Stamping prediction method based on multi-machine learning model and automatic compensation device

Abstract

The invention relates to the technical field of computer science, in particular to a stamping prediction method based on a multi-machine learning model and an automatic compensation device.The stamping prediction method comprises the steps that firstly, a sheet microstructure and material characteristics are obtained through a transmission electron microscope and an atomic probe tomography technology, and then finite element simulation sampling is conducted; a data set is expanded by using an adaptive algorithm, multiple models are constructed and trained to obtain a final springback prediction model, an automatic compensation device is embedded to realize automatic compensation, and the automatic compensation device covers visual parameter input, embedded springback prediction, springback compensation calculation and an automatic control module. Parameters can be visually input, springback can be accurately predicted, a compensation value can be calculated, and the process can be adjusted. The method aims at solving the problems that in the high-strength plate stamping process, due to springback, the size and shape of a part deviate from the design requirement, and a traditional prediction and compensation technology is insufficient in precision and efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of computer science and technology, and relates to a stamping prediction method and an automatic compensation device based on multiple machine learning models. Background Art

[0002] Sheet metal stamping is a metal plastic forming method widely used in the industrial field. By applying pressure to the metal sheet with a mold and stamping equipment, plastic deformation is generated to obtain stamping parts with specific shapes, dimensions, and properties. This technology plays a key role in industries such as aerospace, automotive manufacturing, electrical machinery and appliances, and daily necessities production, and is an important means to achieve large-scale and high-efficiency production of various metal parts. With the development of material science, high-strength sheet materials are increasingly widely used in the stamping field due to their excellent properties. However, during the stamping process of high-strength sheet materials, the springback phenomenon has become a key factor restricting product quality and production efficiency. Springback refers to the elastic recovery deformation generated due to the release of internal stress when the external force is removed after the metal sheet undergoes plastic deformation during stamping. This deformation will cause the dimensions and shapes of the stamping parts to deviate from the design requirements, seriously affecting the accuracy and quality of the products.

[0003] In the stamping production practice of high-strength sheet materials, the springback problem has brought many challenges. From the perspective of product quality, springback makes it difficult to ensure the dimensional accuracy of stamping parts, increasing the costs of subsequent processing and debugging. For example, in automotive manufacturing, the springback of body panels will lead to a decrease in the assembly accuracy between parts, affecting the appearance and performance of the entire vehicle. In the aerospace field, where extremely high precision requirements are imposed on parts, springback may cause parts to fail to meet the design standards, even affecting flight safety. In terms of production efficiency, in order to compensate for the errors caused by springback, it is often necessary to conduct multiple trial moldings and adjust the molds, which not only prolongs the product development cycle but also increases the production cost. Moreover, the springback amount of high-strength sheet materials is relatively large, and it is difficult for traditional stamping processes to effectively control, further exacerbating the severity of these problems. For different batches of high-strength sheet materials, due to the slight differences in material properties, the springback performances are also different, making it more difficult to control the quality during the production process.

[0004] Currently, for the stamping springback problem of high-strength sheet materials, there are mainly the following solutions: Relying on the process experience of technicians, compensating for springback by manually adjusting mold designs such as changing the bending radius, mold opening, etc. or stamping parameters such as pressure, speed, etc. For example, in some small stamping processing factories, technicians rely on their long-term accumulated experience to roughly estimate and adjust the springback of simple stamping parts. However, this method has obvious limitations. The experience differences among different technicians are relatively large, making it difficult to ensure the accuracy and consistency of compensation, and for complex-shaped parts and new high-strength sheet materials, the empirical method often fails to work.

[0005] Springback is predicted by building a physical model, such as building a simple mathematical model based on the principles of material mechanics and plasticity to describe the deformation and springback behavior of sheet metal during stamping. However, since the stamping process involves many complex factors, such as material anisotropy, work hardening, and friction between the die and the sheet metal, physical models are often difficult to accurately simulate the actual situation, resulting in limited prediction accuracy.

[0006] Finite element analysis (FEA) technology is used to construct finite element models of sheet metal, mold and stamping process, combined with material constitutive model, friction model and boundary conditions, and CAE software is used to predict the springback amount and perform springback compensation calculations. Patent CN202410211644.9, the patent name is a method and device for springback compensation of stamping parts based on machine learning model. The principle is to obtain the springback data of stamping parts through finite element simulation, and then reversely compensate the mold surface according to the springback amount. This method improves the accuracy of springback prediction to a certain extent, but it has high requirements on the parameter setting and calculation accuracy of the model, and the calculation process is complicated and time-consuming. Different CAE software and different settings and usage levels of the software will affect the reliability of the prediction results.

[0007] For example, patent number CN202411909498.3, a method and device for rapid springback prediction of complex components based on CNN-LSTM, performs convolution operations on three-dimensional model images through the CNN model, which can effectively capture the macro-structural characteristics of the components, such as the overall size, rib distribution, and curvature radius. For example, in the case of high-ribbed wall panels, CNN can automatically identify the correlation between key geometric parameters such as rib height and thickness and springback. By using pixel matrix conversion and pooling layer dimensionality reduction, the data preprocessing time is greatly reduced compared with traditional finite element meshing, which is suitable for rapid generation of training data sets. The LSTM module can handle the serial dependence of process parameters (such as aging temperature and time), such as analyzing the stress release law in different aging stages, avoiding the limitation of traditional methods that regard process parameters as independent variables. However, it only relies on the macro-geometric features of the three-dimensional model and completely ignores the decisive influence of the microstructure of the material, such as grain boundary distribution, dislocation density, and element segregation on springback. Although the macro-micro constitutive equation is introduced, it is not combined with the microscopic image data, and the accuracy of the microscopic parameters cannot be verified by experimental data. In addition, the "image-value" data is only processed through the CNN-LSTM series structure, and different feature types, such as the differentiated weights of the spatial features of the image and the numerical features of the process parameters, are not considered, which may lead to the weakening of key information. An automatic compensation scheme linked with the stamping equipment is not proposed. In actual applications, it is still necessary to manually adjust the mold parameters based on the prediction results, which cannot avoid the problems of human errors and extended mold trial cycles.

[0008] In summary, traditional solutions to the problem of springback in high-strength sheet metal stamping generally have the disadvantages of heavy reliance on process experience, low precision, and complex calculations. In order to effectively solve the problems in actual production, a stamping prediction method and automatic compensation device based on multiple machine learning models are adopted. This method uses machine learning algorithms to process and analyze stamping data, which can mine the potential laws in the data, more accurately predict springback, and realize automatic compensation. It is expected to overcome the shortcomings of traditional methods and improve the quality and efficiency of stamping production. Summary of the invention

[0009] The present invention provides a stamping prediction method and automatic compensation device based on a multi-machine learning model to solve the problem that the size and shape of parts deviate from the design requirements due to springback during the stamping process of high-strength sheet metal, and the traditional prediction and compensation technologies have insufficient accuracy and efficiency.

