Integrated intelligent management and control method and system for automobile manufacturing system based on digital twinning

Abstract

This application relates to the field of intelligent manufacturing technology and discloses an integrated intelligent control method and system for automobile manufacturing systems that integrates digital twins. The method includes: constructing a cross-process error propagation sensitivity matrix S using workpiece numbers as indices; introducing the matrix S into the control objectives of each process's digital twin to solve for the cross-process collaborative process parameter combination and issuing it for execution; obtaining the error attribution contribution rate through inverse decomposition of the matrix S; storing the production condition feature vector and batch-to-batch variations in a drift scenario knowledge base and performing feedforward correction to obtain a feedforward correction sensitivity matrix; and using the feedforward correction sensitivity matrix to solve for the cross-process collaborative process parameter combination for the current batch. This application achieves batch-level feedforward adaptive correction of cross-process collaborative control and error propagation laws.

Description

Technical Field

[0001] This application relates to the field of intelligent manufacturing technology, and in particular to an intelligent control method and system for automobile manufacturing system integration that incorporates digital twins. Background Technology

[0002] The automotive manufacturing industry generally adopts a sequential production model consisting of four major processes: stamping, welding, painting, and final assembly. These processes have strong sequential dependencies, with workpieces flowing sequentially between them. The processing results of upstream processes directly determine the incoming material status of downstream processes. With the development of intelligent manufacturing technology, digital twin technology is gradually being introduced into automotive manufacturing systems. By constructing corresponding digital twins for each process, real-time monitoring, simulation prediction, and control decisions of the production process can be achieved. Each process's digital twin builds a simulation model based on sensor data collected for that process. With the optimal quality indicators for that process as the control objective, the process parameters for that process are optimized and then executed, forming a closed-loop control system within the process.

[0003] However, existing automotive manufacturing control technologies integrating digital twins have fundamental flaws: the digital twins for each process operate independently, focusing solely on achieving optimal quality within their own process. They fail to recognize the propagation of error parameters from their current process to downstream processes. This leads to upstream processes, in their pursuit of optimal quality, actually locking errors into the workpiece as residual stress and geometric deviations, which are then passed on to downstream processes. For example, the forced clamping strategy used in welding to compensate for deformation of stamped parts may achieve the welding process's quality standards, but it locks a large amount of residual stress into the car body. During the painting and baking stage, this stress release causes secondary deformation of the car body, ultimately resulting in irreversible quality defects such as excessive door and hood gaps during final assembly. At this point, it is impossible to correct these defects by returning to upstream processes, necessitating scrapping or costly rework. Furthermore, the error parameter data for each process twin is stored independently in its own system, lacking cross-process data linking and causal relationship modeling. Quality defects discovered during final assembly cannot be quantitatively traced back to specific error parameters in upstream processes, and control decisions lack quantitative cross-process basis. Summary of the Invention

[0004] This application provides an integrated intelligent control method and system for automobile manufacturing systems that integrates digital twins. It solves the problems in existing digital twin automobile manufacturing control systems, such as isolated optimization of each process leading to unidirectional error propagation lock-up across processes, and static lag of the error propagation model which cannot feed forward to respond to changes in production conditions. It realizes cross-process collaborative control and batch-level feedforward adaptive correction of error propagation laws.

[0005] Firstly, this application provides an intelligent control method for automotive manufacturing system integration that integrates digital twins, the method comprising:

[0006] Step S1: Using the workpiece number as an index, connect all the production data of the four major processes into a full-process dataset, and construct a cross-process error propagation sensitivity matrix S based on the full-process dataset;

[0007] Step S2: Introduce the cross-process error propagation sensitivity matrix S into the control target of the digital twin of each upstream process, solve for the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value, and send the cross-process collaborative process parameter combination to the corresponding physical process for execution;

[0008] Step S3: Perform a full inspection of the key quality parameters of the vehicle body after the final assembly process to obtain the final assembly quality measured vector. Subtract the final assembly quality measured vector from the final assembly quality predicted value to obtain the final assembly quality residual vector. Decompose the final assembly quality residual vector in reverse through the cross-process error propagation sensitivity matrix S to obtain the error attribution contribution rate of each upstream process.

[0009] Step S4: Store the production condition feature vector of each batch and the batch-to-batch variation of the cross-process error propagation sensitivity matrix S of the corresponding batch into the drift scenario knowledge base; before the start of production of a new batch, the drift scenario knowledge base retrieves the historical batch that is closest to the production condition feature vector of the current batch, and uses the batch-to-batch variation of the historical batch to perform feedforward correction on the current cross-process error propagation sensitivity matrix S to obtain the feedforward correction sensitivity matrix. The digital twins of each upstream process are introduced into the control target with the feedforward correction sensitivity matrix to solve for the cross-process collaborative process parameter combination of the current batch.

[0010] Secondly, this application provides an intelligent control system for automotive manufacturing systems integrating digital twins, the intelligent control system for automotive manufacturing systems integrating digital twins includes:

[0011] The concatenation module is used to concatenate the full-element production data of the four major processes into a full-process dataset using the workpiece number as an index, and to construct a cross-process error propagation sensitivity matrix S based on the full-process dataset;

[0012] The solution module is used to introduce the cross-process error propagation sensitivity matrix S into the control target of the digital twin of each upstream process, solve for the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value, and send the cross-process collaborative process parameter combination to the corresponding physical process for execution.

[0013] The decomposition module is used to perform full inspection of the key quality parameters of the vehicle body after the final assembly process to obtain the final assembly quality measured vector. The final assembly quality residual vector is obtained by subtracting the final assembly quality measured vector from the final assembly quality predicted value. The final assembly quality residual vector is decomposed in reverse by the cross-process error propagation sensitivity matrix S to obtain the error attribution contribution rate of each upstream process.

[0014] The correction module is used to store the production condition feature vector of each batch and the batch-to-batch change of the cross-process error propagation sensitivity matrix S of the corresponding batch into the drift scenario knowledge base. Before the start of production of a new batch, the drift scenario knowledge base retrieves the historical batch that is closest to the production condition feature vector of the current batch, and uses the batch-to-batch change of the historical batch to perform feedforward correction on the current cross-process error propagation sensitivity matrix S to obtain the feedforward correction sensitivity matrix. The digital twins of each upstream process are introduced into the control target with the feedforward correction sensitivity matrix to solve for the cross-process collaborative process parameter combination of the current batch.

