limit difference, the parameter deviation degree is calculated ;

[0079] The parameter deviation degree is verified for effectiveness; if , it indicates that the measured value of the parameter exceeds the matching threshold range, and the parameter quality is unqualified. The parameter is marked as exceeding the threshold; if , it indicates that the measured value of the parameter is within the matching threshold range, and the parameter quality is qualified.

[0080] The calculated parameter deviation degree and the threshold exceeding mark are supplemented to the threshold matching table to ensure that the deviation degree of each parameter is traceable to a specific vehicle-grade product. At the same time, the vehicle-grade product with the threshold exceeding mark is directly marked as unqualified, directly triggering the mechanical interception instruction of the sorting device, guiding the unqualified product to the waste area through the action of the sorting mechanical arm controlled by the PLC, to screen out unqualified products in advance. The remaining vehicle-grade products are marked as qualified products. The relevant data of the qualified products in the threshold matching table, such as product identification code, parameter ID, parameter name, measured value, matching threshold range, threshold median, and parameter deviation degree, are integrated to generate a single parameter deviation table for each qualified product, which serves as the basic data for subsequent device fine-tuning;

[0081] To avoid the subjectivity of manually setting weights, such as mistakenly setting a secondary parameter weight too high, an entropy weight method is used to objectively calculate the weight of each parameter; for the same batch of vehicle rule products, each vehicle rule product contains parameters, based on a single parameter deviation table, a batch deviation matrix of the current batch is constructed , , , , , , ;

[0082] Based on the batch deviation matrix, the entropy value of the th parameter is calculated, and the difference coefficient of the th parameter is calculated, and the difference coefficient of the th parameter is normalized to obtain the weight of the th parameter, so as to directly associate the control priority of the equipment parameter;

[0083] The parameter deviation degree of multiple parameters is comprehensively calculated by the approximation ideal solution sorting method to reflect the overall quality deviation degree of the parts, for the th parameter, the minimum value of the parameter deviation degree in the batch deviation matrix is obtained, defined as the ideal solution , representing the optimal quality state, and the maximum value of the parameter deviation degree in the batch deviation matrix is obtained, defined as the negative ideal solution , representing the worst qualified state, at this time each parameter has a corresponding ideal solution and negative ideal;

[0084] For the th vehicle rule product, based on the weighted distance, the ideal distance between the measured value and the ideal solution, and the negative ideal distance between the measured value and the negative ideal solution are calculated, based on the ratio of the ideal distance to the sum of the ideal distance and the negative ideal distance, the quality deviation degree is calculated, the value is closer to 1, indicating that the overall quality is closer to the lower limit of the qualified state, and the quality deviation degree of all vehicle rule products in the current batch, the weight of each parameter and the batch deviation matrix table are integrated to generate a comprehensive deviation table, including product identification code and corresponding single parameter deviation table, parameter weight and quality deviation degree;

[0085] The historical quality deviation data of the edge pieces in the past period is obtained, the minimum value of the quality deviation degree is obtained , a buffer coefficient is set to avoid missing judgment, based on the difference between the minimum value and the buffer coefficient, an edge state threshold is set For each vehicle regulation product in the comprehensive deviation table, edge parts are identified by comparing the quality deviation value with the edge state threshold value. If the first vehicle regulation product is an edge part, an edge part marker is generated, such as appending "-E" after the product identification code, and the determination basis is recorded, and the edge parts are integrated to generate an edge identification table, which includes batch number, product identification code, determination basis, edge part quantity or proportion, wherein the edge part proportion is the ratio of the number of edge parts to the number of all vehicle regulation products. If the first vehicle regulation product is not an edge part, it is marked as a standard qualified part and does not need subsequent differentiated management and control. Specifically, the specific steps of the coupling identification strategy include:

[0086] Specifically, the specific steps of the coupling identification strategy include:

[0087] The edge identification table is obtained, a high-fidelity digital twin is constructed for the edge parts, and the standard qualified parts do not need to construct a high-fidelity model, only the basic parameter archives are retained to avoid waste of computing power and modeling cost, and a unique twin ID is assigned to each edge part, which is bound to the product identification code;

[0088] The design drawings of the edge parts, such as CAD files, are imported, and an industrial-level modeling software is used to construct a geometric model to ensure that the key dimensions and the measured dimensions in the single-parameter deviation table are completely consistent, and the measured data of the material physical properties of the edge parts, such as the yield strength of aluminum alloy and the heat resistance temperature of silicone rubber, are obtained from the raw material test report, and are assigned to the corresponding components of the geometric model for physical property integration, realizing virtual mapping of materials and performance, deploying data interaction interface for the twin, using OPC UA industrial protocol to interface with the quality offset unit, realizing synchronization of measured data and simulation data, and setting device parameter correction trigger conditions in the twin model, when the deviation between physical test data and twin simulation data exceeds the deviation threshold, automatically starting model parameter calibration to construct the twin model;

[0089] The constructed twin is subjected to static precision verification, the measured value of the key parameters of the edge parts is compared with the corresponding parameters of the twin model, the simulation deviation is calculated, if the simulation deviation is less than the simulation deviation threshold, the twin verification is passed, otherwise, the simulation deviation is out of standard, the geometric modeling and attribute assignment are rechecked; After verification, the twin is stored in the digital twin database, including the twin ID, the product identification code association relationship, the model file, the interface configuration information, and a twin construction report is generated to record the modeling process, the data sources used and the verification results;

[0090] ​The production equipment control parameters and measured performance data of historical automotive-grade products are acquired, including parameter values, covering qualified parts, marginal parts, and unqualified parts, all batch single parameter deviation tables are synchronously acquired, combined with the physical property data in the twin body model, associated according to product batches and parameter IDs, a coupled analysis data set is formed, and each item of parameter data can be traced back to a specific batch and product;

[0091] The coupled analysis data set is preprocessed, including removing outliers, filling missing values, and data normalization, to generate a standardized coupled data set;

[0092] Variance analysis is used to screen key parameters that significantly affect the quality of automotive-grade products. The variance of each parameter in the standardized coupled data set is calculated. The greater the variance, the more significant the impact of parameter fluctuation on quality. The top 30% of all parameter variances is set as the variance threshold. Parameters with variances greater than the variance threshold are retained, and redundant parameters with small variances are removed. A list of key parameters is generated after screening, ensuring that subsequent coupled analysis focuses on core influencing factors and avoids wasting computing power.

[0093] For any two key parameters in the list of key parameters, the linear coupling coefficient is calculated using the Pearson correlation coefficient, and the nonlinear coupling coefficient is calculated using mutual information. If the linear coupling coefficient is less than the linear coupling threshold and the nonlinear coupling coefficient is less than the nonlinear coupling threshold, it is determined that there is no coupling relationship between the two key parameters. Otherwise, it is determined that there is a coupling relationship between the two parameters, and coupled parameter pairs are generated, such as injection temperature and product warpage, and welding current and weld strength.

[0094] For coupled parameter pairs, based on the partial correlation coefficient, the coupling strength is calculated. The parameters in the list of key parameters are used as rows and columns to construct a coupling weight matrix for the coordinated adjustment of production equipment parameters. The coupling strength is used as the coupling weight, and the diagonal elements are 1, indicating the self-regulation weight of the parameter. The coupling weight of parameter pairs without a coupling relationship is 0.

