An automatic coding intelligent agent control system and method for multi-task scenarios

Abstract

The present application relates to the technical field of data processing, and especially to an automatic coding intelligent agent control system and method for multi-task scenarios, comprising a data acquisition module, a coding processing module, a quality evaluation module, a strategy optimization module, an execution control module, a defect traceability module, a parameter correction module and an iterative optimization module; the quality evaluation module is used for calculating the quality deviation index of each process core quality index based on a unified feature vector; the strategy optimization module is used for optimizing each process parameter regulation strategy by fusing multi-source information. The present application breaks the limitation of traditional single process coding, effectively avoids cross-process data representation distortion, significantly improves the coding accuracy and data link integrity of multi-process linkage control; has the functions of integrating data coding, quality correlation and parameter iteration, effectively solves the problems of insufficient generalization and closed loop loss of traditional systems, and assists in realizing multi-process quality accurate control in the manufacturing scene, improving production stability and product consistency.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an automatic coding intelligent agent control system and method for multi-task scenarios. Background Technology

[0002] Traditional automatic coding methods are mostly single-task customized designs, relying on manual preset coding rules and fixed model parameters. They can generally only adapt to the data processing needs of a single process (such as stamping or painting), and cannot adapt to the needs of multi-process linkage control.

[0003] Taking the core scenario of continuous production of automotive parts, including stamping, welding, and painting, as an example, this multi-task processing scenario needs to handle cross-modal, high-dimensional, and strongly correlated data such as stamping sheet characteristics, welding process parameters, and coating film thickness distribution. Automatic coding technology, as the core means of standardizing and extracting features from such data, directly affects the accuracy of subsequent process quality assessment and process strategy optimization. Therefore, existing automatic coding systems and methods have the following shortcomings when applied to multi-process manufacturing scenarios for automotive parts:

[0004] First, traditional systems optimize parameters for a single process. When switching between stamping, welding, and painting processes, the coding logic and model structure need to be re-adjusted, resulting in the break of data association between processes. This makes it impossible to build a full-process data link, leading to low coding efficiency and easy distortion of the characteristic representation of welding process parameters and stamping sheet characteristics, which affects the accuracy of quality control throughout the entire process.

[0005] Second, existing technologies only address data encoding at the level of a single process, without linking with core quality indicators such as weld joint strength and coating film thickness uniformity. They lack dynamic parameter correction and iterative optimization capabilities, and the encoding results are disconnected from the quality control requirements of multiple processes in automotive parts. This makes it difficult to support precise process decisions, and the response to sudden quality defects such as weld porosity and coating color difference is lagging, making it unable to adapt to the dynamic needs of continuous production.

[0006] Based on this, it is of great necessity to design an automatic coding intelligent agent control system and method that is adapted to, but not limited to, the entire process of stamping, welding and painting of automotive parts, and has cross-process adaptive coding capability and quality-driven closed-loop optimization. Summary of the Invention

[0007] To solve one of the above-mentioned technical problems, the present invention adopts the following technical solution: an automatic coding intelligent agent control system for multi-task scenarios, including a data acquisition module, a coding processing module, a quality assessment module, a strategy optimization module, an execution control module, a defect tracing module, a parameter correction module, and an iterative optimization module.

[0008] The data acquisition module includes a stamping process acquisition unit, which is used to collect data on quality inspection, process parameters, and material properties of automotive parts during the stamping process.

[0009] The welding process acquisition unit is used to collect quality inspection, process parameters, and material characteristic data for the welding process of automotive parts. The painting process acquisition unit is used to collect quality inspection, process parameters, and material characteristic data for the painting process of automotive parts. The encoding processing module is used to perform unified encoding processing on the collected multi-process data to generate a unified feature vector that integrates process correlation information.

[0010] The quality assessment module is used to calculate the quality deviation index of the core quality indicators of each process based on a unified feature vector.

[0011] The strategy optimization module is used to integrate multi-source information to optimize the process parameter control strategy for each process step.

[0012] The execution control module is used to issue process parameter adjustment instructions and monitor the execution status of the instructions.

[0013] The defect tracing module is used to locate the responsible process for quality defects and the cross-process influencing factors.

[0014] The parameter correction module is used to correct the coding and adjust relevant parameters based on the defect tracing results.

[0015] The iterative optimization module is used to trigger the cyclic execution function of each module according to the production batch.

[0016] Each module and unit establishes a data interaction connection through an industrial Ethernet network to achieve coordinated operation.

[0017] Based on any of the above technical solutions, the following further optimization is made: the specific data collected by the stamping process acquisition unit includes: quality inspection data including sheet flatness image and sheet size error value; process parameter data including stamping pressure, stamping speed, and die temperature time sequence change data; and material characteristic data including hardness, tensile strength, and thickness uniformity test data of the stamped sheet.

[0018] The specific data collected by the welding process acquisition unit includes: quality inspection data including weld joint strength test values ​​and weld formation images; process parameter data including time-series variation data of welding current, welding voltage, welding speed, and shielding gas flow rate; and material characteristic data including test data of welding base material composition and weld joint metallographic structure.

[0019] The specific data collected by the coating process acquisition unit includes: quality inspection data such as coating film thickness distribution images and coating color difference detection values; process parameter data such as time-series changes in spraying pressure, spraying distance, paint viscosity, and drying temperature; and material characteristic data such as test data on the surface cleanliness of the coated sheet and paint adhesion.

[0020] Based on any of the above technical solutions, a further optimization is made: the calculation for unified encoding processing by the encoding processing module follows the following formula:

[0021] ;

[0022] in, Here is the encoding loss function; in the double summation symbol, the summation range of k is 1 to 3, and the summation range of m is 1 to 3; For process-data dynamic weighting coefficients; Material-data adaptation coefficient; The square of the L2 norm is used to quantify the data error of the original data of type k in the m-th process after being encoded by the encoder and reconstructed by the decoder, so as to ensure the consistency between the encoded features and the original data. This is the matrix form of the k-th type of data in the m-th process;

[0023] This is the mapping function for the encoder network; This is the mapping function for the decoder network; This refers to the process correlation coefficient; Pearson correlation coefficient; This is the process parameter matrix for the m-th process step.

[0024] Based on any of the above technical solutions, a further optimization is made to the specific formula for calculating the quality deviation index by the quality assessment module:

[0025] ;

[0026] In the formula, The quality deviation index for the m-th process; This represents the real-time detection value of the core quality indicator for the m-th process; The m represents the industry standard value or enterprise internal control standard value of the core quality indicator of the m-th process; m is the process identifier, where 1 is the stamping process, 2 is the welding process, and 3 is the painting process.

[0027] Based on any of the above technical solutions, a further optimization is made to the specific steps of the quality assessment module in calculating the quality deviation index:

[0028] The first step is to retrieve the standard thresholds corresponding to the core quality indicators of each process in the stamping, welding, and painting of automotive parts.

[0029] The second step is to extract the real-time detection values ​​of the core quality indicators of each process based on the unified feature vector;

[0030] The third step is to substitute the values ​​into the quality deviation index calculation formula to complete the calculation and generate a quality deviation assessment report.

[0031] Based on any of the above technical solutions, a further optimization is made: the calculation of the process parameter control strategy optimized by the strategy optimization module follows the following formula:

[0032] ;

[0033] In the formula, Let m be the objective function of the strategy network for the m-th process. Let be the parameter vector of the policy network for the m-th process; E is the mathematical expectation symbol, used to calculate the mean of the random variable; clip(.) is the cutoff function, used to restrict the input value to a specified interval to avoid excessive policy update amplitude; The strategy update ratio for the m-th process; Let m be the dominance function of the m-th process; The process stability weight is exp(.), which is the natural exponential function. This is the quality sensitivity coefficient; The quality deviation index for the m-th process; Let be the policy entropy of the policy network for the m-th process; The strategy distribution for the m-th process is given by the parameter vector. The determination is based on the probability distribution of the adjustment of process parameters.

[0034] Based on any of the above technical solutions, the following further optimization is made: the process parameter adjustment instructions issued by the execution control module are specifically: the stamping pressure adjustment range of the stamping process is ±5%, the stamping speed adjustment range is ±3%, and the mold temperature adjustment range is ±2℃.

[0035] The welding current adjustment range for the welding process is ±10A, the welding voltage adjustment range is ±0.5V, the welding speed adjustment range is ±0.2mm / s, and the shielding gas flow rate adjustment range is ±5L / min.

[0036] The spraying pressure adjustment range for the coating process is ±0.1MPa, the spraying distance adjustment range is ±5mm, the paint viscosity adjustment range is ±5mPa・s, and the drying temperature adjustment range is ±3℃.

[0037] Based on any of the above technical solutions, a further optimization is made: the probability calculation of the defect tracing module in locating the responsible process for quality defects follows the following formula:

[0038] ;

[0039] In the formula, This is the conditional probability, and its value ranges from [0,1].

[0040] Each process contributes a weight to the process, representing the probability weight of each process causing quality defects;

[0041] The cosine similarity value ranges from [0,1].

[0042] Let m be the material property influence coefficient for the m-th process;

[0043] D represents the currently detected quality defect feature data of automotive parts;

[0044] Let m be the historical defect data for the m-th process, which is a two-dimensional real matrix.

[0045] j is the process identifier, with values ​​of 1, 2, and 3, corresponding to stamping, welding, and painting processes respectively, and is a summation and traversal variable.

[0046] Based on any of the above technical solutions, a further optimization is made to the specific steps of the parameter correction module in correcting parameters: dynamically weighting the process-data corresponding to the defect responsibility process. Improve by 20%, adjust the material-data fit coefficient based on defect type. Adjust by ±10%; increase the update step size of the strategy network parameters corresponding to the defect responsibility process by 1.5 times, and iterate and optimize the process parameter control strategy again.