[0015] Thirdly, an intelligent control device for automotive manufacturing system integration that integrates digital twins is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the intelligent control device for automotive manufacturing system integration that integrates digital twins to execute the aforementioned intelligent control method for automotive manufacturing system integration that integrates digital twins.

[0016] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the aforementioned intelligent control method for automotive manufacturing system integration fused with digital twins.

[0017] In the technical solution provided in this application, the production data of all four major processes are linked together into a full-process dataset using the workpiece number as an index. Based on partial least squares regression, a cross-process error propagation sensitivity matrix is ​​constructed. After normalization of each column element of the matrix, the relative influence weight of each dimension error parameter in the three upstream processes of stamping, welding, and painting on the final assembly door cover gap value, surface difference value, and sealing strip compression amount is directly quantified. This breaks the limitation of independent storage of data for each process and the inability to quantify the error propagation relationship in the prior art, and makes the cross-process error propagation law expressed in a computable matrix form for the first time. Based on this, a cross-process error propagation sensitivity matrix is ​​introduced into the control objectives of the digital twins of each upstream process. The optimization objective is to minimize the comprehensive impact of the error parameters of each upstream process on the final assembly quality parameters after being weighted by this matrix. Under the constraint of the quality qualification domain, the cross-process collaborative process parameter combination is solved. This fundamentally changes the isolated optimization logic of each process twin in the existing technology, which takes the optimal quality of its own process as the control objective. It enables the upstream process to actively select the process parameter combination that has the least impact on the downstream process while ensuring the quality qualification of its own process, thus cutting off the chain of error propagation along the process chain and locking it.

[0018] Furthermore, this application obtains the final assembly quality residual vector by subtracting the final assembly quality measured vector from the final assembly quality predicted value, and obtains the error attribution contribution rate of each upstream process by inverse decomposition of the cross-process error propagation sensitivity matrix. This realizes quantitative traceability from the final assembly quality result to the error parameters of the upstream process, enabling the control system to have the ability to accurately attribute the source of quality deviation, which is different from the single function of the final assembly quality inspection result in the prior art, which is only used to determine whether the product is qualified or not. Driven by this attribution result, this application constructs a drift scenario knowledge base, continuously accumulating the batch-to-batch changes of the production condition feature vector and the corresponding batch cross-process error propagation sensitivity matrix in the form of key-value pairs. Before the start of a new batch of production, historical batches are retrieved based on the similarity of production conditions. The current cross-process error propagation sensitivity matrix is ​​then feedforward corrected using the weighted average of the changes between corresponding batches in the historical batches, resulting in a feedforward corrected sensitivity matrix. This matrix then drives the digital twins of each upstream process to solve and distribute the cross-process collaborative process parameter combination for the current batch, moving the control intervention node forward to before the final assembly quality deviation of the current batch occurs, thus realizing a paradigm shift from passive feedback control to active feedforward control. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1This is a schematic diagram of an embodiment of the intelligent control method for automobile manufacturing system integration that incorporates digital twins, as described in this application.

[0021] Figure 2 This is a schematic diagram of an embodiment of the intelligent control system for automobile manufacturing systems that integrates digital twins, as described in this application.

[0022] Figure 3 This is a schematic block diagram of the integrated intelligent control equipment for automobile manufacturing systems that incorporates digital twins, as described in this embodiment of the invention. Detailed Implementation

[0023] This application provides an intelligent control method and system for automotive manufacturing systems integrating digital twins. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0024] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the intelligent control method for automotive manufacturing systems integrating digital twins in this application includes:

[0025] Step S1: Using the workpiece number as an index, connect all the production data of the four major processes into a full-process dataset, and construct a cross-process error propagation sensitivity matrix S based on the full-process dataset.

[0026] It is understood that the executing entity of this application can be an intelligent control system integrating digital twins in automotive manufacturing systems, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0027] Specifically, the full-element production data for the four major processes includes: stamping process data such as stamping pressure, die clearance, sheet metal thickness deviation, and springback; welding process data such as weld point position deviation, clamping force, welding current, and body positioning reference offset; painting process data such as baking temperature, body surface stress distribution, and coating thickness uniformity; and final assembly process data such as door / hood gap value, surface difference value, and sealing strip compression. The reason for aligning and concatenating these data one by one using the workpiece number as an index is that the same workpiece flows sequentially through the four processes, and the data from each process only have a causal relationship after being aligned under the same workpiece number. Data splicing across workpieces does not reflect the true error propagation relationship. The cross-process error propagation sensitivity matrix is ​​constructed using partial least squares regression instead of ordinary least squares regression because there is significant multicollinearity between stamping springback and sheet metal thickness deviation, and between welding clamping force and positioning reference offset. Ordinary least squares regression has unstable regression coefficients and ambiguous physical meanings under multicollinearity conditions. Partial least squares regression eliminates the influence of multicollinearity by extracting latent components, giving each regression coefficient a stable and interpretable physical meaning. After normalizing each column of the regression coefficient matrix by absolute value, the sum of the elements in each column is one. Its physical meaning is that among all upstream error parameters that affect the same final assembly quality index, the sum of the relative contribution ratios of each parameter is 100%, so that error parameters of different processes and different dimensions have a weight relationship that can be directly compared on the same scale.

[0028] Step S2: Introduce the cross-process error propagation sensitivity matrix S into the control target of the digital twin of each upstream process, solve for the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value, and send the cross-process collaborative process parameter combination to the corresponding physical process for execution.