[0010] In order to solve the above problems, the technical solution adopted by the invention is: A stamping prediction method based on multiple machine learning models, the method comprising: S01 Microstructure and feature acquisition: Use transmission electron microscopy combined with atomic probe tomography to obtain the microstructure of the sheet material, measure the microstructural characteristics of element distribution, grain boundary characteristics and dislocation density at the atomic scale, and obtain the characteristics of the material's elastic modulus and hardening index; S02 Initial simulation sampling: Use the finite element method to sample stamping parts, use ABAQUS combined with the improved UMAT subroutine to perform stamping springback simulation, and record the springback vector data; S03 Dataset Expansion: For a variety of stamping materials, the adaptive parameter adjustment algorithm is used to traverse the initial process parameters, the parameter step size is adjusted according to the material properties and the previous simulation results, and the stamping springback data set is obtained through simulation sampling; S04 Model construction and training: The improved ResNet-50 model is used to integrate the attention mechanism to extract features from the microstructure map, and the extracted features are integrated with the material characteristic parameters and process parameters. The data set is divided into training set, validation set and test set. The five models of support vector regression, kernel ridge regression, KNN, random forest regression and gradient boosting tree are trained using transfer learning and ensemble learning strategies. The root mean square error (RMSE), mean absolute percentage error (MAPE) and determination coefficient (C As the evaluation index, the dynamic weight allocation algorithm is used to integrate the final rebound prediction model according to the evaluation results, where: It is used to measure the degree of fit of the model to the data. The closer it is to 1, the better the model fit is. , the formula is

[0011] in is the true value, is the predicted value, is the mean of the true values, n is the number of samples; MSE (Mean Squared Error), which measures the average of the squared differences between the predicted values and the true values. The formula is

[0012] where is the true value, is the predicted value, n is the number of samples; MAPE (Mean Absolute Percentage Error), which measures the relative error between the predicted values and the true values, expressed as a percentage. The formula is

[0013] According to the ratio of each model determine the weight of each model in the final prediction, integrate the models into a compensation prediction model, input the material properties and process parameters, and the model outputs the springback prediction value; S05 Automatic Compensation: Embed the final springback prediction model into an automatic compensation device. The visual parameter input module of this device receives the process parameters and material property parameters input by the operator; the embedded springback prediction model predicts the springback vector; the springback compensation calculation module calculates the springback compensation value according to the prediction result and the preset compensation rule; the automatic compensation control module generates a control signal according to the compensation value to adjust the process parameters of the stamping equipment for automatic compensation of the springback of the stamping parts.

[0014] An automatic compensation device for high-strength sheet metal stamping parts based on multiple machine learning models, comprising the following modules: A visual parameter input module, used to visually set the process parameters and material property parameters for the automatic springback compensation device and input the parameters into the embedded springback prediction module; An embedded springback prediction module, the prediction model of which is generated according to the springback prediction method for high-strength sheet metal stamping parts based on multiple machine learning models and then embedded in this device, used to predict the springback vector value during the stamping process on the premise of providing the process parameters and material property parameters; A springback compensation calculation module, used to calculate the springback compensation vector value; An automatic control module, used to generate a control signal to control and adjust the process.

[0015] The principle and advantages of this solution are as follows: By using transmission electron microscopy and atom probe tomography techniques, the microscopic structural characteristics such as elemental distribution, grain boundary characteristics, and dislocation density of the sheet metal at the atomic scale, as well as material characteristics such as elastic modulus and hardening index, are obtained to reveal the influence of the material's essential properties on stamping performance at the microscopic level. Secondly, with the help of the finite element method, sampling and springback simulation of stamping parts are carried out using ABAQUS combined with an improved UMAT subroutine, and the springback vector data is recorded to obtain the basic simulation data. Then, an adaptive parameter adjustment algorithm is adopted to adjust the parameter step size according to the material characteristics and the previous simulation results, and the process parameters of various stamping materials are traversed and sampled to expand the stamping springback data set to provide rich data for model training. Then, the ResNet-50 model is improved and an attention mechanism is incorporated to extract deep features from the microscopic structure diagrams. After integrating them with the material characteristic parameters and process parameters, transfer learning and ensemble learning strategies are used to train five models such as support vector regression and kernel ridge regression, and through the dynamic weight allocation algorithm, with the root mean square error RMSE, mean absolute percentage error MAPE, and coefficient of determination as evaluation indicators, multiple models are integrated into the final springback prediction model. Finally, the prediction model is embedded in the automatic compensation device to predict the springback according to the parameters input by the operator, and through the springback compensation calculation module and the automatic compensation control module, the automatic adjustment of the process parameters of the stamping equipment is realized to complete the automatic compensation of the springback of the stamping parts.

[0016] Compared with the existing technology, the traditional empirical method and manual adjustment method rely on personal experience, and the accuracy is difficult to guarantee, resulting in large dimensional deviations of parts. However, this solution uses the accurate data obtained from the microscopic structure and characteristics, combined with complex model construction and training, to improve the springback prediction accuracy. For example, in the production of automotive parts, the dimensional deviation of high-strength sheet metal stampings produced by the traditional method may reach ±1 mm, and after adopting this solution, the dimensional deviation can be controlled within ±0.2 mm, significantly improving the product quality and reducing the scrap rate; this method realizes data acquisition by combining the finite element CAE simulation software with the material microscopic structure, and introduces microscopic structure characteristics to significantly improve the prediction performance. After separately training five models and integrating them into a unified model, and embedding it in the automatic compensation device, automatic compensation based on material characteristics and process parameters is realized.

[0017] Although traditional finite element analysis uses numerical simulation, it has poor adaptability to changes in material characteristics and process parameters. The adaptive parameter adjustment algorithm and multi-model integration strategy of this solution enable it to quickly adapt to different stamping materials and process conditions. When producing high-strength sheet metal stampings for different models of automobiles, there is no need for a large amount of re-adjustment, and it can be quickly switched and maintain high-precision prediction and compensation, effectively shortening the product development cycle and improving production flexibility.

[0018] Springback compensation often requires multiple manual interventions, resulting in low efficiency. The automatic compensation device of this solution realizes a fully automated process from parameter input, springback prediction to process parameter adjustment. On a continuous stamping production line, it can monitor and compensate for springback in real time, reducing manual operations and downtime. The production efficiency is improved compared with traditional methods, the labor cost is reduced, and at the same time, the stability and consistency of the production process are enhanced. Since this solution improves the prediction accuracy and production efficiency, reduces the scrap rate and manual intervention, the cost in the mold development and production process is reduced.