[0047] Based on any of the above technical solutions, the following further optimization is made: the specific rules for triggering the iterative optimization module to execute each module in a loop are as follows: each production batch consists of 50 to 100 automotive parts, and the entire process module linkage is triggered once after the production of a batch is completed; if a severe quality defect is detected during the production process, the defect tracing module and parameter correction module are immediately triggered to perform their functions, the parameters are corrected, production continues, and the number of production batches is dynamically adjusted according to the optimization effect.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] 1. This invention enables collaborative coding and data processing of cross-modal data across the entire process of stamping, welding, and painting of automotive parts. It breaks through the limitations of traditional single-process coding, constructs a full-link data association between sheet material characteristics, welding parameters, and painting indicators, effectively avoids distortion of cross-process data representation, and significantly improves the coding accuracy and data link integrity of multi-process linkage control.

[0050] 2. This invention establishes a quality-driven closed-loop coding optimization mechanism that deeply links core quality indicators such as weld joint strength and coating film thickness uniformity with the coding process. It can dynamically correct coding parameters for sudden defects such as weld porosity and coating color difference, solving the problems of disconnection between traditional coding and quality control and delayed defect response, and adapting to the dynamic needs of continuous production.

[0051] 3. This invention has cross-process adaptive coding capability, which can adapt to the switching of stamping, welding and painting processes without manual re-adjustment of coding logic and model structure. It greatly reduces the coding debugging cost in multi-process scenarios, improves coding efficiency, and strengthens the correlation of data of different processes, providing reliable data support for the optimization of the whole process strategy.

[0052] 4. This invention integrates data encoding, quality correlation, parameter iteration and other functions through the full-process collaborative design of the automatic coding intelligent agent, effectively making up for the problems of insufficient universality and lack of closed loop in traditional systems. It assists in the manufacturing of automotive parts to achieve precise quality control of multiple processes, reduce defect rate and improve production stability and product consistency. Attached Figure Description

[0053] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or components are generally identified by similar reference numerals. In the drawings, the elements or components are not necessarily drawn to scale.

[0054] Figure 1 This is a connection block diagram of the automatic coding intelligent agent control system of the present invention.

[0055] Figure 2 The flowchart shows the operation of calculating the quality deviation index for the quality assessment module of this invention. Detailed Implementation

[0056] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore merely examples and should not be used to limit the scope of protection of the present invention. The specific composition and process of the present invention are as follows: Figures 1-2 As shown in the image.

[0057] Example 1: An automatic coding intelligent agent control system for multi-task scenarios. The system is applied to the stamping, welding and painting manufacturing process of automotive parts, and includes a data acquisition module, a coding processing module, a quality assessment module, a strategy optimization module, an execution control module, a defect tracing module, a parameter correction module and an iterative optimization module.

[0058] The data acquisition module includes a stamping process acquisition unit, a welding process acquisition unit, and a coating process acquisition unit. Each of these units establishes a one-way data transmission connection with the main body of the data acquisition module via an industrial Ethernet network. They are used to collect quality inspection, process parameters, and material characteristic data for the stamping, welding, and coating processes, respectively, and upload the collected raw data to the main body of the data acquisition module in real time. The main body of the data acquisition module then performs data aggregation and preliminary standardization processing.

[0059] The encoding processing module establishes a bidirectional data interaction connection with the main body of the data acquisition module via industrial Ethernet. On the one hand, it receives standardized multi-process data uploaded by the data acquisition module, and on the other hand, it feeds back data anomalies during the encoding process to the data acquisition module to trigger the acquisition unit to re-acquire data. The encoding processing module is used to perform unified encoding processing on the acquired multi-process data to generate a unified feature vector that integrates process-related information.

[0060] The quality assessment module establishes a one-way data transmission connection with the encoding processing module via industrial Ethernet, and receives the unified feature vector output by the encoding processing module. This unified feature vector is used to calculate the quality deviation index of the core quality indicators of each process. At the same time, the quality assessment module establishes a two-way data interaction connection with the defect tracing module and the iterative optimization module via industrial Ethernet. On the one hand, it synchronizes the quality deviation index and quality status data with both modules. On the other hand, it receives the quality defect association information fed back by the defect tracing module and the evaluation threshold adjustment instructions fed back by the iterative optimization module.

[0061] The strategy optimization module establishes a one-way data transmission connection with the quality assessment module via industrial Ethernet, and receives the quality deviation index output by the quality assessment module; at the same time, it establishes a two-way data interaction connection with the parameter correction module, receives the encoding and control parameter correction values ​​output by the parameter correction module, and feeds back the strategy optimization results; the strategy optimization module is used to integrate multi-source information to optimize the process parameter control strategy of each process.

[0062] The execution control module establishes a one-way data transmission connection with the strategy optimization module via industrial Ethernet to receive process parameter adjustment instructions output by the strategy optimization module. At the same time, it establishes bidirectional data interaction connections with the iterative optimization module and the data acquisition module respectively. On the one hand, it feeds back the instruction execution status (including parameter adjustment signal and equipment operation status) to the iterative optimization module. On the other hand, it obtains real-time process parameters of each process through the data acquisition module to monitor the instruction execution accuracy. The execution control module is used to issue process parameter adjustment instructions and monitor the instruction execution status.

[0063] The defect tracing module establishes a one-way data input connection with the quality assessment module via industrial Ethernet to receive quality defect warnings and quality deviation data; at the same time, it establishes a one-way data transmission connection with the parameter correction module to feed back the responsible process of the quality defect and the cross-process influencing factors to the parameter correction module; the defect tracing module is used to locate the responsible process of the quality defect and the cross-process influencing factors.

[0064] The parameter correction module establishes a one-way data receiving connection with the defect tracing module via industrial Ethernet, and at the same time establishes a two-way data interaction connection with the encoding processing module and the strategy optimization module respectively. It synchronizes the corrected encoding parameters to the encoding processing module and the corrected control parameters to the strategy optimization module, and receives the parameter adaptation effect fed back by the two. The parameter correction module is used to correct the encoding and control related parameters according to the defect tracing results.

[0065] The iterative optimization module establishes bidirectional data interaction connections with the quality assessment module, execution control module, and parameter correction module via industrial Ethernet, respectively, and receives the quality assessment results, instruction execution status, and parameter correction effects from the three modules. At the same time, it establishes unidirectional control signal connections with the encoding processing module and strategy optimization module, and issues module cyclic execution trigger instructions according to production batches. The iterative optimization module is used to trigger the cyclic execution function of each module according to production batches.

[0066] All modules and units communicate and operate in tandem via industrial Ethernet.

[0067] This invention is based on multi-process, multi-dimensional data acquisition. It achieves integrated representation of cross-modal data through unified coding, and then forms process parameter control instructions through quality assessment and strategy optimization. Combined with defect tracing and parameter correction, it completes closed-loop rectification. Finally, it achieves continuous upgrading of the entire process through iterative optimization, transforming each process from independent control to an organically collaborative whole, solving the problems of cross-process data fragmentation, blind process control, and difficulty in defect tracing.

[0068] The workflow is as follows: First, through the data acquisition units of each process, the three-dimensional data of stamping, welding and coating quality, process and material are acquired synchronously, and real-time data interaction between multiple modules is realized through industrial Ethernet.

[0069] Subsequently, the encoding processing module performs unified encoding on the multi-source cross-modal data to generate a unified feature vector that integrates process-related information, providing a standardized data foundation for subsequent analysis;

[0070] The quality assessment module calculates the quality deviation index of each process based on this feature vector, quantifying the actual quality status of the process;

[0071] The strategy optimization module integrates multi-source information such as quality deviation and process correlation to generate the optimal process parameter control strategy for each process.

[0072] The execution control module translates the control strategy into specific instructions and sends them to the production equipment, while monitoring the execution status of the instructions in real time.

[0073] If a quality defect is detected, the defect tracing module immediately locates the responsible process and cross-process influencing factors, and the parameter correction module makes targeted corrections to the coding coefficients and strategy optimization parameters based on the tracing results.

[0074] Finally, the iterative optimization module triggers the entire process module to execute cyclically according to the production batch, and dynamically adjusts the iteration rules based on the production optimization effect to achieve closed-loop intelligent control of the entire process.

[0075] This invention achieves integrated intelligent control of multi-process manufacturing through the time-series linkage design of eight modules and the full-link data interaction of industrial Ethernet, thereby changing the limitations of traditional single-process control at the architectural level.

[0076] Compared to traditional distributed control systems, this solution organically integrates functional modules such as data acquisition, encoding processing, and quality assessment. Each module is linked in a fixed sequence and the data is exchanged in real time, enabling process control to be based on the quality status and process-related information of the entire process, rather than local data of a single process, thus greatly improving the accuracy of control.

[0077] Meanwhile, the closed-loop design of defect tracing and parameter correction can quickly locate and solve quality problems, avoiding the mass production of defective products; the batch triggering and dynamic adjustment rules of the iterative optimization module enable the system to adapt to different production conditions and manufacturing needs of different parts types, realize continuous self-optimization of the process, and adapt to the large-scale and multi-variety production characteristics of automotive parts.

[0078] The selection of core parameters in this solution is based on industry standards for automotive parts stamping, welding, and painting manufacturing, actual production needs, and general standards for intelligent manufacturing technologies.