[0029] Specifically, the logic for introducing the cross-process error propagation sensitivity matrix into the control objectives of the digital twin of each upstream process is as follows: Each dimension of the error parameters of the three upstream processes—stamping, welding, and painting—is multiplied by the weight value of the corresponding row in the cross-process error propagation sensitivity matrix, and then summed to obtain the weighted comprehensive influence of each upstream process error on three quality indicators: the gap value of the final assembly door cover, the surface difference value, and the compression amount of the sealing strip. The optimization objective is to minimize this weighted comprehensive influence. The quality qualification domain constraint refers to the pre-set allowable range of process parameters and the qualification judgment boundary of quality indicators within each process. For example, the qualification domain constraint for the welding process is that the weld point position deviation does not exceed 0.5 mm and the torsional stiffness of the vehicle body is not less than 18,000 Nm per degree. The solution is performed within this constraint range to ensure that while the upstream process pursues the minimum error of the entire process chain, the quality of this process does not fall below the qualification standard. The fundamental difference between the control objectives of this invention and existing technologies lies in the following: In existing technologies, the digital twin of the welding process aims to achieve the highest quality score in the welding process and selects the maximum clamping force strategy to compensate for the deformation of the stamped parts. Although this ensures that the internal quality of the welding process meets the standards, it locks a large amount of residual stress into the car body, causing secondary deformation due to the release of residual stress during the painting and baking stage, resulting in irreversible door and hood gap deviations during the final assembly stage. This invention, on the other hand, proactively selects the clamping force scheme with the least impact on the final assembly quality, rather than the scheme with the maximum clamping force, under the premise that the welding process quality is qualified. This fundamentally cuts off the chain of unidirectional propagation and locking of residual stress along the process chain. The predicted value of the final assembly quality is obtained by the digital twins of each upstream process in the process of solving the cross-process collaborative process parameter combination. The process parameter combination obtained is substituted into the cross-process error propagation sensitivity matrix for forward calculation and output synchronously with the cross-process collaborative process parameter combination, serving as the benchmark value for calculating the final assembly quality residual vector in step three.

[0030] Step S3: Perform a full inspection of the key quality parameters of the vehicle body that comes off the assembly line in the final assembly process to obtain the final assembly quality measured vector. Subtract the final assembly quality measured vector from the final assembly quality predicted value to obtain the final assembly quality residual vector. Decompose the final assembly quality residual vector in reverse through the cross-process error propagation sensitivity matrix S to obtain the error attribution contribution rate of each upstream process.

[0031] Specifically, the full inspection of key quality parameters of the vehicle body after final assembly involves: measuring all door / hood gap, surface difference, and sealing strip compression points for each vehicle body one by one; calculating the root mean square (RMS) value of each type of measurement point; and organizing the RMS values ​​of door / hood gap, surface difference, and sealing strip compression into a final assembly quality measurement vector. The reason for using the RMS value instead of the arithmetic mean is that the RMS value assigns higher weight to measurement points with larger deviations, more sensitively reflecting local quality anomalies and preventing local deviations from being masked by the overall average value. After obtaining the final assembly quality residual vector by subtracting the actual and predicted final assembly quality values, the specific operation of reverse decomposition based on the weights of each column of the cross-process error propagation sensitivity matrix is ​​as follows: For the residual value of each dimension in the final assembly quality residual vector, the weights of each row in the corresponding column of the cross-process error propagation sensitivity matrix are weighted and allocated. The values ​​of all error parameter dimensions allocated to the same upstream process are summed to obtain the original attribution value of the process to the final assembly quality residual of the current batch. After normalizing the original attribution values ​​of each upstream process, the error attribution contribution rate of each upstream process is obtained. The sum of the error attribution contribution rates of the three upstream processes is one, which represents the relative contribution ratio of each upstream process to the final assembly quality deviation of the current batch. When the error attribution contribution rate of a certain process exceeds 0.4, the process is determined to be the main source process of the final assembly quality deviation of the current batch.

[0032] Step S4: Store the batch-to-batch variation of the production condition feature vector of each batch and the cross-process error propagation sensitivity matrix S of the corresponding batch into the drift scenario knowledge base; before the start of production of a new batch, the drift scenario knowledge base retrieves the historical batch that is closest to the production condition feature vector of the current batch, and uses the batch-to-batch variation of the historical batch to perform feedforward correction on the current cross-process error propagation sensitivity matrix S to obtain the feedforward correction sensitivity matrix. The digital twins of each upstream process are introduced into the control target with the feedforward correction sensitivity matrix to solve for the cross-process collaborative process parameter combination of the current batch.

[0033] Specifically, the drift scenario knowledge base is constructed as follows: After each batch of production is completed, the batch number of the material supplier, the cumulative number of die punches, the workshop ambient temperature, humidity, and production cycle time are recorded and organized into a production condition feature vector for that batch. Simultaneously, the element-wise difference between the cross-process error propagation sensitivity matrix of that batch and the cross-process error propagation sensitivity matrix of the previous batch is calculated to obtain the batch-to-batch variation. The production condition feature vector and the batch-to-batch variation are stored in the drift scenario knowledge base as key-value pairs, using the production condition feature vector as the retrieval key and the batch-to-batch variation as the retrieval value, continuously expanding the knowledge base as batches accumulate. The material supplier batch number in the production condition feature vector is processed using one-hot encoding. Other numerical features are normalized according to their respective historical mean and standard deviation and then concatenated with the one-hot encoding to ensure that different types of features have comparable dimensions in Euclidean distance calculation. Before a new batch of production begins, the feature vector of the current batch's production conditions is obtained. The Euclidean distance between the feature vector and the feature vectors of all historical batches' production conditions in the knowledge base is calculated. The five historical batches with the closest Euclidean distance are taken as the search results. The inter-batch changes corresponding to the five historical batches are weighted and summed using the reciprocal of the Euclidean distance between each historical batch and the feature vector of the current batch's production conditions as the weight. The predicted drift is obtained by summing the changes between five historical batches instead of a single most recent batch. The reason for choosing five historical batches instead of a single most recent batch is that the inter-batch changes of a single historical batch may be distorted due to occasional disturbances. The weighted summation of multiple batches can smooth out occasional noise and make the predicted drift more stably reflect the systematic change trend of the cross-process error propagation law under the current production conditions. After adding the predicted drift amount element-by-element to the current cross-process error propagation sensitivity matrix, the absolute value of each column of the sum is normalized to obtain the feedforward correction sensitivity matrix. All the above calculations are completed before the upstream process of the current batch starts production. After the digital twins of each upstream process replace the cross-process error propagation sensitivity matrix with the feedforward correction sensitivity matrix and introduce it into the control target, the particle swarm optimization algorithm is used to solve it again under the constraint of the quality qualification domain of each upstream process to obtain the cross-process collaborative process parameter combination of the current batch and issue it for execution, so that the control intervention node is moved forward to before the final assembly quality deviation of the current batch occurs.