[0095] Specifically, the steps of the difference test strategy include:

[0096] Based on the marginal identification table, the product identification code and quality deviation of the target marginal part are acquired as the basis for virtual working condition severity classification. Based on the preset mapping relationship between the product identification code and the twin body identification, the twin body model of the corresponding marginal part is retrieved from the digital twin database to generate a target marginal test set;

[0097] Referring to the industry standards related to the extreme working conditions of vehicle-grade, and combining the typical quality failure modes of the target edge part, such as fatigue fracture for mechanical parts and high-temperature leakage current exceeding the standard for semiconductor parts, a parameterized virtual working condition load scheme is developed, such as temperature variation range, vibration frequency interval, and electric stress intensity, and is converted into simulation instructions recognizable by the twin, such as temperature-time curve and vibration acceleration spectrum. Through the industrial data interaction interface configured in advance for the twin, the instructions are sent to the twin model to ensure that the virtual working conditions fully reproduce the quality failure inducing environment of the vehicle-grade product in actual use.

[0098] The multi-physics field coupled simulation of the twin model is started, the coupling weight matrix is called, and the multi-parameter coupled simulation of the edge part is realized to avoid risk omission caused by isolated parameter simulation. The key failure indicators of the edge part twin, such as the maximum stress value of the mechanical part and the number of times the leakage current exceeds the standard for the semiconductor part, are monitored in real time. If a certain indicator reaches the preset failure threshold, such as the maximum stress of the gear exceeding the preset proportion of the material yield strength or the leakage current exceeding the upper limit of the vehicle-grade, it is automatically marked as a high-risk point, and the working condition parameters corresponding to the risk point are recorded. After the simulation is completed, the various data recorded during the simulation are sorted and analyzed to form a virtual test report of the edge part, including the key parameter evolution curve, the high-risk point list, the virtual test pass rate, and the product identification code, twin ID, and quality deviation value of the target edge part are marked.

[0099] The virtual test risk report is analyzed in detail, and the high-risk point information identified in the report is extracted, including the location of the high-risk occurrence, the working condition condition triggering the risk, and the corresponding failure parameter indicator. At the same time, combined with the single-parameter deviation table of the target edge part, through cross analysis, the weak point is located through the double superposition of high-risk point position and high-deviation parameter. The high-deviation parameter is a parameter whose parameter deviation exceeds the deviation threshold. The physical test resources are accurately focused on the key risk points to avoid wasting resources in non-critical links.

[0100] The vehicle-grade physical test related industry standards are called, and the physical test content including basic test items and special test items is customized for the target edge part according to the standard requirements and combined with the weak point. The basic test items must strictly cover the test links required by the vehicle-grade standards to ensure the compliance of the test scheme. The special test items are designed for the weak point, such as increasing local stress monitoring for high stress concentration areas and increasing dynamic parameter tracking for parameters prone to exceed the standard. At the same time, referring to the parameter coupling relationship, coupling verification test items are supplemented to verify whether the coupling effect between parameters will cause new failure risks, which not only meets the compliance requirements but also comprehensively and specifically verifies the quality of the edge part.

[0101] The quality deviation degree of the target edge part is obtained, and a deviation risk threshold is set to divide the risk level of the target edge part, including low, medium and high risk. Based on the risk level, the test parameters are dynamically adjusted to realize accurate matching of the edge part risk level and the test severity, which not only ensures that the high-risk parts are fully tested, but also avoids the increase in cost and component loss caused by excessive testing of medium and low-risk parts. All test items, test parameters and judgment criteria are arranged into a test scheme, and a parameter adjustment interface is reserved in the template. The test scheme is verified for compliance, and the test items in the scheme are checked one by one to see if they completely cover the standard requirements, the test parameters meet the standard limits, and the test result judgment criteria are stricter than the standard lower limit, to ensure that the scheme completely meets the vehicle regulation certification requirements and avoids subsequent product failure due to non-compliance of the scheme. After verification, the physical test scheme of the target edge part is generated, and the target edge part identification, test station number and scheme effective time are marked. The scheme is issued to the corresponding physical test station programmable logic controller through the industrial network, and the scheme is synchronized to the subsequent dynamic traceability file, providing a basis for subsequent test process tracing and auditing.