[0019] Compared with traditional numerical models or physical models, this method and device have more advantages in terms of accuracy, adaptability, real-time performance and cost control, especially showing excellent performance when dealing with complex high-strength sheet metals and variable process conditions. Compared with the prediction method of a single model, this solution is more flexible and robust, can better cope with the complexity and variability in the stamping process, reduce the risk of model errors and improve the prediction accuracy.

[0020] By integrating the advantages of multiple models, this method can handle diverse data and feature relationships more comprehensively, and has significant superiority in improving prediction accuracy, calculation efficiency and model optimization. At the same time, this method fully considers the influence of process parameters and material properties on springback, further improving the accuracy and efficiency of the stamping process.

[0021] Furthermore, in S01, a deep learning algorithm is used to perform secondary analysis on the data obtained from transmission electron microscopy and atom probe tomography techniques, automatically identifying and marking regions with special microstructural features. Transmission electron microscopy and atom probe tomography techniques generate a large amount of data, which contains numerous microstructural information. The deep learning algorithm can quickly process this massive amount of data, automatically identifying regions with special microstructural features, avoiding the cumbersome process of manually checking and analyzing the data one by one, and greatly saving time and labor costs. For example, when analyzing the microstructure of large plates, it may take days or even weeks for manual screening, while the deep learning algorithm may be able to complete the same task within a few hours; the real-time processing ability of the deep learning algorithm can provide analysis results in a short time, enabling operators to adjust production parameters in a timely manner according to the results. For example, when the algorithm identifies a special microstructural region in the plate that affects springback, the operator can immediately adjust the stamping process to avoid producing unqualified products. At the same time, during the subsequent construction and training process of the springback prediction model, the data of the accurately marked special microstructural feature regions can be used as important input parameters. These data can more accurately reflect the relationship between the microstructure of the material and the springback performance, thereby improving the accuracy and reliability of the springback prediction model. For example, when training models such as support vector regression and kernel ridge regression, using the data processed by the deep learning algorithm can enable the model to better learn the mapping relationship between microstructural features and springback, improving the prediction accuracy.

[0022] Furthermore, in S02, the improved UMAT subroutine adopts parallel computing optimization technology, using the parallel computing ability of the graphics processing unit GPU to accelerate the stamping springback simulation process. By distributing the computing tasks in the UMAT subroutine to multiple cores of the GPU for parallel execution, the time required for the simulation can be significantly shortened. For example, for the springback simulation of a complex stamping part, using traditional CPU serial computing may take hours or even days, while using GPU parallel computing may only take dozens of minutes, greatly improving the simulation efficiency; the stamping springback simulation accelerated by GPU can provide accurate results in a shorter time, enabling technicians to quickly adjust the production process according to the simulation results and reducing the production cycle; for large stamping parts or complex stamping processes, the simulation calculation amount will increase sharply. The parallel computing ability of the GPU can effectively handle these large-scale problems, avoiding model simplification or incomplete calculation caused by insufficient computing resources. For example, in the stamping springback simulation of automotive body panels, using GPU can perform accurate simulation analysis on the entire body panel without excessive simplification of the model.

[0023] Furthermore, in S04, the attention mechanism of the improved ResNet-50 model adopts a dynamic weight update strategy, which adjusts the attention weights in real time according to the characteristics of different microstructure images. Different microstructure images have differences in element distribution, grain boundary characteristics, dislocation density, etc., and the degrees of influence of these differences on stamping springback prediction are also different. The dynamic weight update strategy enables the attention mechanism to automatically focus on the regions and features that are most critical for springback prediction according to the characteristics of each image. For example, in some images, the special morphology of grain boundaries may be the key factor affecting springback. Dynamic weight adjustment can make the model pay more attention to the characteristics of the grain boundary region, so as to more accurately extract these important information and provide strong support for subsequent springback prediction. There may be some information in the microstructure image that has little relevance to springback prediction. If the model gives the same attention to all regions and features, it will increase the computational load and may also introduce noise interference, reducing the accuracy of the model. By dynamically adjusting the attention weights, the model can reduce the attention to this redundant information and allocate more computational resources to process key features, improving the efficiency and quality of feature extraction.

[0024] Furthermore, the adaptive parameter adjustment algorithm in S03 dynamically adjusts the change step of process parameters based on the historical simulation data and characteristics of the material. In the initial stage of stamping simulation, the value range of process parameters is usually relatively wide. If a fixed step is used for parameter traversal, a large amount of computational resources and time may be wasted in the invalid parameter region. The adaptive parameter adjustment algorithm can quickly identify the parameter intervals that may produce effective results based on historical simulation data and material characteristics, and appropriately reduce the step size for fine search within these intervals, while increasing the step size to quickly skip in the invalid intervals, thus greatly shortening the time required for simulation.

[0025] Furthermore, the improved ResNet-50 model in S04 incorporates an attention mechanism that can automatically focus on regions in the microstructure diagram. The attention mechanism can automatically identify regions closely related to stamping springback, such as special grain boundaries, dislocation dense regions, etc., and assign higher attention weights to these key regions. In this way, the model can focus more on key information during feature extraction, avoiding being interfered by irrelevant or secondary information, and thus more accurately capturing the key features affecting stamping springback. The attention mechanism can keenly capture these subtle features and give sufficient attention. For example, differences in element distribution at the atomic scale may enable the model to extract these subtle but key features through the focus of the attention mechanism, providing richer and more accurate information for subsequent prediction.

[0026] Furthermore, the transfer learning strategy in S04 initializes five machine learning models with the model parameters pre-trained in related fields to accelerate model convergence and reduce training time.