[0034] In one specific embodiment, step S1 includes:

[0035] The sensor data collected from the four major processes of stamping, welding, painting and final assembly are aligned one by one with the workpiece number as the index and connected in series to form a complete process dataset.

[0036] The error parameters of the three upstream processes of stamping, welding and painting in the whole process data are organized into an upstream error parameter matrix, and the quality parameters of the final assembly process are organized into a final assembly quality parameter matrix. The upstream error parameter matrix and the final assembly quality parameter matrix are standardized respectively to obtain the standardized upstream error parameter matrix and the standardized final assembly quality parameter matrix.

[0037] Based on the standardized upstream error parameter matrix and the standardized final assembly quality parameter matrix, partial least squares regression is used to solve the regression coefficient matrix.

[0038] Normalize the absolute value of each column element of the regression coefficient matrix to obtain the cross-process error propagation sensitivity matrix S. The sum of each column element of the cross-process error propagation sensitivity matrix S is 1.

[0039] Specifically, the full-element production data for the four major processes includes: stamping pressure, die clearance, sheet metal thickness deviation, and springback in the stamping process; weld point position deviation, fixture clamping force, welding current, and body positioning reference offset in the welding process; baking temperature, body surface stress distribution, and coating thickness uniformity in the painting process; and door / hood gap values, surface difference values, and sealing strip compression in the final assembly process. The reason for aligning each part by workpiece number is that the same workpiece flows sequentially through the four processes, and only data aligned under the same workpiece number has a true cross-process causal relationship. Data splicing between different workpiece numbers does not reflect the true error propagation relationship. Standardization uses zero-mean unit variance normalization, which involves subtracting the mean of each column from the upstream error parameter matrix and the final assembly quality parameter matrix, then dividing by the standard deviation of that column. This eliminates the interference of different parameter dimensions on the regression coefficients, ensuring that each parameter has a comparable numerical scale in the regression calculation.

[0040] The reason for using partial least squares regression instead of ordinary least squares regression is that there is significant multicollinearity between stamping springback and sheet metal thickness deviation, and between welding clamping force and positioning datum offset. Ordinary least squares regression suffers from unstable regression coefficients and ambiguous physical meanings under multicollinearity conditions. Partial least squares regression extracts latent components from both the standardized upstream error parameter matrix and the standardized final assembly quality parameter matrix, maximizing the covariance between the two sets of variables while eliminating the effects of multicollinearity. This ensures that each element in the obtained regression coefficient matrix has a stable and interpretable physical meaning. After normalizing each column of the regression coefficient matrix by absolute value, the cross-process error propagation sensitivity matrix is ​​obtained, with the sum of each column's elements being one. The physical meaning is that for the same final assembly quality indicator, the sum of the relative contributions of all upstream process error parameters is 100%. This ensures that the error parameters of different dimensions in the stamping, welding, and painting processes have a directly comparable weight relationship under a unified scale. This provides a quantifiable mathematical basis for constructing cross-process control targets based on this weight relationship in subsequent steps using digital twins of each upstream process.

[0041] In one specific embodiment, step S2 introduces the cross-process error propagation sensitivity matrix S into the control objectives of the digital twins of each upstream process, and solves for the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value, including:

[0042] The optimization objective is to minimize the overall impact of the error parameters of each upstream process on the final assembly quality parameters after being weighted by the cross-process error propagation sensitivity matrix S.

[0043] Under the constraint of the quality qualification domain of each upstream process, the particle swarm optimization algorithm is used to solve the optimization objective, and the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value are obtained.

[0044] Specifically, the logic for introducing the cross-process error propagation sensitivity matrix into the control objectives of the digital twins of upstream processes is as follows: For each of the three upstream processes—stamping, welding, and painting—the current controllable combination of process parameters is used as the decision variable. The simulation model of the digital twin of that process calculates the predicted values ​​of each dimension of error parameters corresponding to the process parameter combination. The predicted values ​​of each dimension of error parameters are multiplied by the weight value of the corresponding row in the cross-process error propagation sensitivity matrix and then summed column by column to obtain the weighted comprehensive influence of the process error parameters on the three quality indicators of the final assembly door / cover gap value, surface difference value, and sealing strip compression amount. The optimization objective is to minimize the weighted sum of this weighted comprehensive influence. The weight of the door / cover gap value is set to 0.4, the weight of the surface difference value is set to 0.4, and the weight of the sealing strip compression amount is set to 0.2. The sum of the weights of the three quality indicators is 1. The weight values ​​are pre-set by the product engineer based on the degree of influence of each indicator on the user's perceived quality in the vehicle quality design specifications. The quality acceptance domain constraint refers to the allowable range of process parameters and the acceptance judgment boundary of quality indicators within each process. Taking the welding process as an example, the quality acceptance domain constraint is that the deviation of the weld point position does not exceed 0.5 mm and the torsional stiffness of the car body is not less than 18,000 Nm per degree. The solution is performed within this constraint range to ensure that the upstream process pursues the minimum error of the entire process chain while the quality of this process is not lower than the acceptance standard.