[0102] After receiving the issued physical test scheme, the physical test station automatically starts the test preparation work. According to the test parameter requirements in the scheme, the hardware devices of the station are configured, such as loading the preset temperature change curve in the temperature test box, setting the specified vibration frequency and acceleration in the vibration test table, and calibrating the test voltage and current range in the electrical performance test equipment. After configuration, the device will automatically feed back the hardware ready signal, and activate various types of sensor acquisition channels related to the test on the station. According to the scheme requirements, the data sampling frequency is set to ensure that the collected test data meets the analysis requirements. After the preparation work is completed, the operator scans the exclusive identification of the target edge part, which will automatically check whether the identification matches the currently loaded test scheme to prevent scheme misuse. Through automatic preparation and identification verification, manual operation errors can be greatly reduced to ensure accurate testing from the start stage.

[0103] Real-time acquisition of measured data, including test working condition data and test performance data, and through the data interface of the twin body, the real-time acquisition of test performance data is transmitted to the digital twin model of the target edge part, real-time simulation calculation is carried out based on the test performance data, and the simulation result is compared with the measured data. If the deviation between the two exceeds the preset range, the test is suspended, the deviation reason is investigated, the abnormality is handled in time, the invalid test is avoided to continue, and the reliability of the test data is ensured. The test working condition data is used to monitor whether the actual working condition parameters of the test equipment are consistent with the scheme requirements, to ensure that the test process is strictly carried out according to the preset scene. If the working condition parameters deviate, the test is suspended in time, and the test is restarted after adjusting the equipment parameters. The test performance data is used to record the changes of various performance parameters of the edge part in the test process, which can directly reflect the quality performance of the edge part under the test working condition.

[0104] After the test is completed, the test performance data is preprocessed, and each performance data is determined according to the preset result determination standard in the physical test scheme. If any performance data exceeds the determination standard, it is determined that the physical test fails, and the target edge part is marked as a high-risk unqualified part. If all performance data meets the determination standard, it is determined that the physical test is passed.

[0105] After the determination is completed, a physical test report of the target edge part is generated, which records the target edge part identification, test whole process data curve, final determination result and abnormality explanation in the data processing process in detail.

[0106] The absolute deviation rate is calculated by calling the parameter evolution curve of virtual test and the performance data curve of actual measurement. If the absolute deviation rate is within the preset reasonable range, it means that the accuracy of the current twin body model meets the virtual test requirements and does not need to be adjusted. If the absolute deviation rate exceeds the reasonable range, the key parameters affecting the model accuracy, such as material attribute parameters and structural mechanics parameters, are analyzed according to the deviation condition, and the corresponding parameters in the twin body model are adjusted and optimized. After adjustment, the virtual simulation is run again until the deviation rate of the virtual curve and the actual measurement curve is reduced to the reasonable range.

[0107] Specifically, the specific steps of executing differentiated decision-making include:

[0108] A traceability file of the target edge part is constructed, each file is assigned a unique traceability file ID, and a bidirectional mapping is performed through the product identification code and the traceability file ID, so that the complete file can be called up for any identification input, including basic identity information, quality deviation data, digital twin data, physical test data, and decision-making and flow record, wherein the basic identity information includes a product identification code, a twin ID, a production batch / shift / equipment number, and a raw material batch number, the quality deviation data includes a single parameter deviation table, a quality deviation, and an edge part determination basis, the digital twin data includes a twin construction report and a coupling matrix association, the virtual test data includes a virtual test risk report and a virtual test pass rate, the physical test data includes a physical test scheme, a physical test report, and a twin calibration record, and the decision-making and flow record includes a decision result (release / monitoring / interception), a flow node, an operator ID, and a timestamp;