[0027] Furthermore, the dynamic weight allocation algorithm dynamically adjusts the weights of each model in the integrated model according to the evaluation indicators of the model under different working conditions. The working conditions during the stamping process are complex and variable, and factors such as different material properties, process parameters, and environmental conditions will all affect springback. It is difficult for a single model to maintain good performance under all working conditions. The dynamic weight allocation algorithm can dynamically adjust the weights according to the real-time working conditions and the evaluation results of each model, enabling the integrated model to better adapt to various complex working conditions and provide more accurate springback predictions. The distribution of stamping data may change with factors such as time and production batches. The dynamic weight allocation algorithm can monitor the performance of each model on new data in real time and dynamically adjust the weights according to the evaluation indicators, enabling the integrated model to quickly adapt to the changes in data distribution and maintain stable prediction performance. At the same time, the performance of a single model may fluctuate due to factors such as data noise and outliers. Through the dynamic weight allocation algorithm, the integrated model can synthesize the prediction results of multiple models and dynamically adjust the weights according to the evaluation indicators of each model, reducing the impact of the fluctuations of a single model on the overall prediction result and improving the stability of the model. Description of the Drawings

[0028] Figure 1 It is a flowchart of the steps of the method for predicting springback of high-strength sheet metal stamping parts based on multiple machine learning models according to an embodiment of the present invention; Figure 2 It is a flowchart of data set collection and processing and model training of the method for predicting springback of high-strength sheet metal stamping parts based on multiple machine learning models according to an embodiment of the present invention; Figure 3 It is a flowchart of KNN execution of the method for predicting springback of high-strength sheet metal stamping parts based on multiple machine learning models according to an embodiment of the present invention; Figure 4 It is a flowchart of RFE execution of the method for predicting springback of high-strength sheet metal stamping parts based on multiple machine learning models according to an embodiment of the present invention; Figure 5 It is a flowchart of SVM execution of the method for predicting springback of high-strength sheet metal stamping parts based on multiple machine learning models according to an embodiment of the present invention; Figure 6 It is a flowchart of kernel ridge regression execution of the method for predicting springback of high-strength sheet metal stamping parts based on multiple machine learning models according to an embodiment of the present invention; Figure 7 It is a flowchart of boosting tree execution of the method for predicting springback of high-strength sheet metal stamping parts based on multiple machine learning models according to an embodiment of the present invention; Figure 8 It is the final model structure diagram of the method for predicting springback of high-strength sheet metal stamping parts based on multiple machine learning models according to an embodiment of the present invention; Figure 9The restNet-50 network structure diagram of the high-strength sheet metal stamping part springback prediction method based on multiple machine learning models according to the embodiments of the present invention; Figure 10 The structural schematic diagram of the automatic compensation device for high-strength sheet metal stamping parts based on multiple machine learning models according to the embodiments of the present invention. Specific implementation manners

[0029] Example 1, as Figure 1-10 shown, an automatic compensation device for high-strength sheet metal stamping parts based on multiple machine learning models includes the following modules: A visual parameter input module for visually setting process parameters and material characteristic parameters for the automatic springback compensation device and inputting the parameters into the embedded springback prediction module; An embedded springback prediction module, the prediction model of which is generated according to the high-strength sheet metal stamping part springback prediction method based on multiple machine learning models and then embedded in the device for predicting the springback vector value during the stamping process on the premise of providing process parameter material characteristic parameters; A springback compensation calculation module for calculating the springback compensation vector value; An automatic control module for generating control signals to control and adjust the process.

[0030] A stamping prediction method based on multiple machine learning models, the method includes: S01 Microstructure and feature acquisition: Use a transmission electron microscope combined with atom probe tomography technology to obtain the microstructure diagram of the sheet metal, measure the microstructure characteristics of element distribution, grain boundary characteristics and dislocation density at the atomic scale, and obtain the characteristics of the elastic modulus and hardening index of the material; S02 Initial simulation sampling: Use the finite element method to sample the stamping parts, and perform stamping springback simulation with the help of ABAQUS combined with the improved UMAT subroutine, and record the springback vector data; S03 Dataset expansion: For various stamping materials, use the adaptive parameter adjustment algorithm to traverse the initial process parameters, adjust the parameter step size according to the material characteristics and the previous simulation results, and obtain the stamping springback dataset through simulation sampling; S04 Model construction and training: Use the improved ResNet-50 model, integrate the attention mechanism to extract features from the microstructure diagram. Integrate the extracted features with the material characteristic parameters and process parameters, divide the dataset into a training set, a validation set and a test set, and use transfer learning and ensemble learning strategies to train five models of support vector regression, kernel ridge regression, KNN, random forest regression and gradient boosting tree. With the root mean square error RMSE, mean absolute percentage error MAPE and coefficient of determination As the evaluation index, according to the evaluation results, a dynamic weight allocation algorithm is used to integrate into the final springback prediction model. Among them, is used to measure the fitting degree of the model to the data. The closer it is to 1, the better the fitting effect of the model. Calculate , and the formula is

[0031] where is the true value, is the predicted value, is the mean of the true values, n is the number of samples; MSE (Mean Squared Error), which measures the average of the squares of the differences between the predicted values and the true values. The formula is

[0032] where is the true value, is the predicted value, n is the number of samples; MAPE (Mean Absolute Percentage Error), which measures the relative error between the predicted values and the true values, expressed as a percentage. The formula is

[0033] According to the ratio of each model , determine the weight of each model in the final prediction, and integrate the models into a compensation prediction model. Input the material properties and process parameters, and this model outputs the springback prediction value; S05 Automatic Compensation: Embed the final springback prediction model into an automatic compensation device. The visual parameter input module of this device receives the process parameters and material property parameters input by the operator; the embedded springback prediction model predicts the springback vector; the springback compensation calculation module calculates the springback compensation value according to the prediction results and the preset compensation rules; the automatic compensation control module generates a control signal according to the compensation value to adjust the process parameters of the stamping equipment for the automatic compensation of the springback of the stamping parts.

[0034] The principle of this solution is based on the deep integration of multi-disciplinary technologies. In the stage of obtaining the microstructure and characteristics, a transmission electron microscope combined with atom probe tomography technology is used to deeply analyze the sheet metal microstructure at the atomic scale, accurately measure key characteristics such as element distribution, grain boundary characteristics, and dislocation density, and at the same time obtain the elastic modulus and hardening index of the material, providing a solid microstructural basic data for subsequent analysis, and revealing the influence mechanism of the material's inherent characteristics on springback.

[0035] In the initial simulation sampling stage, the stamping parts are sampled by means of the finite element method. The stamping springback simulation is carried out by using the ABAQUS software and combining with the improved UMAT subroutine. In this way, the microstructural characteristics are incorporated into the simulation process to more accurately simulate the actual deformation of the sheet metal during the stamping process, so as to record reliable springback vector data and provide real and effective samples for subsequent model training.

[0036] In the dataset expansion stage, for various stamping materials, the adaptive parameter adjustment algorithm is used to traverse the initial process parameters. This algorithm dynamically adjusts the parameter step size according to the material characteristics and the previous simulation results, and intelligently obtains the stamping springback data under different process parameter combinations, effectively expanding the diversity and representativeness of the dataset, and ensuring that the subsequent model can learn rich stamping springback characteristics.