[0045] The execution logic of the particle swarm optimization algorithm in this step is as follows: using the controllable process parameter combination of each upstream process as the particle position, the particle swarm is randomly initialized within the quality qualified domain constraint range, with the number of particles set to 50, the number of iterations set to 200, and the convergence accuracy set to 0.0001. In each iteration, each particle updates its own velocity and position based on its individual historical best position and the global best position of the population. The fitness function is used to evaluate the quality of each particle, with the minimum weighted comprehensive influence. After iteration to convergence, the position of the particle with the best fitness is output as the cross-process collaborative process parameter combination. The calculation method for the final assembly quality prediction value is as follows: Substitute the cross-process collaborative process parameter combination obtained from the solution into the simulation model of the digital twin of each upstream process to obtain the error parameter prediction value corresponding to each process. Multiply the error parameter prediction value of each dimension by the row weight corresponding to the cross-process error propagation sensitivity matrix and sum them by column to obtain the prediction values ​​of three quality indicators: the gap value of the final assembly door cover, the surface difference value, and the compression amount of the sealing strip. Organize them into the final assembly quality prediction value, and output them synchronously with the cross-process collaborative process parameter combination as the benchmark value for calculating the final assembly quality residual vector in step S3.

[0046] In one specific embodiment, step S3 involves a full inspection of the key quality parameters of the vehicle body coming off the assembly line, yielding a measured assembly quality vector. The measured assembly quality vector is then subtracted from the predicted assembly quality value to obtain a residual assembly quality vector. This residual vector is then decomposed using the cross-process error propagation sensitivity matrix S to obtain the error attribution contribution rate of each upstream process, including:

[0047] A full inspection was conducted on the door and hood gap values, surface difference values, and sealing strip compression values ​​of the vehicle body after the final assembly process. The root mean square values ​​of the door and hood gap values, surface difference values, and sealing strip compression values ​​were organized into a final assembly quality measurement vector.

[0048] The difference between the measured final assembly quality vector and the predicted final assembly quality vector is used to obtain the final assembly quality residual vector.

[0049] Based on the weights of each column of the cross-process error propagation sensitivity matrix S, the final assembly quality residual vector is decomposed into the error parameter dimensions of each upstream process to obtain the error attribution contribution rate of each upstream process.

[0050] Specifically, the full inspection of the vehicle body coming off the assembly line involves measuring each of the 24 door / hood gap measurement points, 24 surface difference measurement points, and 12 sealing strip compression measurement points for each vehicle body. The root mean square (RMS) values ​​of the 24 measured door / hood gap values, 24 measured surface difference values, and 12 measured sealing strip compression values ​​are calculated. These three RMS values ​​are then organized into a final assembly quality measurement vector. The RMS value is used instead of the arithmetic mean because it assigns higher weight to measurement points with larger deviations, more sensitively reflecting local quality anomalies and preventing local deviations from being masked by the overall average. The final assembly quality measurement vector is then subtracted element-wise from the final assembly quality prediction value output in step S2 to obtain a final assembly quality residual vector containing three elements: door / hood gap residual, surface difference residual, and sealing strip compression residual.

[0051] The specific calculation steps for inverse decomposing the final assembly quality residual vector based on the weights of each column of the cross-process error propagation sensitivity matrix are as follows: Multiply the residual value of the j-th element in the final assembly quality residual vector by the weight value of the i-th row in the j-th column of the cross-process error propagation sensitivity matrix to obtain the attribution allocation value of the j-th dimension final assembly quality residual to the i-th dimension upstream process error parameter. Summate the attribution allocation values ​​of the three final assembly quality residual dimensions according to the upstream process error parameter dimensions to obtain the original attribution value of the i-th dimension upstream process error parameter. Summate the original attribution values ​​of all error parameter dimensions belonging to the same upstream process to obtain the original attribution values ​​of the three upstream processes: stamping, welding, and painting. Normalize the original attribution values ​​of the three upstream processes, i.e., divide the original attribution value of each process by the sum of the original attribution values ​​of the three processes to obtain the error attribution contribution rate of each upstream process. The sum of the error attribution contribution rates of the three upstream processes is 1. When the error attribution contribution rate of a certain upstream process exceeds 0.4, the process is determined to be the main source of the batch final assembly quality deviation. The threshold of 0.4 is set on the basis that if the three upstream processes contribute equally, each accounts for about 0.33. If it exceeds 0.4, it indicates that the contribution rate of the process is significantly higher than the average level and has statistical dominance.

[0052] In one specific embodiment, step S4 involves storing the production condition feature vector of each batch and the batch-to-batch variation of the cross-process error propagation sensitivity matrix S of the corresponding batch into a drift scenario knowledge base, including:

[0053] The production condition feature vector includes material supplier batch number, cumulative number of die punches, workshop ambient temperature, ambient humidity, and production cycle time;

[0054] The production condition feature vectors and the corresponding batch-to-batch variations are stored in the drift scenario knowledge base as key-value pairs.

[0055] Specifically, the material supplier batch number reflects the impact of raw material composition fluctuations on the cross-process error propagation pattern. Differences in yield strength and elongation of different batches of raw materials directly alter the propagation ratio of stamping springback and welding residual stress. The cumulative number of die stampings reflects the impact of die wear on the distribution of stamping error parameters. As the cumulative number of die stampings increases, the systematic shift pattern of stamping springback changes. The workshop ambient temperature and humidity reflect the impact of the thermal field distribution during the painting and baking process on the stress release pattern of the vehicle body. The production cycle reflects the impact of the production line load state on the response characteristics of equipment in each process. The material supplier batch number is a discrete variable, processed using one-hot encoding. It is converted into a binary vector of the same length as the number of suppliers and then concatenated with four types of numerical features: cumulative number of die stampings, workshop ambient temperature, ambient humidity, and production cycle. The numerical features are normalized according to their respective historical means and standard deviations before concatenation to ensure that different types of features have comparable numerical scales in subsequent Euclidean distance calculations.

[0056] After each batch of production is completed, the cross-process error propagation sensitivity matrix of the current batch, updated in step S4 using the recursive least squares method with a forgetting factor, is subtracted element-wise from the cross-process error propagation sensitivity matrix of the previous batch to obtain the inter-batch variation of the current batch. The inter-batch variation has the same dimension as the cross-process error propagation sensitivity matrix, and each element represents the magnitude and direction of the change in sensitivity weight at the corresponding position between adjacent batches. The feature vector of the production conditions of the current batch is used as the retrieval key, and the inter-batch variation of the current batch is used as the retrieval value. These are stored in the drift scenario knowledge base in key-value pairs. The knowledge base is continuously expanded as batches accumulate. Each record in the knowledge base corresponds to the mapping relationship between the production conditions of a historical batch and the cross-process error propagation variation of that batch, providing a data foundation for retrieving historical drift patterns based on the similarity of production conditions before the start of subsequent new batch production.