[0109] Based on the traceability file of the target edge part, a hierarchical edge library is constructed, which is divided into three layers according to data sensitivity and use scenarios, including a basic layer, a test layer, and a core layer, each layer is provided with independent access permission, the hierarchical edge library adopts a distributed storage architecture, supports real-time data writing and second-level retrieval; wherein the basic layer stores basic identity information and decision-making and flow record, which is used for production line operators to query the current state of the edge part, improves the production line flow efficiency, the test layer is superimposed with quality deviation data, virtual / physical test data, which is used for quality team to reproduce the test process, and the core layer is superimposed with digital twin data and raw material / equipment parameters, which is used for audit agencies to trace compliance and R&D teams to analyze failure root causes;

[0110] Every month, simulate the unqualified scenarios of the target edge part, quickly locate the problem source through the traceability file, such as querying the quality of the same batch of raw materials, whether the operation parameters of the production equipment are abnormal, and whether there is deviation in the test process, record the traceability time consumption and positioning accuracy during the exercise process, continuously optimize the retrieval algorithm of the database, and finally realize the effect of locking the root cause and starting rectification in a short time after the abnormality occurs, avoiding the problem of wider quality risk caused by the spread of the problem;

[0111] Define decision parameters for virtual risk, physical results, and coupling effects, combine coupling weight matrix and quality deviation, assign weights to each decision parameter using the analytic hierarchy process, construct a multi-dimensional decision matrix, and perform weighted calculation based on the dimension score value to obtain a comprehensive decision score; wherein the virtual test pass rate is used to configure the weight of the virtual risk, the physical test decision result is used to configure the weight of the physical result, and the coupling weight matrix is used to calculate the coupling strength average of the high coupling parameter pairs of the target edge part to configure the weight of the coupling effect;

[0112] A secondary difference decision threshold is set to divide the comprehensive decision score into scenes, including a first scene, a second scene and a third scene, to execute differentiated decision and link to a trace file, wherein the first scene is to grant release, mark the target edge part as a reinforced verified qualified part, update the transfer record of the trace file, such as release to the final assembly workshop, generate a qualified release sheet including the trace file ID and the judgment matrix for verification by the downstream process, the second scene is cautious release, marked as a key monitoring part, the product identification code and the vehicle VIN code are associated through the trace file, the vehicle networking monitoring rule is configured, the monitoring record is updated, the monitoring period and the alarm threshold are recorded, the third scene is interception and locking, marked as unqualified product, guided to the waste area, the transfer record is updated, the reverse trace is started, the production batch data, the equipment operation parameters and the raw material batch report are retrieved, the equipment maintenance work order and the batch re-inspection work order are generated, and the trace result and the work order information are supplemented to the trace file.

[0113] Specifically, the decision module includes a judgment unit and a feedback unit.

[0114] The judgment unit is configured to build a trace file of the target edge part, ensure accurate information retrieval, and build a hierarchical edge library using distributed storage, blockchain storage to prevent tampering, taking into account production line efficiency, quality analysis and audit compliance, ensuring data credibility, simulating unqualified scenarios regularly, optimizing search algorithms, and quickly locking the root cause of abnormalities, using virtual risk, physical results and coupled impact as decision parameters, calculating a comprehensive decision score based on a multi-dimensional judgment matrix, and performing scene division to execute differentiated decision, accurately control quality and effectively prevent risk spread.

[0115] The feedback unit is configured to build a double-objective optimization function with minimum misjudgment rate and test cost, define the calculation logic, clean and standardize historical full-link data, eliminate abnormalities to ensure sample quality, design a reinforcement learning triple model containing state, action and reward, output the optimal parameter set after training, perform autonomous optimization, verify the parameter effect through small batch pilot, compare multi-dimensional indicators to avoid large-scale risks, deploy the whole batch after meeting the standards, synchronize the parameters to related data sets and trace files, and establish a monthly iteration mechanism to dynamically adjust parameters to adapt to production changes, generate an optimization report, and ultimately achieve the long-term goal of continuously improving the accuracy of quality control and gradually reducing costs.