[0037] In the process of model construction and training, the improved ResNet-50 model incorporating the attention mechanism is used to extract features from the microstructure diagram, highlighting the key microstructural features. The extracted features are integrated with the material characteristic parameters and process parameters. After dividing the dataset, five models such as support vector regression and kernel ridge regression are trained by using transfer learning and ensemble learning strategies. Taking the root mean square error, mean absolute percentage error and coefficient of determination as evaluation indicators, the five models are integrated into the final springback prediction model by using the dynamic weight allocation algorithm, giving full play to the advantages of each model and improving the accuracy and reliability of the prediction.

[0038] Finally, in the automatic compensation stage, the final springback prediction model is embedded in the automatic compensation device. The parameters input by the operator are obtained through the visual parameter input module. The embedded springback prediction model predicts the springback vector. The springback compensation calculation module calculates the compensation value according to the preset rules, and the automatic compensation control module adjusts the process parameters of the stamping equipment according to the compensation value to achieve real-time and accurate compensation for the springback of the stamping parts.

[0039] Compared with the existing technology, the traditional empirical method and manual adjustment method rely on personal experience, and the accuracy is difficult to guarantee, and the part size deviation is large. However, this solution uses the accurate data obtained from the microstructure and features, combined with complex model construction and training, to improve the springback prediction accuracy. For example, in the production of automotive parts, the size deviation of the high-strength sheet metal stamping parts produced by the traditional method may reach ±1 mm. After adopting this solution, the size deviation can be controlled within ±0.2 mm, significantly improving the product quality and reducing the scrap rate.

[0040] Although traditional finite element analysis uses numerical simulation, it has poor adaptability to changes in material properties and process parameters. The adaptive parameter adjustment algorithm and multi-model integration strategy of this solution enable it to quickly adapt to different stamping materials and process conditions. When producing high-strength sheet metal stampings for different models of automobiles, it can quickly switch and maintain high-precision prediction and compensation without a large number of readjustments, effectively shortening the product development cycle and improving production flexibility.

[0041] Springback compensation often requires multiple manual interventions and has low efficiency. The automatic compensation device of this solution realizes a fully automated process from parameter input, springback prediction to process parameter adjustment. On a continuous stamping production line, it can monitor and compensate for springback in real time, reducing manual operations and downtime. The production efficiency is improved compared with traditional methods, the labor cost is reduced, and at the same time, the stability and consistency of the production process are improved. Since this solution improves the prediction accuracy and production efficiency, reduces the scrap rate and manual intervention, the cost in the mold development and production process is reduced.

[0042] In S01, a deep learning algorithm is used to perform secondary analysis on the data obtained by transmission electron microscopy and atom probe tomography technologies, automatically identifying and marking regions with special microstructural features. Transmission electron microscopy and atom probe tomography technologies generate a large amount of data, which contains numerous microstructural information. The deep learning algorithm can quickly process this massive data and automatically identify regions with special microstructural features, avoiding the cumbersome process of manually viewing and analyzing data one by one, and greatly saving time and labor costs. For example, when analyzing the microstructure of large sheets, it may take days or even weeks for manual screening, while the deep learning algorithm may be able to complete the same task within a few hours; the real-time processing ability of the deep learning algorithm can provide analysis results in a short time, enabling operators to adjust production parameters in a timely manner according to the results. For example, when the algorithm identifies a special microstructural region in the sheet that affects springback, the operator can immediately adjust the stamping process to avoid producing unqualified products. At the same time, during the subsequent construction and training of the springback prediction model, the data of the accurately marked special microstructural feature regions can be used as important input parameters. These data can more accurately reflect the relationship between the microstructure of the material and the springback performance, thereby improving the accuracy and reliability of the springback prediction model. For example, when training models such as support vector regression and kernel ridge regression, using the data processed by the deep learning algorithm can enable the model to better learn the mapping relationship between microstructural features and springback and improve the prediction accuracy.

[0043] In S02, the improved UMAT subroutine adopts parallel computing optimization technology and utilizes the parallel computing power of the Graphics Processing Unit (GPU) to accelerate the stamping springback simulation process. By distributing the computing tasks in the UMAT subroutine to multiple cores of the GPU for parallel execution, the simulation time can be significantly shortened. For example, for the springback simulation of a complex stamping part, it may take several hours or even days using traditional CPU serial computing, while it may only take dozens of minutes using GPU parallel computing, greatly improving the simulation efficiency. The stamping springback simulation accelerated by GPU can provide accurate results in a shorter time, enabling technicians to quickly adjust the production process according to the simulation results and reduce the production cycle. For large stamping parts or complex stamping processes, the simulation computing volume will increase sharply. The parallel computing power of the GPU can effectively handle these large-scale problems and avoid model simplification or incomplete calculations caused by insufficient computing resources. For example, in the stamping springback simulation of automotive body panels, the GPU can be used to perform accurate simulation analysis on the entire body panel without excessive model simplification.

[0044] In S04, the attention mechanism of the improved ResNet-50 model adopts a dynamic weight update strategy, which adjusts the attention weights in real time according to the characteristics of different microstructure images. Different microstructure images have differences in element distribution, grain boundary characteristics, dislocation density, etc., and the degrees of influence of these differences on stamping springback prediction are also different. The dynamic weight update strategy enables the attention mechanism to automatically focus on the regions and features that are most critical for springback prediction according to the characteristics of each image. For example, in some images, the special morphology of the grain boundary may be a key factor affecting springback. The dynamic weight adjustment can make the model pay more attention to the characteristics of the grain boundary region, thus more accurately extracting this important information and providing strong support for subsequent springback prediction. There may be some information in the microstructure image that has little relevance to springback prediction. If the model gives the same attention to all regions and features, it will increase the computing volume and may also introduce noise interference, reducing the accuracy of the model. By dynamically adjusting the attention weights, the model can reduce the attention to this redundant information and allocate more computing resources to process key features, improving the efficiency and quality of feature extraction.

[0045] The adaptive parameter adjustment algorithm in S03 dynamically adjusts the change step of process parameters based on the historical simulation data and characteristics of the material. In the initial stage of stamping simulation, the value range of process parameters is usually relatively wide. If a fixed step is used for parameter traversal, a large amount of computing resources and time may be wasted in the invalid parameter area. The adaptive parameter adjustment algorithm can quickly identify the parameter intervals that may produce effective results based on historical simulation data and material characteristics, and appropriately reduce the step within these intervals for fine search, while increasing the step to quickly skip in the invalid intervals, thus greatly shortening the time required for simulation.