[0057] In one specific embodiment, in step S4, before the start of production of a new batch, the historical batch with the most similar feature vector to the current batch's production conditions is retrieved from the drift scenario knowledge base. The current cross-process error propagation sensitivity matrix S is then feedforward corrected using the inter-batch variation corresponding to the historical batch, resulting in a feedforward corrected sensitivity matrix, including:

[0058] Calculate the Euclidean distance between the feature vector of the current batch production conditions and the feature vectors of the production conditions of each historical batch in the drift scenario knowledge base, and retrieve the k historical batches with the closest Euclidean distance;

[0059] Using the reciprocal of the Euclidean distance between each historical batch and the feature vector of the production conditions of the current batch as the weight, the inter-batch changes corresponding to each historical batch are weighted and summed to obtain the predicted drift.

[0060] Add the predicted drift amount to the current cross-process error propagation sensitivity matrix S, and normalize the absolute value of each column element of the added matrix to obtain the feedforward correction sensitivity matrix.

[0061] Specifically, the Euclidean distance between the feature vector of the current batch production conditions and the feature vector of each historical batch in the drift scenario knowledge base is obtained by summing the squares of the differences element by element and taking the square root. The Euclidean distances of all historical batch records in the knowledge base are sorted in ascending order, and the five historical batches with the smallest Euclidean distances are selected as the search results. The reason for setting k to 5 is as follows: when selecting a single most recent historical batch, the batch-to-batch variation of that batch may be distorted due to occasional disturbances, causing the predicted drift to deviate from the actual drift pattern; when selecting too many historical batches, historical batches with large differences in production conditions will introduce drift information that is unrelated to the current production conditions, interfering with the accuracy of the predicted drift; selecting five historical batches can achieve a balance between smoothing out occasional noise and maintaining the correlation of production conditions. The specific calculation logic for using the reciprocal of the Euclidean distance between each historical batch and the feature vector of the production conditions of the current batch as the weight is as follows: take the reciprocal of the Euclidean distance corresponding to each of the five historical batches, divide each reciprocal value by the sum of the five reciprocal values ​​to obtain the normalized weight, multiply the normalized weight of each historical batch by the corresponding inter-batch change amount and sum them to obtain the predicted drift amount. The predicted drift amount has the same dimension as the cross-process error propagation sensitivity matrix, and each element of it represents the predicted change magnitude and direction of the cross-process error propagation sensitivity matrix of the current batch at the corresponding position.

[0062] After element-wise addition of the predicted drift to the current cross-process error propagation sensitivity matrix, the absolute value of each column of the sum is divided by the sum of the absolute values ​​of that column to obtain the feedforward correction sensitivity matrix. Normalization ensures that the sum of each column of the feedforward correction sensitivity matrix is ​​1, maintaining its physical meaning as a relative sensitivity weight consistent with the cross-process error propagation sensitivity matrix. This allows the digital twins of upstream processes to directly replace the cross-process error propagation sensitivity matrix with the feedforward correction sensitivity matrix when introducing it into the control target. All the above Euclidean distance calculations, weighted summations, and normalization operations are completed before the start of production in the three upstream processes of stamping, welding, and painting for that batch. The digital twins of each upstream process re-solve the problem using a particle swarm optimization algorithm under the constraint of the quality qualification domain, based on the feedforward correction sensitivity matrix, to obtain the cross-process collaborative process parameter combination for that batch, which is then issued to the corresponding physical processes for execution. This moves the control intervention node forward to before the final assembly quality deviation occurs in that batch, unlike the feedback control paradigm in existing technologies that triggers correction only after quality deviations are detected during final assembly inspection.

[0063] In one specific embodiment, in step S4, the digital twins of each upstream process introduce the control objective with a feedforward correction sensitivity matrix, and solve for the cross-process collaborative process parameter combination for the current batch, including:

[0064] Based on the differences between each element in the feedforward correction sensitivity matrix and the corresponding element in the cross-process error propagation sensitivity matrix S, determine the direction and magnitude of the change in the influence weight of each upstream process error parameter on the final assembly quality parameter in the current batch.

[0065] The digital twins of each upstream process replace the cross-process error propagation sensitivity matrix S with the feedforward correction sensitivity matrix to introduce the control objective. Under the constraint of the quality qualification domain of each upstream process, the particle swarm optimization algorithm is used to solve the control objective again to obtain the cross-process collaborative process parameter combination for the current batch.

[0066] Specifically, the operational logic for determining the direction and magnitude of change based on the differences between each element in the feedforward correction sensitivity matrix and the corresponding element in the cross-process error propagation sensitivity matrix is ​​as follows: The difference between each element of the feedforward correction sensitivity matrix and the cross-process error propagation sensitivity matrix is ​​calculated. A positive difference indicates that the weight of the upstream process error parameter on the corresponding final assembly quality indicator under the current batch production conditions has increased compared to the historical benchmark. Therefore, the digital twins of each upstream process need to apply stricter suppression to this dimension of the error parameter when solving for the control objectives. A negative difference indicates that the influence weight has decreased, and the control over this dimension of the error parameter can be appropriately relaxed. The absolute value of the difference represents the magnitude of the weight change. A larger magnitude indicates that the change in batch production conditions has a more significant impact on the cross-process propagation law of this dimension of the error parameter, and the larger the adjustment amount of the corresponding weight in the control objectives. This difference analysis provides an interpretable physical basis for the solution results of the digital twins of each upstream process after introducing the feedforward correction sensitivity matrix, enabling those skilled in the art to understand the reasons why the combination of cross-process collaborative process parameters changes compared to the previous batch.