[0116] Specifically, the specific steps of autonomous optimization include:

[0117] Based on the target edge part, a double-objective optimization function is built, including minimizing misjudgment rate and minimizing test cost, wherein the misjudgment rate is calculated based on the ratio of the sum of the number of qualified parts misjudged as edge parts and the number of edge parts missed to the total number of parts, and the test cost is calculated by summing the product of the virtual test computing power cost and the time length, and the product of the physical test equipment cost and the time length.

[0118] A historical full-link data set is acquired, including threshold parameters of the target edge part identification stage, working condition settings of the virtual test, scheme parameters of the physical test, corresponding decision results, misjudgment records, and test cost statistics. The historical full-link data set is cleaned to eliminate abnormal data caused by equipment failure and human operation errors, and is standardized to unify data format and magnitude. A historical data set for model training is constructed to provide sufficient sample support for reinforcement learning.

[0119] A triple including state, action, and reward is defined, a reinforcement learning model based on deep Q network is designed, and model training is performed to output an optimal parameter combination to achieve autonomous optimization. The state is the quality characteristics of the previous production batch, such as the overall quality level of the recent edge part, the historical misjudgment rate, and the current test cost, which comprehensively reflects the current management and control state of the system. The action is the parameter to be adjusted, including the edge state threshold of the edge part identification stage, the extreme working condition intensity of the virtual test stage, and the test duration and sampling frequency of the physical test stage, covering the key management and control links. The reward is a double-objective optimization function value. If the misjudgment rate decreases and the test cost is controllable after adjusting the parameters, a positive reward is given. If the parameter adjustment violates the vehicle standard or leads to a decrease in production line efficiency, a negative penalty is given.

[0120] Based on the optimal parameter combination, small-batch pilot is performed. The edge parts of a certain production batch are selected, and the optimal parameter combination is used to execute all previous management and control links from edge part identification to differential test. The misjudgment rate, test cost, and production line efficiency of the pilot batch and the historical batch are compared to verify the actual effect of the optimal parameter combination. If the pilot results meet the expectations, the optimal parameter combination is updated to the production line, and is simultaneously written into the coupling analysis data set and the trace file to ensure that the management and control of all subsequent batches are based on the optimized parameters.

[0121] A monthly optimization iteration mechanism is established. The newly added production and test data are extracted and supplemented to the historical data set every month. The reinforcement learning model is retrained to dynamically adjust the optimal parameters to adapt to the dynamic changes of the production environment, such as raw material batch update, equipment aging, and vehicle standard fine-tuning. A production optimization report is generated to analyze the change trend of the misjudgment rate and test cost, evaluate the improvement effect of the optimization on the vehicle quality, and achieve the long-term goal of continuously improving the quality management and control accuracy and gradually reducing the management and control cost.

[0122] Working principle and effect:

[0123] By multi-sensor and automatic test equipment to collect vehicle product multi-dimensional performance parameters, combined with the pre-set classification threshold library to construct threshold matching table and calculate threshold median, and then calculate the parameter deviation degree to eliminate unqualified products; again based on the same batch of qualified product data to construct deviation matrix, use entropy weight method to objectively allocate parameter weight, combined with ideal solution and negative ideal solution to calculate quality deviation degree, accurately identify edge pieces, replace traditional binary judgment, avoid edge pieces and ordinary qualified pieces confusion; for edge pieces to construct digital twin, excavate parameter coupling relationship to generate coupling weight matrix, through multi-physics field simulation to simulate vehicle extreme working condition positioning weak point, then customize the strengthening physical test containing special and basic items, compare virtual and real data to calibrate twin, detect the performance degradation risk of edge pieces in extreme working condition in advance, avoid missing hidden dangers in regular detection, construct full life cycle traceability file and hierarchical library, combined with virtual risk, physical results, coupling influence to calculate comprehensive decision score, execute differentiated control, optimize parameters and monthly iteration, ensure that edge pieces do not flow into assembly, meet the vehicle full working condition reliability requirements, and continuously improve the control precision and reduce the cost.