[0046] The improved ResNet-50 model in S04 incorporates an attention mechanism, which can automatically focus on the regions in the microstructure diagram. The attention mechanism can automatically identify the regions closely related to stamping springback, such as special grain boundaries, dislocation dense regions, etc., and assign higher attention weights to these key regions. In this way, the model can focus more on key information during feature extraction, avoiding being interfered by irrelevant or secondary information, and thus more accurately capture the key features affecting stamping springback; the attention mechanism can sensitively capture these subtle features and give sufficient attention. For example, the differences in element distribution at the atomic scale may enable the model to extract these subtle but key features through the focusing of the attention mechanism, providing richer and more accurate information for subsequent predictions.

[0047] The transfer learning strategy in S04 initializes five machine learning models with the model parameters pre-trained in related fields to accelerate model convergence and reduce training time.

[0048] The dynamic weight allocation algorithm dynamically adjusts the weights of each model in the integrated model according to the evaluation indicators of the model under different working conditions. The working conditions during the stamping process are complex and variable, and factors such as different material characteristics, process parameters, and environmental conditions will all affect springback. It is difficult for a single model to maintain good performance under all working conditions. The dynamic weight allocation algorithm can dynamically adjust the weights according to the real-time working conditions and the evaluation results of each model, enabling the integrated model to better adapt to various complex working conditions and provide more accurate springback predictions. The distribution of stamping data may change with factors such as time and production batches. The dynamic weight allocation algorithm can monitor the performance of each model on new data in real time and dynamically adjust the weights according to the evaluation indicators, enabling the integrated model to quickly adapt to the change in data distribution and maintain stable prediction performance; at the same time, the performance of a single model may fluctuate due to factors such as data noise and outliers. Through the dynamic weight allocation algorithm, the integrated model can synthesize the prediction results of multiple models and dynamically adjust the weights according to the evaluation indicators of each model, reducing the impact of the fluctuation of a single model on the overall prediction result and improving the stability of the model.

[0049] In actual application Experimental equipment preparation: Select a high-resolution transmission electron microscope (TEM) and an atom probe tomography (APT) device to ensure that the performance indicators such as resolution and precision of the equipment meet the requirements for observing the microstructure of high-strength sheet materials. Calibrate and debug the equipment to ensure the accuracy of the acquired data.

[0050] Sample preparation: Select representative samples from high-strength sheet materials of different batches and different production processes. According to the requirements of TEM and APT, prepare the samples into appropriate sizes and shapes. For TEM, the sample thickness needs to be controlled at about dozens of nanometers; for APT, the samples are usually prepared into needle shapes.

[0051] Data acquisition: Use TEM to observe the samples, obtain the microstructure diagrams of the sheet materials, and record information such as element distribution, grain boundary characteristics, and dislocation density. At the same time, use APT technology to accurately measure the microstructure characteristics such as element distribution and grain boundary chemical composition at the atomic scale.

[0052] Material property testing: Use a dynamic mechanical analyzer (DMA) to measure the elastic modulus of the material, and obtain the hardening index of the material through tensile tests combined with a microstructure evolution model. Record these material property data and store them corresponding to the microstructure data.

[0053] Deep learning secondary analysis: Import the acquired microstructure data into an analysis system based on deep learning algorithms. This system uses a convolutional neural network (CNN) architecture and is trained with a large amount of labeled microstructure image data to learn the patterns of special microstructure features. After training, perform secondary analysis on the data obtained by TEM and APT, automatically identify and mark the regions with special microstructure features, such as abnormal grain boundaries and high-density dislocation regions. Integrate the analysis results with the original data to provide more accurate data support for subsequent simulations and model training.

[0054] Finite element model establishment: Use professional finite element analysis software such as ABAQUS to establish an accurate three-dimensional finite element model according to the design drawings and actual dimensions of the stamping parts. Perform reasonable mesh division on the model, and use finer meshes in key areas such as those prone to springback to improve the calculation accuracy.

[0055] Improved UMAT subroutine writing: Based on the microstructure characteristics and mechanical properties of the material, write an improved UMAT subroutine. In the subroutine, consider the influence of the microstructure on the constitutive relationship of the material, such as grain boundary strengthening and dislocation slip mechanisms. Import the written UMAT subroutine into the ABAQUS software to ensure its compatibility and stability with the software.

[0056] GPU Parallel Computing Configuration: Build a computing platform with a high-performance Graphics Processing Unit (GPU), install the corresponding driver programs and parallel computing software such as CUDA. Parallelize the improved UMAT subroutine, and reasonably distribute the computing tasks to multiple cores of the GPU. Set it up in the ABAQUS software to enable the GPU acceleration computing function.

[0057] Stamping Springback Simulation: Set the initial process parameters for the stamping material, such as forming blank holding force, lubrication conditions, forming speed, etc. Sample the stamping parts using the finite element method, and conduct stamping springback simulation in the ABAQUS software combined with the improved UMAT subroutine. During the simulation process, monitor the computing progress and results in real time to ensure the stability and accuracy of the simulation. Record the springback vector data, including displacement, stress, strain and other information of each sampling point, as the sample data under the initial process parameter conditions.

[0058] Implementation of Adaptive Parameter Adjustment Algorithm: Develop an adaptive parameter adjustment algorithm. Based on the historical simulation data and characteristics of the material, this algorithm adopts the idea of reinforcement learning, aiming to maximize the prediction accuracy, and dynamically adjusts the change step of the process parameters. According to the previous simulation results, the algorithm evaluates the influence of different process parameter combinations on springback, and intelligently selects the parameter step for the next simulation.

[0059] Multi-Material and Multi-Parameter Traversal Simulation: For multiple stamping materials, use the adaptive parameter adjustment algorithm to traverse the initial process parameters. During the traversal process, adjust the parameter values of each parameter according to the algorithm to obtain different process parameter conditions. Under each corresponding process parameter condition, repeat the steps of the initial simulation sampling, conduct finite element sampling on the stamping parts, and conduct stamping springback simulation through ABAQUS combined with the improved UMAT subroutine, and record the springback vector data. Integrate the sample data under the initial process parameter conditions and the sample data under each corresponding process parameter condition to form a stamping springback dataset.

[0060] Improved ResNet-50 Model Setting: Adopt the improved ResNet-50 model, and integrate the attention mechanism module into its convolutional layer. The attention mechanism module adopts a dynamic weight update strategy, and adjusts the attention weights in real time according to the characteristics of different microstructure images. Initialize the improved ResNet-50 model and load the pre-trained model parameters in related fields, such as the pre-trained parameters in the field of material microstructure analysis.

[0061] Dataset Processing: Integrate the microstructure diagrams, material property parameters and process parameters to construct a dataset. Divide the dataset into a training set, a validation set and a test set according to the ratio of 70%, 15%, 15%. Normalize the data to make the data with different features have the same scale, and improve the model training efficiency and accuracy.