[0079] For example, in an automotive body welding production line, the process parameter data of the welding robot needs to be transmitted to the strategy optimization module in real time. Industrial Ethernet can achieve millisecond-level data transmission, ensuring the real-time nature of control commands. The triggering method for iterative optimization is selected based on production batches. Automotive parts manufacturing uses batches as the basic production unit, and a batch size of 50-100 pieces is the general standard for small and medium-batch production. In an automotive door stamping-welding-painting production line, 80 pieces are considered as one production batch. After the batch production is completed, the entire process module linkage is triggered, which ensures the timeliness of process optimization while avoiding excessively frequent iterations that may affect production efficiency.

[0080] For example, the timing linkage rules for each module are set according to the following sequence: data acquisition - encoding processing - quality assessment - strategy optimization - execution control - defect tracing - parameter correction - iterative optimization. This sequence is based on the process logic of automobile manufacturing and the technical flow of data processing, and it aligns with the actual production operation process.

[0081] Taking the stamping-welding-painting production scenario of automobile wheel hubs as an example, using the parameter settings of this solution, the data acquisition module synchronously acquires data such as the hardness of the stamping sheet metal, the welding current and voltage, and the film thickness distribution of the coating. This data is transmitted via industrial Ethernet to the encoding processing module to generate a unified feature vector. The quality assessment module calculates the flatness deviation index of the stamping sheet metal, and the strategy optimization module then generates a control strategy to reduce the stamping pressure by 3%. The execution control module sends the instruction to the stamping CNC equipment, and the equipment adjusts the parameters in real time. The entire process is completed within 1 minute. If a color difference defect in the wheel hub coating is subsequently detected, the defect tracing module quickly locates it as unevenness of the coating substrate caused by welding thermal deformation. The parameter correction module performs targeted correction of the coding coefficient of the welding process. The iterative optimization module triggers full-process optimization after completing the production of 100 wheel hubs, thereby improving the yield of subsequent wheel hub products.

[0082] Based on any of the above technical solutions, the following further optimization is made: the specific data collected by the stamping process acquisition unit includes: quality inspection data including sheet flatness image and sheet size error value; process parameter data including stamping pressure, stamping speed, and die temperature time sequence change data; and material characteristic data including hardness, tensile strength, and thickness uniformity test data of the stamped sheet.

[0083] The specific data collected by the welding process acquisition unit includes: quality inspection data including weld joint strength test values ​​and weld formation images; process parameter data including time-series variation data of welding current, welding voltage, welding speed, and shielding gas flow rate; and material characteristic data including test data of welding base material composition and weld joint metallographic structure.

[0084] The specific data collected by the coating process acquisition unit includes: quality inspection data such as coating film thickness distribution images and coating color difference detection values; process parameter data such as time-series changes in spraying pressure, spraying distance, paint viscosity, and drying temperature; and material characteristic data such as test data on the surface cleanliness of the coated sheet and paint adhesion.

[0085] It should be noted that: the forming precision of the sheet metal is directly determined by the material properties, which in turn determine the formability of the stamping process. Therefore, the focus is on collecting material data such as sheet hardness and tensile strength, as well as quality data such as flatness and dimensional error. The core of the welding process is the joint strength and weld formation. Process parameters directly affect the metallographic structure and joint performance. Therefore, the focus is on collecting process timing data such as current and voltage, as well as material data such as base material composition and metallographic structure. The core of the coating process is the uniformity of film thickness and coating adhesion. The surface characteristics of the sheet metal and spraying parameters directly determine the film quality. The focus is on collecting material data such as sheet cleanliness and coating adhesion, as well as quality data such as film thickness distribution and color difference.

[0086] During operation, each process data acquisition unit is equipped with corresponding existing testing equipment and data acquisition terminals. Each acquisition terminal standardizes the collected data of different types, such as images, values, and time series, and then transmits it to the encoding processing module in real time via industrial Ethernet. This provides comprehensive, accurate, and standardized raw data for subsequent unified encoding, ensuring that the collected data is highly matched with the process requirements and quality control points of each process.

[0087] Compared to traditional data acquisition schemes that only collect single process parameter values, this scheme targets the core quality influencing factors of each process, including stamping, welding, and painting. It designs data collection content encompassing material characteristics, process parameters, and quality inspection, covering both static material data and dynamic time-series process parameter data. It also incorporates intuitive image-based quality inspection data, enabling the collected data to comprehensively reflect the production status and quality level of each process. Furthermore, the data collection content for each process is designed according to industry process specifications. The type, dimension, and accuracy of the collected data must meet the quality control requirements of automotive manufacturing. After standardization, it can be directly used for subsequent unified coding without additional secondary processing, significantly improving data processing efficiency. In addition, images of sheet metal flatness and weld formation make quality inspection more intuitive and accurate, capturing subtle quality issues that numerical data cannot reflect.

[0088] Taking the stamping-welding-painting production scenario of an automobile engine hood as an example, using the data collection content and parameter settings of this solution, the stamping process data collection unit acquires the flatness image of the cold-rolled sheet metal of the engine hood through a laser flatness detector, and simultaneously collects the timing data of stamping pressure of 120MPa and stamping speed of 300mm / s, as well as the material data of sheet hardness of 180HV and tensile strength of 350MPa; the welding process data collection unit collects the timing data of welding current of 180A and voltage of 28V, as well as the material data of base material composition (carbon 0.12%, manganese 0.6%) and the metallographic structure of the weld joint being fine-grained ferrite; the painting process data collection unit collects the quality data of spraying pressure of 0.4MPa and film thickness distribution of 85±5μm, as well as the material data of sheet surface cleanliness and coating adhesion.

[0089] All collected data is transmitted to the encoding processing module in real time after standardization processing, providing comprehensive and accurate raw data for subsequent unified encoding. The unified feature vector generated by the encoding processing module based on this data can accurately reflect the material characteristics, process status and quality level of each process in the engine cover production. The strategy optimization module generates process control strategies based on this, which improves the production yield of the engine cover compared with traditional data collection methods.

[0090] Based on any of the above technical solutions, a further optimization is made: the calculation for unified encoding processing by the encoding processing module follows the following formula:

[0091] .

[0092] in, The encoding loss function is used to quantify the overall error of multi-process cross-modal data in the encoding and reconstruction process. It is the core objective of the iterative optimization of the encoding network, so that the encoding result can retain the process characteristics and correlation information of the original data as much as possible.

[0093] In the double summation symbol, the summation range of k is 1 to 3, and the summation range of m is 1 to 3. It is used to perform full-dimensional traversal calculations on three types of data in the three processes of stamping, welding, and painting, so as to realize the overall encoding of multi-process and multi-type data.

[0094] The process-data dynamic weighting coefficient is a non-negative real number used to characterize the importance weight of the k-th type of data in the m-th process, enabling differentiated coding representation of different processes and different types of data.

[0095] The material-data matching coefficient is a positive real number used to match the k-th type of data in the m-th process with the material characteristics of that process, thereby improving the coding accuracy of the core material data.

[0096] The square of the L2 norm, i.e. the square of the Euclidean distance, is used to quantify the data error of the original data of class k in the m-th process after being encoded by the encoder and reconstructed by the decoder, to ensure the consistency between the encoded features and the original data.

[0097] It is a matrix of the k-th type of data in the m-th process. It is a two-dimensional real matrix. The row and column dimensions of the matrix are determined by the sample size and feature dimensions of the collected data. It is used to store the standardized original collected data.

[0098] This is the mapping function for the encoder network; the values ​​in parentheses are the function input terms, used to map the original data matrix. The data is mapped to a high-dimensional feature vector to achieve feature extraction.

[0099] This is the mapping function of the decoder network, and the part in parentheses is the function input. It is used to reconstruct the high-dimensional feature vector output by the encoder into a matrix with the same dimension as the original data, thereby restoring the data.

[0100] The process correlation coefficient is a positive real number used to characterize the correlation strength between the data of the m-th process and the process parameters of that process, thus strengthening the intrinsic connection between the process parameters and the data.

[0101] is the Pearson correlation coefficient, with a value range of [-1, 1]. is the linear correlation coefficient between the k-th type of data in the m-th process and the process parameters of that process, used to quantify the degree of linear correlation between the data and the process parameters.

[0102] Let be the process parameter matrix for the m-th process. It is a two-dimensional real matrix, and the row and column dimensions of the matrix are determined by the acquisition time and parameter type of the process parameters. It is used to store the standardized process parameter data of the process.

[0103] This solution addresses the characteristics of multi-process and cross-modal data in automotive parts stamping, welding, and painting. It breaks through the limitations of traditional coding techniques, such as single-modal coding, lack of process weights, and detachment from process logic, and constructs a unified coding model for multi-source data with dual-dimensional weighting of process and data, process correlation constraints, and minimization of reconstruction errors.

[0104] By employing double summation, a full-dimensional traversal of the three processes and three types of data is achieved, ensuring the comprehensiveness of the coding; through and Differentiated weighting of processes and data is implemented to highlight the coding priority of core processes and core material data; the reconstruction error is calculated by using the square of the L2 norm to ensure the consistency of information between the coded features and the original data; and then... The Pearson correlation coefficient integrates the inherent relationship between process parameters and data into the coding process, so that the coding results not only include the characteristics of the data itself, but also integrate the process logic of automobile manufacturing, solving the problems of cross-modal data being difficult to represent in an integrated manner and the coding results being disconnected from the actual production process.