[0124] The above only describes the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the concept of the present application shall be deemed to fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall also be deemed to fall within the protection scope of the present application.

Claims

1. A good production real-time monitoring system for automotive-grade devices, characterized by, Comprise: A perception module, a reinforcement test module and a decision module; The perception module is used to collect physical parameters and performance test physical data of vehicle regulation product production process in real time, construct batch deviation matrix, calculate quality deviation degree combined with ideal solution and negative ideal solution, generate edge identification table, and screen key control parameters affecting physical performance; The reinforcement test module is used to obtain target edge piece based on the edge identification table, perform virtual test on the target edge piece, perform multi-physical field simulation by using parameter coupling matrix, generate equipment control scheme, dynamically adjust parameters, compare virtual and real curves to calibrate twin body, optimize digital twin body, and reissue parameters; The decision module is used to construct a trace file of the target edge piece, define decision parameters, calculate comprehensive decision score through a multi-dimensional decision matrix, and execute differentiated decisions in different scenarios to generate equipment control instructions.

2. The real-time monitoring system for good production of vehicle regulation equipment according to claim 1, wherein: The perception module comprises a data acquisition unit, a quality deviation unit and a coupling unit; The data acquisition unit is used to test multiple performance parameters of vehicle regulation products and configure product identification codes to obtain finished product performance data; The quality deviation unit is configured with an edge identification strategy, which is used to receive finished product performance data, construct a threshold matching table, calculate parameter deviation by comparing measured difference and upper limit difference, and eliminate unqualified products, calculate the quality deviation of vehicle regulation products by combining entropy weight method, and set edge state threshold to identify edge pieces; The coupling unit is configured with a coupling identification strategy, which is used to create a digital twin, judge coupling relationship by Pearson coefficient and mutual information, screen coupling parameter pairs, calculate coupling strength, and generate a coupling weight matrix.

3. The real-time monitoring system for good production of vehicle regulation equipment according to claim 2, wherein: The edge identification strategy comprises: Establish a classification threshold library according to parameter types, assign a unique parameter ID to each parameter, construct a threshold matching table combined with finished product performance data; For each parameter in the threshold matching table, calculate the threshold median of the matching threshold range; The measured difference between the measured value and the threshold value, the upper limit difference between the threshold upper limit and the threshold value, and the parameter deviation degree are calculated Once The parameter is marked as over-threshold. For vehicle regulation products with an over-threshold mark, mark them as unqualified products and guide them to the waste area to generate a single-parameter deviation table for each qualified product.

4. The real-time monitoring system for good production of vehicle regulation equipment according to claim 3, wherein: The edge identification strategy further comprises: constructing a batch deviation matrix wherein, is a parameter deviation degree of a parameter of a first vehicle rule product, , , is a number of vehicle rule products, is a number of parameter types;​ Calculate the first The entropy value of the parameter is calculated, and the first parameter is calculated. The difference coefficients of the parameters are normalized to obtain the first parameter. The weights of the item parameters; For the first item parameter, the ideal solution, negative ideal solution are obtained, and the ideal distance, negative ideal distance of the first vehicle rule product are calculated, so as to calculate the quality deviation , and generate a comprehensive deviation table; Based on historical quality deviation data, obtain the minimum value of quality deviation degree, introduce a buffer coefficient, and set an edge state threshold ; like Determine the first For each automotive-grade product as an edge component, an edge component marker is generated, and an edge recognition table is created; if Determine the first These automotive-grade products are standard qualified parts.