[0062] Model training: Using transfer learning and ensemble learning strategies, five models including Support Vector Regression (SVM), Kernel Ridge Regression, K-Nearest Neighbor (KNN), Random Forest Regression, and Gradient Boosting Tree are trained. During the training process, the grid search method is used to find the optimal parameters for each model. The Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R²) are used as evaluation metrics to evaluate the model performance. According to the evaluation results, a dynamic weight allocation algorithm is used to calculate the ratio of R² of each model and determine the weight of each model in the final prediction. The five models are integrated into the final springback prediction model. During the training process, the training progress and performance metrics of the model are monitored in real-time, and the training parameters are adjusted according to the results of the validation set to prevent overfitting of the model.

[0063] Automatic compensation device construction: An automatic compensation device is constructed, including a visual parameter input module, an embedded springback prediction module, a springback compensation calculation module, and an automatic control module. The visual parameter input module is designed with a graphical interface to provide an intuitive and convenient operation interface for operators. Operators can input process parameters such as forming blank holding force, lubrication condition parameters, forming speed parameters, etc. and material property parameters such as material thickness, yield strength, tensile strength, etc. on the interface. The embedded springback prediction module embeds the final springback prediction model, receives the parameters transmitted by the visual parameter input module, and predicts the springback vector value during the stamping process. The springback compensation calculation module calculates the springback compensation vector value based on the prediction results and preset compensation rules, which are formulated based on the principles of material mechanics and empirical formulas. The automatic control module generates control signals according to the compensation value and adjusts the process parameters of the stamping equipment, such as die clearance, stamping pressure, etc., through the interface with the stamping equipment to achieve automatic compensation for the springback of stamping parts.

[0064] System integration and debugging: The automatic compensation device is integrated with the stamping production equipment, and a data transmission channel is established to ensure smooth data interaction between modules. The entire system is debugged, and the prediction accuracy and compensation effect of the system are tested by inputting different process parameters and material property parameters. During the debugging process, the system is optimized and adjusted to improve the stability and reliability of the system.

[0065] Production application and optimization: The stamping prediction method based on multiple machine learning models and the automatic compensation device are applied to actual stamping production. During the production process, new stamping data, including process parameters, material properties, and springback data, are continuously collected to update the stamping springback dataset. The final springback prediction model is retrained and optimized regularly, and the model parameters and weights are adjusted according to the new data to improve the prediction accuracy and adaptability of the model. At the same time, according to the actual production situation, the preset compensation rules and control strategies of the automatic compensation device are adjusted and optimized to further improve the quality and efficiency of stamping production.

[0066] Example 2 As shown in the Figure 1-10 appendix of the specification, the method for predicting springback of high-strength sheet metal stamping parts based on multiple machine learning models in the embodiment of the present invention includes four steps: S101: Using experimental techniques such as scanning electron microscope (SEM), transmission electron microscope (TEM), or X-ray diffraction (XRD) to obtain the microstructure diagram of the high-strength sheet metal, the grain morphology, grain size, orientation, texture and other microstructure characteristics of the high-strength sheet metal can be clearly observed. These structure characteristics directly affect the strength, toughness, springback, fatigue resistance and other properties of the high-strength sheet metal. The microstructure affects the springback of the material to a great extent, and the addition of microstructure characteristics can greatly enhance the springback prediction ability of the model.

[0067] S102: The process parameter conditions include factors such as forming blank holding force, part size parameters, lubrication conditions, and forming speed. Sampling the stamping parts by the finite element method, combining with the material definition module of CAE software, such as the UMAT subroutine of ABAQUS and microstructure analysis, to complete springback prediction and generate sample data. Specifically, specify the sampling positions in the finite element model of the stamping parts, use CAE simulation to evaluate the springback situation, obtain the size deviation vectors of each sampling point after springback, and use them as the sample data of the initial process parameters.

[0068] S103: The material properties of a certain specification of material are fixed. As long as the process parameters are changed by looping through and using the sampling method in step S102, the data of different process parameters of one material can be obtained. Similarly, by selecting different specifications of raw materials and then looping through to change the process parameters and using CAE forming springback prediction, the data of different process parameters of multiple materials can be obtained, and thus a data set can be obtained.

[0069] Step S104, use the pre-trained resNet-50 model to extract features from the microstructure diagram, and re-integrate the data set by combining the corresponding material property parameters and process parameters obtained. Then divide the test set and the training set according to the ratio of 2:8, and then train five models. Use the trained models to test on the test set and calculate the , MSE, MAPE and other performance indicators of each model, and then as Figure 10 shown, combine the models according to the ratio to form the final model. The essence of this integrated model is that when calling the trained models, first input the predicted parameters into each model, and finally the output results of the five models are combined according to the The proportional combinations are used to obtain the final predicted data. Compared with the prediction method of a single model, this multi-model integration method has stronger flexibility and robustness, can effectively cope with the variability and complexity in the stamping process of high-strength sheet metal, reduce the risk of the model, and improve the prediction accuracy. By combining the advantages of multiple models, different types of data and feature relationships can be processed more comprehensively, and greater advantages are achieved in terms of prediction accuracy, computational efficiency, and model optimization. Moreover, this method fully considers the influence of process parameters and material properties on springback.

[0070] In regression analysis, R² is an index to measure the goodness of fit of the model, indicating the proportion of the explanation of the input of independent variables to the output of dependent variables. The value range of R² is from 0 to 1, and the closer it is to 1, the better the model fitting.

[0071]

[0072] where is the true value, is the predicted value, is the mean of the true values, n is the number of samples; MSE, the mean squared error, measures the average of the squares of the differences between the predicted values and the true values. The formula is

[0073] where is the true value, is the predicted value, n is the number of samples; MAPE, the mean absolute percentage error, measures the relative error between the predicted values and the true values, expressed as a percentage. The formula is

[0074] MSE is a commonly used loss function in regression problems, representing the average of the squares of the differences between the predicted values and the true values. The formula:

[0075] where is the true value, is the predicted value, n is the number of samples; MAPE measures the relative error between the predicted values and the true values, usually expressed as a percentage. It can be used to evaluate the accuracy of regression models. Especially in prediction problems with different dimensions, MAPE provides a normalized error measure.

[0076]

[0077] where is the true value, Is the predicted value As shown in the attached instructions Figure 10 As shown, the automatic springback compensation device of the embodiment of the present invention includes: A visual parameter input module, which is used to visually set process parameters and material characteristic parameters for the automatic springback compensation device, and input the parameters into the embedded springback prediction module.

[0078] As an optional example, the visual parameter input module in the embodiment of the present invention is specifically used for: visually inputting the parameters required for prediction into the embedded prediction module.