[0105] During the work process, the standardized raw data collected from each process is first converted into matrix form. Transform process parameter data into a process parameter matrix. And set according to industry standards and process characteristics. , , The initial value; then passed through the encoder network. Original data matrix Mapping to high-dimensional feature vectors enables feature extraction from the data, while simultaneously using a decoder network. The high-dimensional feature vector is reconstructed into a matrix with the same dimension as the original data, completing data restoration; the difference between the original data matrix and the reconstructed matrix is ​​calculated by using the square of the L2 norm to obtain the reconstruction error, quantifying the information loss during the encoding process; and the Pearson correlation coefficient is used to... Calculate the degree of linear correlation between the data of each process step and the process parameters, and combine... Strengthen the weight of process-related information in coding; through and The reconstruction error term and process-related term are weighted, and then a double summation is performed to complete the weighted calculation for all processes and all data types, resulting in the encoding loss function. Finally, with the core objective of minimizing the encoding loss function, the network parameters of the encoder and decoder are iteratively optimized until the loss function converges to the industry-standard threshold. The resulting high-dimensional feature vector is a unified feature vector that integrates process correlation information. This feature vector realizes the integrated representation of multi-process and cross-modal data and includes process correlation information within and between processes.

[0106] This solution achieves high-quality unified coding of multi-process cross-modal data through differentiated weighting, reconstruction of error constraints, and integration of process correlation into multi-dimensional coding. This ensures that the coding results not only guarantee the information integrity of the original data but also deeply integrate the process logic of automobile manufacturing, providing accurate and effective data support for subsequent quality assessment and strategy optimization from the data processing level.

[0107] Compared to traditional single-modal coding schemes, this scheme achieves full-dimensional traversal of three types of data across three processes through double summation, covering multiple data types such as images, numerical data, and time series data, ensuring comprehensive coding; through and Differential weighting can flexibly adjust coding weights according to the importance of the process and the type of data, highlighting the coding priority of core processes such as welding and core material data such as plate hardness, thereby improving the accuracy of coding features.

[0108] By integrating the correlation information between process parameters and data into the coding process, the coding result is no longer a simple mathematical feature, but a feature vector containing process logic. This accurately reflects the inherent relationship between data and process parameters at each stage, making subsequent quality assessment and strategy optimization more aligned with actual production processes. Furthermore, the coding model in this solution employs an iterative optimization approach, allowing for flexible adjustment of coefficient values ​​based on different production scenarios and component types, demonstrating strong adaptability and flexibility.

[0109] The core parameters of the coding formula in this scheme , , The selection of factors such as the convergence threshold of the loss function is based on the technical specifications for industrial big data processing and coding, and the industry process standards for stamping, welding, and painting of automotive parts. Among these, the dynamic weighting coefficient for process-data is... The values ​​are based on the guidelines for weighting automotive manufacturing process data. For stamping, material data is weighted at 0.8-0.9 and process parameters at 0.7-0.8; for welding, quality data is weighted at 0.9-0.95 and process parameters at 0.85-0.9; and for painting, quality data is weighted at 0.8-0.85 and process parameters at 0.75-0.8. These values ​​are set according to the importance of each process in automotive manufacturing and the degree of impact of the data on quality.

[0110] Material-Data Adaptation Coefficient The value range is 0.6-1.2. The hardness data of stamping sheet is 1.0-1.2, the composition data of welding base material is 1.1-1.2, the cleanliness data of coating sheet is 0.9-1.1, and the core material data takes the higher value. The reason is that the material characteristics are the basis for the formation of the quality of automotive parts. Improving the coding adaptability of the core material data can ensure the accuracy of the coding features.

[0111] Process correlation coefficient The values ​​are determined with reference to industry standards for process correlation: 0.8-0.9 for welding, 0.7-0.8 for stamping, and 0.65-0.75 for painting. Welding is assigned a higher value because welding process parameters have the highest correlation with quality and are a core guarantee of automotive component strength. The convergence threshold of the encoding loss function is set to ≤0.05, based on the general error standard for industrial big data encoding. This threshold ensures that information loss during the encoding process is controlled within a reasonable range. Taking the stamping-welding-painting production scenario of an automotive bumper as an example, the parameter settings of this solution are applied... The quality data for welding is set to 0.95, the material data for stamping is set to 0.85, and the process parameters for painting are set to 0.75. The parameters are set as follows: welding base material composition data 1.2, stamping sheet hardness data 1.1, and coating sheet cleanliness data 1.0. The process was set as follows: welding process 0.9, stamping process 0.8, and painting process 0.7. Cross-modal data such as the dimensions of the stamped bumper sheet metal, the weld formation of the welded parts, and the film thickness distribution of the paint were input into the coding model. After iterative optimization, the coding loss function converged to 0.042. The generated unified feature vector successfully characterized the cross-process correlation law of poor uniformity of stamped sheet metal thickness, low weld joint strength, and uneven paint film thickness. The quality assessment module calculated the quality deviation index based on this feature vector, which completely matched the actual production quality status. The strategy optimization module generated a process control strategy based on this, which improved the weld joint strength and the uniformity of the paint film thickness of the bumper.

[0112] The design of the dual summation structure enables full-dimensional coverage of multi-process and multi-type data, avoiding data omissions; the squared term of the L2 norm is the standard form for calculating industrial data reconstruction error, which can accurately quantify the information loss of data encoding and reconstruction, and ensure the consistency between encoded features and original data; the introduction of the Pearson correlation coefficient realizes the linear correlation quantification between process parameters and collected data, and integrates the encoding results into the process logic.

[0113] The parameter values ​​all have clearly defined industry standards and adjustment ranges: The value ranges from 0.7 to 0.95, based on the quality weight of each process in automobile manufacturing (welding is a critical safety process with the highest weight). It can be flexibly adjusted according to the company's product positioning. For example, when producing high-end passenger vehicle parts, the quality data of the welding process will be... Adjusted upwards to 0.95-1.0; The value range is 0.6-1.2. According to the "Classification Standard for the Influence of Material Properties on Manufacturing Quality", the core material data takes the higher value, and the non-core data takes the lower value, such as non-critical dimension data of the stamping process. It can be lowered to 0.6-0.8; The value ranges from 0.65 to 0.9. According to the "Guideline for Evaluating the Process Correlation of Automobile Manufacturing Processes," the welding process has the highest process correlation, followed by the painting process, and then the stamping process. If the produced parts are painted exterior parts, the value of the painting process can be used as the reference value. Increase to 0.8-0.9; the convergence threshold of the coding loss function is ≤0.05. According to the "Industrial Big Data Coding Error Control Standard", if enterprises have higher requirements for data accuracy, the threshold can be lowered to ≤0.03. If coding efficiency is pursued, it can be increased to ≤0.08.

[0114] All parameters can be flexibly adjusted according to the precision of the enterprise's production equipment, product quality requirements, and process characteristics. The direction and range of adjustment conform to the general rules of the industry, and technicians can set them directly according to industry standards and the actual needs of the enterprise.

[0115] This encoding processing module transforms the raw data collected from multiple processes and across modalities into a unified feature vector that integrates process-related information through a process of standardized transformation, feature extraction, reconstruction verification, process association fusion, weighted calculation, and iterative optimization.

[0116] First, the images, numerical data, and time-series data from each process are converted into a matrix with a unified dimension. , This ensures the computability of the data; subsequently, it is passed through an encoder. Feature extraction is performed to map high-dimensional raw data into low-dimensional high-dimensional feature vectors, achieving data dimensionality reduction and feature refinement; Decoder The eigenvectors are reconstructed, and the reconstruction error is calculated using the square of the L2 norm to verify the eigenvectors' ability to represent the original data. Simultaneously, the correlation between the data and process parameters is calculated using the Pearson correlation coefficient, combined with... Strengthen the weight of process-related information in coding; then through and Differentiated weighting is applied to the reconstruction error term and process-related terms to highlight the importance of core processes and core data. Finally, with the goal of minimizing the loss function, the encoder and decoder network parameters are iteratively optimized according to industrial standards using a machine learning model until the loss function converges to a set threshold.

[0117] For example, in the stamping-welding-painting production of automobile bodies, cross-modal data such as the hardness of stamped sheet metal, welding current and voltage, and coating film thickness are collected. After standardization and transformation, a matrix is ​​formed. After the encoder extracts the feature vector, the error between the sheet metal hardness data reconstructed by the decoder and the original data is 0.02. The Pearson correlation coefficient shows that the correlation between welding current and joint strength is 0.92. =0.9 strengthens this association information. After weighted calculation and iterative optimization, the loss function converges to 0.04. The generated unified feature vector successfully integrates the material, process and quality information of each process, providing a standardized and process-oriented data source for subsequent quality assessment.

[0118] This encoding processing module achieves integrated process representation of cross-modal data, overcoming the problem of traditional encoding that only extracts mathematical features and is disconnected from process logic. It enables encoded features to not only include the information of the data itself but also integrate the correlation patterns of automotive manufacturing processes, providing a data source that aligns with actual production for subsequent quality assessment and strategy optimization. This role is not simply multi-data fusion but rather deeply integrates process logic into the encoding process. Secondly, the constructed differentiated weighted encoding system can flexibly adjust weights according to process importance and data type, solving the problem of traditional encoding treating all data equally and diluting core information. This allows encoded features to accurately highlight processes and data that play a key role in product quality. Combined with the customized weighted design of automotive manufacturing process quality weights, it achieves accurate representation of core information. Simultaneously, it possesses adaptive iterative optimization capabilities, flexibly adjusting parameters and convergence thresholds according to different production scenarios and different component types. This adapts to both the efficiency requirements of large-scale mass production and the accuracy requirements of high-end customized products, solving the problem of traditional encoding models being fixed and unable to adapt to diverse manufacturing needs. It adaptively optimizes based on the diverse characteristics of automotive component manufacturing.

[0119] Based on any of the above technical solutions, a further optimization is made to the specific formula for calculating the quality deviation index by the quality assessment module:

[0120] .