5. The system for real-time monitoring of production of good products of automotive equipment according to claim 4, characterized by, The coupling identification strategy comprises: For edge pieces, construct a geometric model, assign material attribute data, and deploy a data interaction interface to construct a twin model; Calculate the simulation deviation; if the simulation deviation is less than the simulation difference threshold, the verification is passed, otherwise, recheck the geometric modeling and attribute assignment to generate a digital twin database; Construct a coupling analysis data set and perform preprocessing to generate a standardized coupling data set; Based on the standardized coupling data set, use variance analysis to screen key parameters to generate a key parameter list; For any two key parameters, the linear coupling coefficient, the nonlinear coupling coefficient are calculated, the coupling parameter pairs are screened out, the coupling strength is calculated based on the partial correlation coefficient, and the coupling weight matrix is constructed. 6.The system according to claim 5, characterized in that: The reinforcement test module is configured with a difference test strategy for virtual testing of the target edge part; The difference test strategy includes: Based on the edge identification table, the target edge part is obtained, the corresponding twin model is called, and the target edge test set is generated; A parameterized virtual working condition load scheme is developed to generate simulation instructions; Multi-physics field coupling simulation is performed on the target edge part, the coupling weight matrix is called, a virtual test report is formed, and the single parameter deviation table of the target edge part is used to locate weak points; A physical test content including basic test items and special test items is developed for the target edge part; The quality deviation of the target edge part is obtained, the deviation risk threshold is set to divide the risk level of the target edge part, the test parameters are dynamically adjusted based on the risk level, and the physical test scheme is generated.

7. The system for real-time monitoring of production of good products of automotive equipment according to claim 6, characterized by, The difference test strategy also includes: The physical test scheme is executed, and real-time test data including test working condition data and test performance data are collected, and real-time simulation calculation is performed based on the test performance data; The test performance data is preprocessed, the result determination criteria preset in the physical test scheme are used to determine each test performance data one by one, and a physical test report of the target edge part is generated; The parameter evolution curve of virtual testing and the measured performance data curve are called to calculate the absolute deviation rate; Once the absolute deviation rate exceeds the reasonable range, the corresponding parameters in the twin model are adjusted and optimized, and the virtual simulation is run again after adjustment until the deviation rate of the virtual curve and the measured curve decreases to a reasonable range. 8.The system according to claim 7, characterized in that: The decision module includes a determination unit and a feedback unit; The determination unit is used to build a traceability file of the target edge part, build a hierarchical edge library, simulate unqualified scenarios regularly, calculate a comprehensive decision score based on a multi-dimensional decision matrix, and perform differentiated decision-making; The feedback unit is used to build a double-objective optimization function, obtain historical full-link data, output an optimal parameter set based on reinforcement learning, and perform autonomous optimization.

9. The system for real-time monitoring of production of good products of automotive equipment according to claim 8, characterized by, The steps of performing differentiated decision-making include: Building a traceability file of the target edge part and assigning a unique traceability file ID to each file; Building a hierarchical edge library including a basic layer, a test layer, and a core layer, and setting independent access permissions for each layer; Simulate unqualified scenarios of the target edge part and optimize the retrieval algorithm of the database; Define decision parameters, combine the coupling weight matrix and the quality deviation, and use the analytic hierarchy process to build a multi-dimensional decision matrix to calculate a comprehensive decision score; Set a two-level differentiated decision threshold to divide the comprehensive decision score into scenes, including a first scene, a second scene, and a third scene, and perform differentiated decision-making.

10. The system for real-time monitoring of production of good car-qualified devices according to claim 9, characterized in that, The steps of autonomous optimization include: Based on the target edge piece, a double target optimization function is constructed, including minimizing the misjudgment rate and minimizing the test cost; Get the historical full-link dataset, clean and standardize it; Define a triple including state, action, and reward, design a reinforcement learning model based on deep Q network, and train the model to output the optimal parameter combination; Based on the optimal parameter combination, carry out small batch pilot and full batch deployment; Establish a monthly optimization iteration mechanism, extract new data to supplement the historical dataset, retrain the reinforcement learning model, and dynamically adjust the optimal parameters.

Images