[0079] An embedded springback prediction module, the prediction model of which is generated according to the high-strength sheet metal stamping part springback prediction method based on multiple machine learning models, and then embedded in the device, which is used to predict the springback vector value during the stamping process on the premise of providing process parameters and material characteristic parameters.

[0080] As an optional example, the embedded springback prediction module of the embodiment of the present invention is specifically used for: predicting the springback during the stamping process under the determined process parameter conditions and material characteristics to obtain the springback vector value. The prediction of the springback vector value can be used to calculate the springback compensation vector to adjust the process to obtain an ideal stamping part; A springback compensation calculation module, which is used to calculate the springback compensation vector value; As an optional example, the springback compensation calculation module of the embodiment of the present invention is specifically used for: calculating and converting the springback amount value predicted by the embedded springback prediction module into a springback compensation value.

[0081] An automatic control module, which is used to generate a control signal to control and adjust the process.

[0082] As an optional example, the automatic control module of the embodiment of the present invention is specifically used for: converting the springback compensation vector value calculated by the springback compensation calculation module into a control signal to adjust the process to obtain an ideal stamping part.

[0083] The above are only embodiments of the present invention. Specific structures and common knowledge such as characteristics that are well-known in the art are not described in detail herein. Those of ordinary skill in the art know all the common general technical knowledge in the technical field to which the invention pertains before the filing date or the priority date, are able to learn all the prior art in this field, and have the ability to apply conventional experimental means before this date. Those of ordinary skill in the art can, under the inspiration given in this application, improve and implement this solution in combination with their own abilities. Some typical well-known structures or well-known methods should not become an obstacle for those of ordinary skill in the art to implement this application. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can still be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application should be based on the content of its claims, and the specific implementation manners and the like recorded in the specification can be used to interpret the content of the claims.

Claims

1. A stamping prediction method based on multiple machine learning models, characterized in that, The method includes: S01 Microstructure and feature acquisition: Using a transmission electron microscope combined with atom probe tomography technology to obtain the microstructure diagram of the sheet metal, measuring the microstructure characteristics of element distribution, grain boundary characteristics, and dislocation density at the atomic scale, and obtaining the characteristics of the elastic modulus and hardening index of the material; S02 Initial simulation sampling: Using the finite element method to sample the stamping parts, and using ABAQUS combined with an improved UMAT subroutine to perform stamping springback simulation, and recording the springback vector data; S03 Dataset expansion: For various stamping materials, using an adaptive parameter adjustment algorithm to traverse the initial process parameters, adjusting the parameter step size according to the material characteristics and the previous simulation results, and obtaining the stamping springback dataset through simulation sampling; S04 Model Construction and Training: An improved ResNet-50 model is adopted, integrating an attention mechanism to extract features from the microstructure diagram. The extracted features are integrated with material property parameters and process parameters, and the dataset is divided into a training set, a validation set, and a test set. Transfer learning and ensemble learning strategies are used to train five models: support vector regression, kernel ridge regression, KNN, random forest regression, and gradient boosting tree. The root mean square error RMSE, mean absolute percentage error MAPE, and coefficient of determination are used as evaluation indicators, and a dynamic weight allocation algorithm is used to integrate them into the final springback prediction model according to the evaluation results. Among them, is used to measure the fitting degree of the model to the data. The closer it is to 1, the better the fitting effect of the model. Calculate , and the formula is wherein is the true value, is the predicted value, is the mean of the true values, n is the number of samples; MSE, the mean squared error, measures the average of the squares of the differences between the predicted values and the true values. The formula is wherein is the true value, is the predicted value, n is the number of samples; MAPE, the mean absolute percentage error, measures the relative error between the predicted value and the true value, expressed as a percentage. The formula is According to the ratios of each model to determine the weight of each model in the final prediction, integrate the models into a compensation prediction model, input the material properties and process parameters, and this model outputs the springback prediction value; S05 Automatic compensation: Embedding the final springback prediction model into the automatic compensation device. The visual parameter input module of this device receives the process parameters and material characteristic parameters input by the operator; the embedded springback prediction model predicts the springback vector; the springback compensation calculation module calculates the springback compensation value according to the prediction result and the preset compensation rule; the automatic compensation control module generates a control signal according to the compensation value to adjust the process parameters of the stamping equipment for automatic compensation of the springback of the stamping parts.

2. The stamping prediction method based on multiple machine learning models according to claim 1, wherein, In the above S01, a deep learning algorithm is used to perform secondary analysis on the data obtained by the transmission electron microscope and atom probe tomography technology, and automatically identify and mark the areas with special microstructure characteristics.

3. A stamping prediction method based on multiple machine learning models according to claim 1, characterized in that In the above S02, the improved UMAT subroutine adopts parallel computing optimization technology, and uses the parallel computing ability of the graphics processing unit GPU to accelerate the stamping springback simulation process.

4. A stamping prediction method based on multiple machine learning models according to claim 1, characterized in that, In the above S04, the attention mechanism of the improved ResNet-50 model adopts a dynamic weight update strategy, and adjusts the attention weights in real time according to the characteristics of different microstructure images.

5. A stamping prediction method based on multiple machine learning models according to claim 1, characterized in that, The adaptive parameter adjustment algorithm in the above S03 dynamically adjusts the change step size of the process parameters based on the historical simulation data and characteristics of the material.

6. The stamping prediction method based on multiple machine learning models according to claim 1, wherein, The improved ResNet-50 model in the above S04 incorporates an attention mechanism, which can automatically focus on the key areas in the microstructure diagram.

7. A stamping prediction method based on multiple machine learning models according to claim 1, characterized in that In the above S04, the transfer learning strategy initializes five machine learning models using the model parameters pre-trained in related fields to accelerate model convergence and reduce training time.

8. A stamping prediction method based on multiple machine learning models according to claim 1, characterized in that, The dynamic weight allocation algorithm dynamically adjusts the weights of each model in the integrated model according to the evaluation indexes of the model under different working conditions.

9. An automatic compensation device for high-strength sheet metal stamping parts based on multiple machine learning models, which uses a stamping prediction method based on multiple machine learning models according to any one of claims 1-8, and is characterized in that, It includes: A visual parameter input module, which is used to visually set the process parameters and material characteristic parameters for the automatic springback compensation device, and input the parameters into the embedded springback prediction module; An embedded springback prediction module, the prediction model of which is generated according to the high-strength sheet metal stamping part springback prediction method based on multiple machine learning models, and then embedded into this device, which is used to predict the springback vector value during the stamping process on the premise of providing the process parameters and material characteristic parameters; A springback compensation calculation module, which is used to calculate the springback compensation vector value; An automatic control module, which is used to generate a control signal to control and adjust the process.

Images