[0121] In the formula, Let be the quality deviation index of the m-th process, which is a non-negative real number used to quantify the degree of deviation between the core quality indicators of the m-th process and the standard value.

[0122] , is the real-time detection value of the core quality indicator of the m-th process, is a positive real number, is extracted from a unified feature vector, and represents the actual quality status of the process.

[0123] The value is the industry standard or internal control standard of the core quality indicator of the m-th process. It is a positive real number and serves as the benchmark for process quality assessment.

[0124] 'm' represents the process identifier, 1 represents the stamping process, 2 represents the welding process, and 3 represents the painting process.

[0125] The absolute value sign is used to eliminate the influence of the positive or negative direction of the deviation, retaining only the magnitude of the deviation.

[0126] This solution addresses the issues of different dimensions of quality indicators and inconsistent evaluation standards in the stamping, welding, and painting processes of automotive parts. It breaks through the limitations of traditional quality assessment methods, which rely on absolute error evaluation, independent judgment of single processes, and inability to compare across processes. Instead, it constructs a multi-process quality deviation index calculation model that quantifies relative deviations, handles them without dimensions, and uses unified standards for evaluation.

[0127] By extracting real-time detection values ​​from the unified feature vector of integrated process-related information, the accuracy of quality assessment is ensured. Secondly, by introducing industry standard values ​​or enterprise internal control standard values ​​as assessment benchmarks, the standardization of quality assessment is ensured. Then, the positive and negative influences of deviations are eliminated by absolute values, focusing only on the magnitude of the deviation. Finally, absolute deviations are converted into relative deviations through division operations, achieving dimensionless unified quantification of quality indicators of different dimensions and processes. This allows for direct comparative analysis of the quality status of different indicators such as stamping flatness (mm), welding joint strength (MPa), and coating film thickness (μm), providing a unified quantitative basis for subsequent cross-process collaborative control.

[0128] During the work process, the first step is to determine the core quality indicators for each process—stamping, welding, and painting—based on industry standards and enterprise quality control requirements for automotive parts manufacturing. These include, for example, the flatness of the sheet metal in the stamping process, the joint strength in the welding process, and the uniformity of film thickness in the painting process. Then, the corresponding industry standard values ​​or enterprise internal control standard values ​​for each core quality indicator are retrieved. This serves as the benchmark for quality assessment; subsequently, real-time detection values ​​of core quality indicators for each process are extracted from the unified feature vector of fused process-related information generated by the encoding processing module. This detection value originates from a standardized, unified feature vector, integrating process correlation information within and between processes. It accurately reflects the actual quality status of the process, avoiding the randomness of single, raw detection data. Real-time detection values ​​are calculated using absolute values. Compared with standard value The absolute deviation is determined, eliminating the influence of positive or negative direction and retaining only the magnitude of the deviation; finally, the absolute deviation is divided by the standard value. Dividing by the absolute deviation converts it into a relative deviation, yielding a dimensionless quality deviation index. , The larger the value, the greater the deviation between the core quality indicators of the process and the standard value, thus achieving a unified quantitative assessment of the quality status of each process.

[0129] By using dimensionless design of relative deviation and extraction of detection values ​​based on unified feature vectors, a unified quantitative assessment of quality indicators for multiple processes is achieved. This enables direct comparison and analysis of the quality status of different dimensions and processes, providing a precise and unified quantitative basis for subsequent cross-process collaborative control from the perspective of quality assessment.

[0130] For example, the absolute deviation of 0.05mm in the flatness of stamped sheet metal and the absolute deviation of 20MPa in the strength of welded joints cannot be directly compared. However, the quality deviation indices of 0.06 and 0.08 obtained after converting them into relative deviations can directly determine the degree of deviation in the welding process. By extracting real-time detection values ​​from a unified feature vector that integrates process-related information, the randomness and one-sidedness of single raw detection data are avoided, making the detection values ​​more reflective of the actual quality status of the process and improving the accuracy of quality assessment results. By using industry standard values ​​or enterprise internal control standard values ​​as evaluation benchmarks, the standardization and universality of quality assessment are ensured. Different enterprises and different production scenarios can directly retrieve the corresponding standard values ​​for evaluation without resetting the evaluation system, which has strong practicality.

[0131] The parameter selection is based on the national / industry standards and enterprise internal control standards corresponding to each process of stamping, welding and painting of automotive parts. The setting of each parameter meets the quality control requirements of automobile manufacturing.

[0132] In the stamping process, the industry standard for sheet flatness is ≤0.8mm / m, and the industry standard for sheet size error is ±0.5mm. In the welding process, the industry standard for tensile strength of low carbon steel welded joints is ≥350MPa, and the standard for weld reinforcement is 0-3mm. In the painting process, the industry standard for film thickness of body coating is 80-120μm, and the industry standard for coating color difference is ΔE≤2.0. Enterprises can set stricter internal control standard values ​​according to their own quality control requirements.

[0133] Taking the stamping-welding-painting production scenario of automobile side panels as an example, using the parameter settings of this solution, the enterprise internal control standard value for the flatness of the sheet metal in the stamping process is set to 0.6 mm / m, and the real-time detection value extracted from the unified feature vector is 0.66 mm / m; the industry standard value for the joint strength in the welding process is 350 MPa, and the real-time detection value is 315 MPa; the enterprise internal control standard value for the film thickness in the painting process is 90 μm, and the real-time detection value is 94.5 μm.

[0134] Calculations using the formula show that the quality deviation index for the stamping process is 0.1, for the welding process is 0.1, and for the painting process is 0.05. Based on these quality deviation indices, the stamping and welding processes can be directly identified as having moderate deviations, while the painting process has slight deviations. The strategy optimization module prioritizes adjusting the process parameters for the stamping and welding processes, lowering the stamping pressure and increasing the welding current. After adjustment, a re-inspection shows that the quality status of each process has improved.

[0135] Enterprises can formulate stricter internal control standards based on their own quality control level and product positioning, on the basis of industry standards. For example, the industry standard for sheet metal flatness is ≤0.8mm / m, while enterprises that produce high-end new energy vehicle parts can set their internal control standards to ≤0.6mm / m.

[0136] The extraction of real-time detection values ​​is based on the feature dimension design of a unified feature vector, which conforms to the industrial big data feature extraction and application specifications. Technicians can directly extract the detection values ​​of the corresponding indicators according to the dimensional mapping relationship of the feature vector.

[0137] The classification of deviation levels for the quality deviation index also has clear industry basis. A quality deviation index ≤ 0.05 is defined as slight deviation, 0.05 < quality deviation index ≤ 0.1 is defined as moderate deviation, and quality deviation index > 0.1 is defined as severe deviation. Enterprises can flexibly adjust the classification threshold according to product quality requirements.

[0138] For example, in the stamping-welding-painting production of automobile wheels, the flatness of the stamped sheet metal is first established. =0.7mm / m, welded joint strength =340MPa, coating film thickness The internal control standard value of 85μm was extracted from the unified feature vector, and the real-time detection values ​​were 0.77mm / m, 306MPa, and 89.25μm, respectively. The quality deviation indices were calculated by formula to be 0.1, 0.1, and 0.05, respectively. It was then determined that the stamping and welding processes had moderate deviations, and the painting process had slight deviations, providing a clear direction for quality issues in subsequent strategy optimization.

[0139] Based on any of the above technical solutions, a further optimization is made to the specific steps of the quality assessment module in calculating the quality deviation index:

[0140] The first step is to retrieve the standard thresholds corresponding to the core quality indicators of each process in the stamping, welding, and painting of automotive parts.

[0141] The second step is to extract the real-time detection values ​​of the core quality indicators of each process based on the unified feature vector;

[0142] The third step is to substitute the values ​​into the quality deviation index calculation formula to complete the calculation and generate a quality deviation assessment report.

[0143] This solution changes the traditional design of quality assessment, which involves independent calculation of each process, scattered data sources, and fragmented assessment results. It deeply integrates quality assessment steps with a unified coding technical architecture, using industry standards or enterprise internal control standards as a unified assessment benchmark to ensure the standardization of the assessment basis; using a unified feature vector that integrates process-related information as the data source to ensure the correlation and consistency of test values ​​across processes; and using a multi-dimensional quality deviation assessment report as the output form to integrate the quality deviation indices of each process into a comprehensive report, providing a full-process, visualized quality basis for subsequent strategy optimization. This solves the problems of traditional quality assessment steps being disconnected from actual production processes and assessment results being unable to support cross-process collaborative control.

[0144] The specific steps during operation are as follows: First, the quality assessment module uses the system's built-in industry and enterprise standard databases to automatically retrieve the standard thresholds corresponding to the core quality indicators of each process, such as stamping, welding, and painting, based on the type of automotive parts produced (e.g., body, doors, wheels). These standard thresholds include industry standard values, enterprise internal control standard values, and deviation level classification thresholds, achieving unified and automated retrieval of assessment benchmarks and avoiding the tediousness and errors of manual retrieval.

[0145] The second step involves establishing real-time data interaction between the quality assessment module and the coding processing module. From the unified feature vector of integrated process-related information generated by the coding processing module, the real-time detection values ​​of the core quality indicators of each process are accurately extracted according to the preset feature extraction rules. This extraction process is based on standardized feature dimensions, ensuring the correlation and consistency of the detection values ​​of each process, and avoiding the problems of data disconnect and poor correlation caused by extracting detection values ​​from different raw data in traditional solutions.