[0067] After the digital twins of each upstream process replace the cross-process error propagation sensitivity matrix with the feedforward correction sensitivity matrix to introduce the control objective, the optimization logic of the control objective is completely consistent with step S2. That is, the optimization objective is to minimize the comprehensive impact of the error parameters of each upstream process on the final assembly quality parameters after being weighted by the feedforward correction sensitivity matrix. Under the constraint of the quality qualified domain of each upstream process, the particle swarm optimization algorithm is used to solve the problem. The number of particles is set to 50, the number of iterations is set to 200, and the convergence accuracy is set to 0.0001, which is consistent with the algorithm execution parameters in step S2. After iteration to convergence, the position of the particle with the best fitness is output as the cross-process collaborative process parameter combination for the current batch. This parameter combination is sent to the physical production lines corresponding to the three upstream processes of stamping, welding, and painting for execution, thus completing the feedforward control closed loop for the current batch before the final assembly quality deviation occurs.

[0068] The above describes the intelligent control method for automobile manufacturing system integration integrating digital twins in the embodiments of this application. The following describes the intelligent control system for automobile manufacturing system integration integrating digital twins in the embodiments of this application. Please refer to [link / reference]. Figure 2One embodiment of the intelligent control system for automobile manufacturing systems integrating digital twins in this application includes:

[0069] The concatenation module is used to concatenate the full-element production data of the four major processes into a full-process dataset using the workpiece number as an index, and to construct a cross-process error propagation sensitivity matrix S based on the full-process dataset;

[0070] The solution module is used to introduce the cross-process error propagation sensitivity matrix S into the control target of the digital twin of each upstream process, solve for the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value, and send the cross-process collaborative process parameter combination to the corresponding physical process for execution.

[0071] The decomposition module is used to perform full inspection of the key quality parameters of the vehicle body after the final assembly process to obtain the final assembly quality measured vector. The final assembly quality residual vector is obtained by subtracting the final assembly quality measured vector from the final assembly quality predicted value. The final assembly quality residual vector is decomposed in reverse by the cross-process error propagation sensitivity matrix S to obtain the error attribution contribution rate of each upstream process.

[0072] The correction module is used to store the production condition feature vector of each batch and the batch-to-batch change of the cross-process error propagation sensitivity matrix S of the corresponding batch into the drift scenario knowledge base. Before the start of production of a new batch, the drift scenario knowledge base retrieves the historical batch that is closest to the production condition feature vector of the current batch, and uses the batch-to-batch change of the historical batch to perform feedforward correction on the current cross-process error propagation sensitivity matrix S to obtain the feedforward correction sensitivity matrix. The digital twins of each upstream process are introduced into the control target with the feedforward correction sensitivity matrix to solve for the cross-process collaborative process parameter combination of the current batch.

[0073] above Figure 2 The intelligent control system for automobile manufacturing system integration with digital twins in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The intelligent control device for automobile manufacturing system integration with digital twins in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0074] Reference Figure 3 This invention also provides an intelligent control device for an automotive manufacturing system integrating digital twins. This intelligent control device for an automotive manufacturing system integrating digital twins can be a server, and its internal structure can be as follows: Figure 3As shown, the integrated intelligent control device for automotive manufacturing systems, which integrates digital twins, includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor, designed as a computer, provides computing and control capabilities. The memory of the integrated intelligent control device for automotive manufacturing systems includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the integrated intelligent control device for automotive manufacturing systems is used to store the data corresponding to this embodiment. The network interface of the integrated intelligent control device for automotive manufacturing systems is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0075] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the intelligent control equipment for the integrated digital twin automotive manufacturing system applied thereto.

[0076] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the integrated intelligent control method for automobile manufacturing systems incorporating digital twins.

[0077] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0078] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an integrated intelligent control device (which can be a personal computer, server, or network device, etc.) of an integrated digital twin automotive manufacturing system to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0079] 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for integrated intelligent control of an automobile manufacturing system that integrates digital twins, characterized in that, The method includes: Step S1: Using the workpiece number as an index, connect all the production data of the four major processes into a full-process dataset, and construct a cross-process error propagation sensitivity matrix S based on the full-process dataset; Step S2: Introduce the cross-process error propagation sensitivity matrix S into the control target of the digital twin of each upstream process, solve for the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value, and send the cross-process collaborative process parameter combination to the corresponding physical process for execution; Step S3: Perform a full inspection of the key quality parameters of the vehicle body after the final assembly process to obtain the final assembly quality measured vector. Subtract the final assembly quality measured vector from the final assembly quality predicted value to obtain the final assembly quality residual vector. Decompose the final assembly quality residual vector in reverse through the cross-process error propagation sensitivity matrix S to obtain the error attribution contribution rate of each upstream process. Step S4: Store the production condition feature vector of each batch and the batch-to-batch variation of the cross-process error propagation sensitivity matrix S of the corresponding batch into the drift scenario knowledge base; before the start of production of a new batch, the drift scenario knowledge base retrieves the historical batch that is closest to the production condition feature vector of the current batch, and uses the batch-to-batch variation of the historical batch to perform feedforward correction on the current cross-process error propagation sensitivity matrix S to obtain the feedforward correction sensitivity matrix. The digital twins of each upstream process are introduced into the control target with the feedforward correction sensitivity matrix to solve for the cross-process collaborative process parameter combination of the current batch.

2. The intelligent control method for automobile manufacturing system integration based on digital twins as described in claim 1, characterized in that, Step S1 includes: The sensor data collected from the four major processes of stamping, welding, painting and final assembly are aligned one by one with the workpiece number as the index and connected in series to form the full process dataset. The error parameters of the three upstream processes of stamping, welding and painting in the whole process data are organized into an upstream error parameter matrix, and the quality parameters of the final assembly process are organized into a final assembly quality parameter matrix. The upstream error parameter matrix and the final assembly quality parameter matrix are standardized respectively to obtain a standardized upstream error parameter matrix and a standardized final assembly quality parameter matrix. Based on the standardized upstream error parameter matrix and the standardized final assembly quality parameter matrix, partial least squares regression is used to solve the regression coefficient matrix. The elements of each column of the regression coefficient matrix are normalized by absolute value to obtain the cross-process error propagation sensitivity matrix S, and the sum of the elements of each column of the cross-process error propagation sensitivity matrix S is 1.