[0146] The third step involves the quality assessment module substituting the extracted real-time detection values ​​into the quality deviation index calculation formula to automatically calculate the quality deviation index for each process. Simultaneously, based on preset deviation level classification standards, it determines the quality status of each process. Finally, it integrates information such as the quality deviation index, deviation level, and problem description for each process to generate a multi-dimensional quality deviation assessment report. The report is presented in a combination of visual charts and text descriptions, clearly demonstrating the quality status of each process and cross-process quality correlation issues. This provides an intuitive and comprehensive quality basis for the strategy optimization module to formulate precise process parameter control strategies.

[0147] Compared to traditional manual quality assessment processes, this solution automatically retrieves standard thresholds, avoiding errors and tediousness associated with manual retrieval and significantly improving assessment efficiency. By extracting real-time detection values ​​from a unified feature vector, it ensures the correlation and consistency of detection values ​​across processes, enabling the quality assessment results to reflect the quality transfer patterns between processes and providing a reliable basis for cross-process collaborative control. Furthermore, by generating multi-dimensional quality deviation assessment reports, it integrates scattered quality data from various processes into an intuitive and visual report, facilitating technicians to quickly grasp the quality status of the entire process and formulate targeted rectification strategies.

[0148] The database design of standard thresholds is based on the quality inspection and evaluation criteria for automotive parts. It includes industry standard values ​​and enterprise internal control standard values ​​for common core quality indicators in stamping, welding, and painting processes. It also supports enterprises to add standard thresholds for special parts to meet the quality control needs of different enterprises. The feature extraction rules are set according to the standardized dimensions of a unified feature vector. The core quality indicators of each process correspond to fixed feature dimensions to ensure the accuracy and consistency of the extracted test values. This rule conforms to the general standards for feature extraction in industrial big data.

[0149] The deviation level classification standard references the quality control experience of the automobile manufacturing industry, and includes a quality deviation index. ≤0.05 is set as slight deviation, 0.05 < ≤0.1 is set as moderate deviation. A value greater than 0.1 is set as a severe deviation. This classification standard can accurately reflect the severity of quality problems in each process and provide a clear priority basis for subsequent strategy optimization.

[0150] Taking the stamping-welding-painting production scenario of a car trunk lid as an example, using the evaluation steps and parameter settings of this solution, the first step is that the system automatically retrieves the standard threshold values ​​for sheet flatness (0.5 mm / m) in the stamping process, joint strength (340 MPa) in the welding process, and film thickness (80 μm) in the painting process. The second step is to accurately extract the real-time detection values ​​for sheet flatness (0.53 mm / m), joint strength (310 MPa), and film thickness (86 μm) from a unified feature vector. The third step is to substitute these values ​​into the formula to calculate the quality deviation index of the stamping process. =0.06, Quality Deviation Index of Welding Process =0.09, Quality Deviation Index of Painting Process =0.075, the system determines that stamping and coating are of moderate deviation and welding is of moderate deviation. The generated multi-dimensional quality deviation assessment report clearly shows the cross-process quality correlation problems of stamping sheet flatness deviation, insufficient weld joint strength and uneven coating thickness. Based on this, the strategy optimization module prioritizes the adjustment of process parameters for the welding process.

[0151] Based on any of the above technical solutions, a further optimization is made: the calculation of the process parameter control strategy optimized by the strategy optimization module follows the following formula:

[0152] ;

[0153] In the formula, Let be the objective function of the strategy network for the m-th process, where is a real number. It is the core basis for optimizing the parameters of the strategy network, and the objective is to maximize the value of this function to achieve optimal process control.

[0154] Let be the parameter vector of the policy network for the m-th process. It is a one-dimensional real number vector containing all the weights and bias parameters of the network and is the adjustment object for policy optimization.

[0155] E is the mathematical expectation symbol, used to calculate the mean of random variables and eliminate the influence of random factors in the production process on strategy optimization;

[0156] The clip(.) function is used to restrict the input value to a specified range to prevent the policy update from being too large.

[0157] Let be the strategy update ratio for the m-th process, which is a positive real number representing the degree of improvement in returns of the current strategy relative to the historical strategy; The truncation parameter is set to a fixed value of 0.2, which is a positive real number, and is used to define the reasonable range for policy updates.

[0158] Let be the advantage function of the m-th process, and be a real number that quantifies the improvement potential of the current process parameters relative to the historical best parameters.

[0159] The process stability weight is a positive real number, with 0.015 for stamping, 0.01 for welding, and 0.012 for painting, used to balance the exploratory nature of process parameters with production stability;

[0160] exp(.) is the natural exponential function, used to convert the quality deviation index into a constraint coefficient on the strategy entropy;

[0161] The quality sensitivity coefficient is a positive real number, with 1.5 for stamping, 1.8 for welding, and 1.3 for painting, representing the degree of influence of process quality deviation on process control.

[0162] The quality deviation index for the m-th process;

[0163] Let be the strategy entropy of the strategy network for the m-th process, which is a non-negative real number, to ensure the diversity of process parameter adjustments and adapt to fluctuations in raw material characteristics.

[0164] The strategy distribution for the m-th process is given by the parameter vector. The determination is based on the probability distribution of the adjustment of process parameters.

[0165] By limiting the policy update range through a truncation function, production stability is ensured; secondly, process stability weights are used. With quality sensitivity coefficient To achieve differentiated control of each process, the welding process is highly sensitive to quality, while the stamping process has high stability requirements. By using the natural exponential function, the quality deviation index is transformed into the constraint coefficient of the strategy entropy, so that the control of process parameters is driven by the quality deviation. Furthermore, the strategy entropy ensures the diversity of parameter adjustments, adapts to the fluctuation of raw material characteristics, and solves the problems of blind optimization and control, poor stability, and inability to adapt to the differentiated needs of multiple processes in traditional processes.

[0166] During the work process, the first step is to set up [the necessary parameters] based on the technological characteristics and quality control requirements of each process. , The initial values ​​of ϵ are set, with ϵ taking the common industrial optimization algorithm value of 0.2. Subsequently, the strategy optimization module receives the quality deviation index output by the quality assessment module, as well as the advantage function of each process. Strategy update ratio Wait for data; update the policy ratio using the clip truncation function. The optimization is limited to the range of 0.8-1.2 to avoid excessively large strategy update fluctuations that could lead to production instability. The mathematical expectation E eliminates the impact of random factors such as raw material fluctuations and equipment precision errors on strategy optimization. The natural exponential function transforms the quality deviation index into a constraint coefficient of the strategy entropy; the larger the quality deviation, the smaller the constraint coefficient, the stronger the constraint on the strategy entropy, and the more focused the parameter adjustments are on solving quality problems. The exploration and stability of strategy optimization are balanced by weighting process stability, resulting in high efficiency in the stamping process. To make parameter adjustments more stable, and to improve welding processes This strengthens the driving effect of quality deviation on regulation; strategy entropy To ensure the diversity of parameter adjustments and adapt to the characteristic fluctuations of different batches of raw materials; finally, the objective function is used. Maximizing the parameter vector of the policy network is the core objective; iterative optimization is performed on this parameter vector. The process continues until the objective function converges, generating the optimal process parameter control strategy for each process. This strategy ensures the effectiveness of process parameter adjustment while also taking into account the stability and adaptability of production.

[0167] The truncation parameter ϵ is set to 0.2, a common value for the Proximal Policy Optimization (PPO) algorithm. This value ensures policy update efficiency while avoiding production instability caused by parameter mutations, thus meeting the stability requirements of industrial control. The process stability weight... The stamping process is assigned a weight of 0.015, the welding process 0.01, and the painting process 0.012. This is because the stamping process, where sheet metal forming is sensitive to sudden parameter changes, requires a higher stability weight; the welding process needs to balance stability with quality rectification efficiency; and the painting process has higher parameter adjustment flexibility, resulting in a moderate weight. The quality sensitivity coefficient... The welding process is set to 1.8, the stamping process to 1.5, and the painting process to 1.3. According to the guidelines for classifying the quality importance of automotive parts manufacturing processes, the welding process is a critical safety process with the highest quality sensitivity, the stamping process is a basic forming process, and the painting process is an appearance protection process, with the sensitivity decreasing in that order. The initial value of the strategy entropy is set to the industry-standard range of 0.5-1.0 to ensure the diversity of parameter adjustments.

[0168] Taking the stamping-welding-painting production scenario of automobile frames as an example, the parameter settings of this solution are as follows: Stamping takes 0.015, welding takes 0.01, and painting takes 0.012. Welding is rated at 1.8, stamping at 1.5, and painting at 1.3, with ϵ=0.2; a severe quality deviation was detected in the welding process. =0.12), the output value of the natural exponential function is greatly reduced, the constraint on the strategy entropy is strengthened, the parameter adjustment focuses on the precise optimization of welding current and voltage, and clip(.) is a cutoff function to limit the current adjustment range to no more than 20% to avoid welding production instability; after iterative optimization, the objective function converges, the welding current adjustment, voltage adjustment, and welding joint strength are improved, the quality deviation index decreases, and the parameter adjustment of the stamping and painting processes is also adapted to the fluctuation of raw material characteristics, thus improving the yield of chassis production.

[0169] This strategy optimization module constructs a quality deviation-driven parameter adjustment mechanism, breaking through the limitations of traditional fixed-rule optimization. It transforms quality problems into strategy constraints through a natural exponential function, enabling parameter adjustments to automatically focus on quality defects, such as prioritizing current and voltage adjustments for severe welding deviations. Simultaneously, it achieves differentiated process control, addressing different needs such as stamping stability, welding quality sensitivity, and coating flexibility. Compared to traditional fixed-rule process optimization schemes, this solution limits the strategy update amplitude through a truncation function, avoiding production failures caused by sudden parameter changes. and The differentiated settings allow the control strategies for each process to be tailored to its technological characteristics, resulting in high-precision welding processes. This allows for the rapid resolution of quality issues and improves the efficiency of the stamping process. This makes production more stable; by combining quality deviation-driven and strategy entropy, parameter adjustments can focus on quality issues while also having the flexibility to adapt to raw material fluctuations.

[0170] The process parameter adjustment instructions issued by the execution control module are as follows: for the stamping process, the stamping pressure adjustment range is ±5%, the stamping speed adjustment range is ±3%, and the mold temperature adjustment range is ±2℃; for the welding process, the welding current adjustment range is ±10A, the welding voltage adjustment range is ±0.5V, the welding speed adjustment range is ±0.2mm / s, and the protective gas flow rate adjustment range is ±5L / min; for the coating process, the spraying pressure adjustment range is ±0.1MPa, the spraying distance adjustment range is ±5mm, the coating viscosity adjustment range is ±5mPa・s, and the drying temperature adjustment range is ±3℃.

[0171] To address the core impacts of process parameters in each step, differentiated adjustment ranges are designed. Stamping pressure affects sheet metal forming, so a slightly larger adjustment range is used, while mold temperature has a smaller adjustment range to ensure forming accuracy. Welding current affects weld penetration, so a slightly larger adjustment range is used, while voltage has a very small adjustment range to ensure current-voltage matching. Coating spraying pressure and distance directly affect film thickness, and small, precise adjustment ranges are designed for both to avoid fluctuations in film quality and to solve the problems of quality issues and production instability caused by excessive adjustment of traditional execution control parameters. The execution control module receives the optimal process parameter adjustment strategy generated by the strategy optimization module. Based on the process characteristics and adjustment range limitations of each process, it transforms the adjustment strategy into specific process parameter adjustment instructions. These instructions are then sent to the production equipment control systems of each process, such as stamping CNC systems, welding robot control systems, and painting automation systems, via industrial Ethernet. The equipment completes the precise adjustment of process parameters according to the instructions. Simultaneously, the execution control module acquires the actual parameter operating values ​​of the equipment in real time through the data acquisition module, compares them with the issued adjustment instructions, and monitors the instruction execution status. If the deviation between the actual parameters and the instructions exceeds a preset threshold (e.g., ±1%), the equipment immediately triggers self-correction. If the deviation still exceeds the range after correction, an early warning is issued and production of that process is suspended to prevent the batch production of defective products.

[0172] Based on any of the above technical solutions, the following further optimization is made: the process parameter adjustment instructions issued by the execution control module are specifically: the stamping pressure adjustment range of the stamping process is ±5%, the stamping speed adjustment range is ±3%, and the mold temperature adjustment range is ±2℃.

[0173] The welding current adjustment range for the welding process is ±10A, the welding voltage adjustment range is ±0.5V, the welding speed adjustment range is ±0.2mm / s, and the shielding gas flow rate adjustment range is ±5L / min.

[0174] The spraying pressure adjustment range for the coating process is ±0.1MPa, the spraying distance adjustment range is ±5mm, the paint viscosity adjustment range is ±5mPa・s, and the drying temperature adjustment range is ±3℃.

[0175] Example 2: Compared with Example 1, this example also includes the following technical features:

[0176] The probability calculation for locating the responsible process for quality defects by the defect tracing module follows the formula below:

[0177] .

[0178] In the formula, The conditional probability has a value range of [0,1], representing the posterior probability that the defect D is caused by the m-th process, given that a quality defect D is detected. The larger the probability value, the higher the probability that the process is the responsible process.

[0179] The weights for each process are positive real numbers, with 0.35 for stamping, 0.4 for welding, and 0.25 for painting. These weights are derived from historical defect data and represent the probability weights of each process in causing quality defects.

[0180] is the cosine similarity, with a value range of [0,1]. is the feature similarity between the current defect D and the historical defect data Dmhist of the m-th process, calculated based on a unified feature vector, representing the degree of matching of defect features.

[0181] is the material property influence coefficient for the m-th process, with a value range of [0.8, 1.2]. It is a positive real number, calculated from the material property data of the corresponding process, and characterizes the degree of influence of material properties on the formation of process defects.

[0182] D represents the currently detected quality defect feature data of automotive parts. It is a one-dimensional real number vector containing core features such as defect type, defect severity, and defect location.

[0183] is the historical defect data for the m-th process, which is a two-dimensional real number matrix that stores the feature information of all past defects in this process; j is the process identifier, with values ​​of 1, 2, and 3, corresponding to stamping, welding, and painting processes, respectively, and is a summation and traversal variable.

[0184] The summation symbol sums the probability calculation terms of the three processes, thereby normalizing the probabilities and ensuring that the sum of the posterior probabilities of each process is 1.

[0185] Based on the technological logic of automotive parts manufacturing, defects in the welding process have a significant impact, thus assigning it the highest contribution weight. The stamping process is fundamental, with the next highest weight; defects in the painting process are mostly localized problems, with the lowest weight; cosine similarity is used to determine the weight. Achieve precise matching of defect features and avoid subjective judgment errors; utilize the influence coefficient of material properties. By incorporating material factors into the traceability process and aligning with the defect formation mechanism of each process, the problem of traditional traceability being unable to locate cross-process defects and having inaccurate traceability results can be solved.

[0186] Compared to traditional traceability schemes, this scheme achieves objectivity through probability calculations, avoiding the subjectivity of manual judgment; it achieves accurate identification of defect features through cosine similarity matching, improving traceability accuracy; it incorporates material factors into traceability by using material characteristic influence coefficients, solving the problem of traditional traceability ignoring material influences; and it reflects the probability of defects occurring in each process by using process contribution weights, making the traceability results more closely reflect actual production. Furthermore, the traceability model can continuously improve traceability accuracy through iterative optimization of historical defect data.

[0187] During operation: First, extract the feature data of the current quality defect D, including core information such as defect type, severity, and location; then retrieve historical defect data for each process from the system's historical defect database. ;Calculate the current defect D and historical defect data for each process based on a unified feature vector. The cosine similarity is used; the closer the similarity is to 1, the higher the matching degree.

[0188] Calculate the material property influence coefficient based on the material property data of each process. Defects in stamped sheet material Take 1.1-1.2, for defects in the composition of the base metal used for welding. Take a score of 1.0-1.2 for cleanliness defects in coated panels. Set the value to 0.8-1.0; assign the process contribution weight. Cosine similarity Material property influence coefficient Multiply the results to obtain the defect correlation score for each process; sum the correlation scores of the three processes using a summation operator, divide the correlation score of each process by the total score to normalize the probability, and obtain the posterior probability that each process causes the current defect. The process with the highest probability value is the responsible process. If the cross-process correlation scores are all high, it is judged as a cross-process related defect.

[0189] Taking the defect tracing scenario of uneven coating thickness on automotive doors as an example, the parameter settings of this solution are adopted. The values ​​are 0.35 for stamping, 0.4 for welding, and 0.25 for painting. The cosine similarity S between the current defect D and the historical defect data of the stamping process is 0.85. For stamping, the value is 1.2; for welding, S = 0.3. =1.0, paint S=0.7, =0.9; the calculated post-test probability for stamping is 0.6, for painting 0.35, and for welding 0.05. Therefore, stamping is determined to be the responsible process due to insufficient flatness of the stamped sheet material. =1.2) This leads to uneven coating thickness in subsequent processes. The traceability results match the actual production situation. Based on this, the parameter correction module makes directional corrections to the coding coefficients of the stamping process, which can improve the uniformity of coating thickness in subsequent production of car doors.

[0190] This solution achieves multi-factor probabilistic traceability, overcoming the limitations of traditional single-feature traceability. It integrates process weight, feature similarity, and material influence to solve the problem of difficult cross-process defect traceability, such as uneven stamping sheet leading to uneven coating thickness. It also introduces a material characteristic influence coefficient to fill the gap in traditional traceability that ignores material factors, such as defects in the composition of the welding base material leading to insufficient joint strength.

[0191] Simultaneously, an iteratively optimized traceability system is constructed to improve matching accuracy through the accumulation of historical data. For example, if a certain vehicle model has a long-term problem with welding porosity, the system will automatically adjust the welding accuracy accordingly. and .

[0192] Based on any of the above technical solutions, a further optimization is made to the specific steps of the parameter correction module in correcting parameters: dynamically weighting the process-data corresponding to the defect responsibility process. Improve by 20%, adjust the material-data fit coefficient based on defect type. Adjust by ±10%; increase the update step size of the strategy network parameters corresponding to the defect responsibility process by 1.5 times, and iterate and optimize the process parameter control strategy again.

[0193] In this technical solution, defect rectification should focus on the responsible process and avoid interfering with non-responsible processes; by improving the responsible process... This ensures that subsequent coding focuses on representing the data from that process; adjustments are made based on the defect type. Material-related defects are adjusted upwards, while non-material-related defects are adjusted downwards. The update step size of the strategy network is increased to accelerate parameter optimization for responsible processes, resolving the problems of blind and inefficient traditional parameter correction and rectification. Compared to traditional parameter correction schemes, this solution improves rectification accuracy by targeting parameters for responsible processes; avoids coding distortion or production instability caused by large-scale parameter corrections through small adjustments to coefficients; accelerates rectification efficiency by increasing the strategy step size; and ensures overall production stability by not interfering with non-responsible processes during the correction process. Furthermore, the correction range can be flexibly adjusted according to the severity of defects to adapt to different rectification needs.