3. The intelligent control method for automobile manufacturing system integration based on digital twins as described in claim 1, characterized in that, In step S2, the cross-process error propagation sensitivity matrix S is introduced into the control objectives of the digital twins of each upstream process, and the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value are obtained by solving, including: The optimization objective is to minimize the overall impact of the error parameters of each upstream process on the final assembly quality parameters after being weighted by the cross-process error propagation sensitivity matrix S. Under the constraint of the quality qualification domain of each upstream process, the particle swarm optimization algorithm is used to solve the optimization objective to obtain the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value.

4. The intelligent control method for automobile manufacturing system integration based on digital twins as described in claim 1, characterized in that, In step S3, a full inspection of the key quality parameters of the vehicle body coming off the assembly line is performed to obtain the measured assembly quality vector. The difference between the measured assembly quality vector and the predicted assembly quality value is used to obtain the assembly quality residual vector. The assembly quality residual vector is then decomposed inversely using the cross-process error propagation sensitivity matrix S to obtain the error attribution contribution rate of each upstream process, including: A full inspection is performed on the door and hood gap value, surface difference value, and sealing strip compression amount of the vehicle body after the final assembly process. The root mean square values ​​of the door and hood gap value, the surface difference value, and the sealing strip compression amount are organized into the final assembly quality measurement vector. The difference between the measured final assembly quality vector and the predicted final assembly quality value is used to obtain the final assembly quality residual vector; Based on the weights of each column of the cross-process error propagation sensitivity matrix S, the final assembly quality residual vector is decomposed into the error parameter dimensions of each upstream process to obtain the error attribution contribution rate of each upstream process.

5. The intelligent control method for automobile manufacturing system integration based on digital twins as described in claim 1, characterized in that, In step S4, the production condition feature vector of each batch and the batch-to-batch variation of the cross-process error propagation sensitivity matrix S of the corresponding batch are stored in the drift scenario knowledge base, including: The production condition feature vector includes the material supplier batch number, the cumulative number of die punches, the workshop ambient temperature, the ambient humidity, and the production cycle time; The production condition feature vector and the batch-to-batch variation of the corresponding batch are stored in the drift scenario knowledge base in the form of key-value pairs.

6. The intelligent control method for automobile manufacturing system integration based on digital twins as described in claim 5, characterized in that, In step S4, before the start of production for a new batch, the drift scenario knowledge base retrieves the historical batch that most closely matches the feature vector of the production conditions for the current batch. The inter-batch variation corresponding to the historical batch is then used to perform a feedforward correction on the current cross-process error propagation sensitivity matrix S, resulting in a feedforward corrected sensitivity matrix, including: Calculate the Euclidean distance between the feature vector of the production conditions of the current batch and the feature vector of the production conditions of each historical batch in the drift scenario knowledge base, and retrieve the k historical batches with the closest Euclidean distance; Using the reciprocal of the Euclidean distance between each historical batch and the feature vector of the production conditions of the current batch as the weight, the batch-to-batch changes corresponding to each historical batch are weighted and summed to obtain the predicted drift. The predicted drift is added to the current cross-process error propagation sensitivity matrix S, and the elements of each column of the added matrix are normalized by absolute value to obtain the feedforward correction sensitivity matrix.

7. The intelligent control method for automobile manufacturing system integration based on digital twins as described in claim 6, characterized in that, In step S4, the digital twins of each upstream process are introduced into the control target using the feedforward correction sensitivity matrix to solve for the cross-process collaborative process parameter combination for the current batch, including: Based on the difference between each element in the feedforward correction sensitivity matrix and the corresponding element in the cross-process error propagation sensitivity matrix S, determine the direction and magnitude of the change in the influence weight of each upstream process error parameter on the final assembly quality parameter in the current batch. The digital twins of each upstream process replace the cross-process error propagation sensitivity matrix S with the feedforward correction sensitivity matrix and introduce it into the control target. Under the constraint of the quality qualification domain of each upstream process, the particle swarm optimization algorithm is used to solve the control target again to obtain the cross-process collaborative process parameter combination for the current batch.

8. An integrated intelligent control system for automobile manufacturing systems that incorporates digital twins, characterized in that, The method for integrating digital twins in automotive manufacturing system integration and intelligent control as described in any one of claims 1-7, wherein the integrated digital twin automotive manufacturing system integration and intelligent control system comprises: The concatenation module is used to concatenate the full-element production data of the four major processes into a full-process dataset using the workpiece number as an index, and to construct a cross-process error propagation sensitivity matrix S based on the full-process dataset; The solution module is used to introduce the cross-process error propagation sensitivity matrix S into the control target of the digital twin of each upstream process, solve for the cross-process collaborative process parameter combination and the corresponding final assembly quality prediction value, and send the cross-process collaborative process parameter combination to the corresponding physical process for execution. The decomposition module is used to perform full inspection of the key quality parameters of the vehicle body after the final assembly process to obtain the final assembly quality measured vector. The final assembly quality residual vector is obtained by subtracting the final assembly quality measured vector from the final assembly quality predicted value. The final assembly quality residual vector is decomposed in reverse by the cross-process error propagation sensitivity matrix S to obtain the error attribution contribution rate of each upstream process. The correction module is used to store the production condition feature vector of each batch and the batch-to-batch change of the cross-process error propagation sensitivity matrix S of the corresponding batch into the drift scenario knowledge base. Before the start of production of a new batch, the drift scenario knowledge base retrieves the historical batch that is closest to the production condition feature vector of the current batch, and uses the batch-to-batch change of the historical batch to perform feedforward correction on the current cross-process error propagation sensitivity matrix S to obtain the feedforward correction sensitivity matrix. The digital twins of each upstream process are introduced into the control target with the feedforward correction sensitivity matrix to solve for the cross-process collaborative process parameter combination of the current batch.

9. An integrated intelligent control device for automobile manufacturing systems that incorporates digital twins, characterized in that, It includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the intelligent control method for automobile manufacturing system integration based on any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the intelligent control method for automotive manufacturing system integration that incorporates digital twins as described in any one of claims 1 to 7.

Images