[0194] Taking the rectification of welding defects in an automotive engine bracket as an example, the defect source is determined to be the welding process, and the defect type is material-related (base material composition is unqualified). Using the rectification method described in this solution, the welding process... Increase by 20%, The algorithm was increased by 10%, and the update step size of the strategy network was increased by 1.5 times. After the quality stabilized, it was restored to 1.0 times. After the correction, the algorithm was iterated and optimized again. The process parameter control strategy of the welding process is more focused on the adaptation of the base material composition, the adjustment of welding current and voltage, the improvement of weld joint strength, and the reduction of defect rate.

[0195] This technical solution constructs a responsibility-based process-oriented correction system, breaking through the limitations of traditional full-process correction. It only adjusts the parameters of the responsible process; for example, welding defects are corrected only in the welding process. and This avoids parameter fluctuations in non-responsible processes (such as stamping); it also enables small, precise corrections, resolving coding distortion and production instability issues caused by traditional large-scale corrections. Increase by 20%, α Adjust by ±10% to ensure a balance between the effectiveness of rectification and system stability. Improve the timeliness of rectification by increasing the strategy step size (1.5 times) for targeted correction and optimizing the step size to accelerate iteration.

[0196] Based on any of the above technical solutions, the following further optimization is made: the specific rules for triggering the iterative optimization module to execute each module in a loop are as follows: each production batch consists of 50 to 100 automotive parts, and the entire process module linkage is triggered once after the production of a batch is completed; if a severe quality defect is detected during the production process, the defect tracing module and parameter correction module are immediately triggered to perform their functions, the parameters are corrected, production continues, and the number of production batches is dynamically adjusted according to the optimization effect.

[0197] Regular iterations are conducted in batches of 50-100 pieces, balancing optimization timeliness and production efficiency; emergency iterations target severe defects, enabling rapid source tracing and rectification to prevent batch defects from occurring; the batch size is dynamically adjusted to balance optimization frequency and production efficiency based on optimization results, solving the problem of traditional iterative optimization's slow response to sudden quality issues and inability to adapt to production changes.

[0198] The iterative optimization module defaults to producing 50-100 pieces per batch. After the batch production is completed, it automatically triggers the linkage of the entire process modules, including data acquisition, coding processing, quality assessment, strategy optimization, execution control, defect tracing, parameter correction, and iterative optimization, to achieve routine optimization of the process.

[0199] If a severe quality defect is detected during production, the defect tracing module and parameter correction module are immediately triggered to quickly locate the responsible process and correct the parameters. Production continues after correction. The batch quantity is dynamically adjusted based on the optimization effect. If the product yield improves and stabilizes, the batch quantity is increased to 80-100 pieces, reducing the iteration frequency. If the quality fluctuates significantly, the batch quantity is decreased to 50-70 pieces, increasing the iteration frequency, thus achieving a dynamic balance between iterative optimization and production efficiency.

[0200] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention. For those skilled in the art, any alternative improvements or transformations made to the implementation of the present invention fall within the protection scope of the present invention.

[0201] Any aspects of this invention not described in detail are well-known to those skilled in the art.

Claims

1. An automatic coding intelligent agent control system for multi-task scenarios, characterized in that, It includes a data acquisition module, an encoding processing module, a quality assessment module, a strategy optimization module, an execution control module, a defect tracing module, a parameter correction module, and an iterative optimization module; The data acquisition module includes a stamping process acquisition unit for acquiring quality inspection, process parameters, and material characteristic data of the stamping process; a welding process acquisition unit for acquiring quality inspection, process parameters, and material characteristic data of the welding process; and a coating process acquisition unit for acquiring quality inspection, process parameters, and material characteristic data of the coating process. The encoding processing module is used to perform unified encoding processing on the collected multi-process data to generate a unified feature vector that integrates process-related information; The quality assessment module is used to calculate the quality deviation index of the core quality indicators of each process based on a unified feature vector. The strategy optimization module is used to integrate multi-source information to optimize the process parameter control strategy for each process step; The execution control module is used to issue process parameter adjustment instructions and monitor the execution status of the instructions; The defect tracing module is used to locate the responsible process for quality defects and the cross-process influencing factors; The parameter correction module is used to correct the code and adjust the relevant parameters based on the defect tracing results. The iterative optimization module is used to trigger the cyclic execution function of each module according to the production batch; Each module and unit establishes a data interaction connection via industrial Ethernet and achieves coordinated operation. The encoding processing module performs unified encoding processing calculations according to the following formula: ； in, Here is the encoding loss function; in the double summation symbol, the summation range of k is 1 to 3, and the summation range of m is 1 to 3; For process-data dynamic weighting coefficients; Material-data adaptation coefficient; The square of the L2 norm is used to quantify the data error of the original data of type k in the m-th process after being encoded by the encoder and reconstructed by the decoder, so as to ensure the consistency between the encoded features and the original data. This is the matrix form of the k-th type of data in the m-th process; This is the mapping function for the encoder network; This is the mapping function for the decoder network; This refers to the process correlation coefficient; Pearson correlation coefficient; This is the process parameter matrix for the m-th process step.

2. The automatic coding intelligent agent control system for multi-task scenarios according to claim 1, characterized in that... The specific data collected by the stamping process acquisition unit includes: quality inspection data including sheet flatness images and sheet size error values; process parameter data including the time-series changes in stamping pressure, stamping speed, and die temperature; and material property data including the hardness, tensile strength, and thickness uniformity test data of the stamped sheet. The specific data collected by the welding process acquisition unit includes: quality inspection data including weld joint strength test values ​​and weld formation images; process parameter data including time-series variation data of welding current, welding voltage, welding speed, and shielding gas flow rate; and material characteristic data including test data of welding base material composition and weld joint metallographic structure. The specific data collected by the coating process acquisition unit includes: quality inspection data such as coating film thickness distribution images and coating color difference detection values; process parameter data such as time-series changes in spraying pressure, spraying distance, paint viscosity, and drying temperature; and material characteristic data such as test data on the surface cleanliness of the coated sheet and paint adhesion.

3. The automatic coding intelligent agent control system for multi-task scenarios according to claim 1, characterized in that, The specific formula for calculating the quality deviation index in the quality assessment module is as follows: ； In the formula, The quality deviation index for the m-th process; This represents the real-time detection value of the core quality indicator for the m-th process; The m represents the industry standard value or enterprise internal control standard value of the core quality indicator of the m-th process; m is the process identifier, where 1 is the stamping process, 2 is the welding process, and 3 is the painting process.

4. The automatic coding intelligent agent control system for multi-task scenarios according to claim 1, characterized in that, The specific steps for the quality assessment module to calculate the quality deviation index are as follows: The first step is to retrieve the standard thresholds corresponding to the core quality indicators of each process in the stamping, welding, and painting of automotive parts. The second step is to extract the real-time detection values ​​of the core quality indicators of each process based on the unified feature vector; The third step is to substitute the values ​​into the quality deviation index calculation formula to complete the calculation and generate a quality deviation assessment report.

5. The automatic coding intelligent agent control system for multi-task scenarios according to claim 1, characterized in that, The strategy optimization module calculates the optimization strategy for process parameter control according to the following formula: ； In the formula, Let m be the objective function of the strategy network for the m-th process. Let be the parameter vector of the strategy network for the m-th process; E is the mathematical expectation symbol, used to calculate the mean of the random variable; This is a truncation function used to restrict input values ​​to a specified range, preventing excessive policy updates. The strategy update ratio for the m-th process; Let m be the dominance function of the m-th process; The process stability weight is exp(.), which is the natural exponential function. This is the quality sensitivity coefficient; The quality deviation index for the m-th process; Let be the policy entropy of the policy network for the m-th process; The strategy distribution for the m-th process is given by the parameter vector. The determination is based on the probability distribution of the adjustment of process parameters.

6. The automatic coding intelligent agent control system for multi-task scenarios according to claim 1, characterized in that, The process parameter adjustment instructions issued by the execution control module are specifically as follows: The stamping pressure adjustment range for the stamping process is ±5%, the stamping speed adjustment range is ±3%, and the die temperature adjustment range is ±2℃. The welding current adjustment range for the welding process is ±10A, the welding voltage adjustment range is ±0.5V, the welding speed adjustment range is ±0.2mm / s, and the shielding gas flow rate adjustment range is ±5L / min. The spraying pressure adjustment range for the coating process is ±0.1MPa, the spraying distance adjustment range is ±5mm, and the paint viscosity adjustment range is ± The drying temperature can be adjusted within ±3℃.

7. The automatic coding intelligent agent control system for multi-task scenarios according to claim 1, characterized in that, The probability calculation for locating the responsible process for quality defects by the defect tracing module follows the formula below: ； In the formula, This is the conditional probability, and its value ranges from [0,1]. Each process contributes a weight to the process, representing the probability weight of each process causing quality defects; The cosine similarity value ranges from [0,1]. denoted as the material property influence coefficient for the m-th process; D represents the currently detected quality defect characteristic data of automotive parts. is the historical defect data for the m-th process, which is a two-dimensional real number matrix; j is the process identifier, with values ​​of 1, 2, and 3, corresponding to stamping, welding, and painting processes respectively, and is a summation and traversal variable.

8. The automatic coding intelligent agent control system for multi-task scenarios according to claim 1, characterized in that, The specific rules for triggering the iterative optimization module to execute each module in a loop are as follows: Each production batch consists of 50 to 100 automotive parts. Once a batch is completed, the entire process module is activated. If a serious quality defect is detected during production, the defect tracing module and parameter correction module are immediately activated. After the parameters are corrected, production continues and the number of production batches is dynamically adjusted based on the optimization effect.

Images