Methods and systems for predicting and compensating for springback during stamping of extended longitudinal beams in automobiles

Abstract

This application discloses a method for predicting and compensating springback during stamping of extended automotive longitudinal beams. The method includes: constructing an initial dataset containing initial predicted values ​​based on the influence of preset process parameters on springback; building a springback prediction model using incremental plasticity theory, then training the springback prediction model using a gradient boosting decision tree and adjusting process parameters using a Bayesian optimization method; simulating the manufacturing of the longitudinal beam using finite element simulation based on the optimal process parameters, generating simulated springback data, and optimizing the die compensation shape using a genetic algorithm and variational autoencoder; calculating auxiliary stamping compensation data based on the die compensation shape, generating longitudinal beam processing data, and using the longitudinal beam processing data as the actual stamping process parameters for the longitudinal beam. This method for predicting and compensating springback during stamping of extended automotive longitudinal beams can improve the accuracy of springback prediction and, based on the optimized die compensation shape and auxiliary stamping compensation data, improve the forming quality of the longitudinal beam.

Description

Technical Field

[0001] This invention relates to the field of automotive parts stamping technology, specifically to a method and system for predicting and compensating for springback during stamping of extended longitudinal beams for automobiles. Background Technology

[0002] The longitudinal beams of an automobile are key load-bearing components in the vehicle body structure, and their machining accuracy directly affects the safety and stability of the entire vehicle. Due to the elastic recovery effect of materials during the stamping process, the geometry of the longitudinal beam parts often experiences springback errors after stamping, causing the actual shape to deviate from the design requirements.

[0003] Currently, stamping process designers typically attempt to minimize springback during the process and die design stages by optimizing stamping process parameters and adjusting die structure. However, due to the complex geometry and diverse material properties of longitudinal beam products, and the multiple processes involved in their forming, stamping springback is difficult to completely eliminate through simple design adjustments.

[0004] Therefore, designers often need to rely on experience to optimize the shape of stamping dies through repeated experiments, mold repairs, and trial molding to keep the product springback error within a reasonable range. However, this experiment-based method is not only time-consuming and costly, but may also lead to unstable compensation effects due to human experience bias.

[0005] Therefore, there is an urgent need for a method for predicting and compensating stamping springback that can optimize process parameters and stamping die shape in the early stages of production, reduce the number of trial die repairs, and lower manufacturing costs. Summary of the Invention

[0006] Purpose of the invention: In order to overcome the above shortcomings, the purpose of this invention is to provide a method and system for predicting and compensating springback during stamping of extended longitudinal beams for automobiles. By combining incremental plasticity theory, gradient boosting decision tree, Bayesian optimization, genetic algorithm and variational autoencoder, the accuracy of springback prediction can be improved and the forming quality of the longitudinal beam can be improved based on optimized die compensation shape and auxiliary stamping compensation data.

[0007] To address the aforementioned technical problems, this invention provides a method for predicting and compensating for springback during stamping of extended longitudinal beams in automobiles, comprising:

[0008] Step S1: Based on the influence of preset process parameters on stamping springback, construct an initial dataset containing initial predicted values;

[0009] Step S2: Construct a springback prediction model using incremental plasticity theory, then train the springback prediction model using gradient boosting decision tree and adjust the process parameters using Bayesian optimization method to generate the optimal process parameters;

[0010] Step S3: Based on the optimal process parameters, use finite element simulation to simulate the manufacturing of the longitudinal beam, generate simulated stamping springback data, and optimize the mold compensation shape using a genetic algorithm and variational autoencoder based on the simulated stamping springback data.

[0011] Step S4: Calculate the longitudinal beam compensation data for auxiliary stamping based on the mold compensation shape, generate longitudinal beam compensation data, and use the longitudinal beam processing data as the actual stamping process parameters of the longitudinal beam.

[0012] In a preferred embodiment of the present invention, step S1 includes:

[0013] Step S11: Select the material model and define the elastic modulus, Poisson's ratio, and yield criterion;

[0014] Step S12: Mesh the material using four-node shell elements or eight-node solid elements;

[0015] Step S13: Set preset process parameters;

[0016] Step S14: Use finite element simulation to simulate the longitudinal beam stamping process, including: applying stamping load to simulate the stress and deformation of the plate, calculating the stress field and strain field after deformation; applying boundary conditions to remove the stamping load, calculating the residual stress and the maximum springback amount and springback angle after deformation;

[0017] Step S15: Extract the maximum rebound amount, residual stress distribution, and rebound angle change under the preset process parameters to generate initial predicted values.

[0018] In a preferred embodiment of the present invention, step S2 includes:

[0019] A springback prediction model is constructed using incremental plasticity theory:

[0020]

[0021] Where Δε is the predicted springback error, E is the material's elastic modulus, and σ res σ represents the residual stress after stamping. y This represents the yield stress of the material.

[0022] In a preferred embodiment of the present invention, step S2 further includes:

[0023] Step S21: Using preset process parameters as input features, train the rebound prediction model using a gradient boosting decision tree:

[0024]

[0025] Among them, F t(x) represents the decision tree in the t-th round, T is the number of decision trees, and R is the number of decision trees. ptredicted This represents the rebound error predicted by the model.

[0026] Step S22: Define the Bayesian optimization objective function:

[0027] f(x) = min|R real -R ptredicted (x)|

[0028] Among them, R real The rebound error is measured experimentally, and x is a preset process parameter;

[0029] Step S23: Optimize the preset process parameters using the Gaussian process regression method:

[0030] p(y|x)=N(μ(x),k(x,x'))

[0031] Where p(y|x) is the probability distribution of the rebound error y given the parameter x, μ(x) is the mean function, representing the estimate of the rebound error, and k(x,x') is the covariance function, representing the correlation between the parameters;

[0032] Step S24: Use five-fold cross-validation to evaluate the generalization ability of the rebound prediction model;

[0033] Step S25: Iterate through steps S21 to S24 for a preset number of times until the preset requirements are met and the optimal process parameters are output.

[0034] In a preferred embodiment of the present invention, step S21 includes:

[0035] Step S210: Initialize the predicted rebound error value:

[0036]

[0037] Where F0(x) is the mean of the rebound error, N is the number of training samples, and R0 is the mean of the rebound error. real,i Let be the rebound error of the i-th sample;

[0038] Step S211: Calculate the deviation between the rebound error and the predicted value:

[0039]

[0040] in, F represents the bias of the i-th sample in the t-th iteration. t-1 (x i ) represents the prediction bounce error for the i-th sample in the (t-1)-th iteration;

[0041] Step S212: Deviation between the fitting rebound error and the predicted value:

[0042]

[0043] Among them, F t (x)' is the decision tree trained in the t-th round, used to fit the residuals, F(x) i Let be the predicted value of the trained decision tree model for the i-th sample, and argmin be the optimization objective that minimizes the sum of squared biases.

[0044] Step S213: Update the predicted values ​​and control the step size with the learning rate η:

[0045] F t (x)=F t-1 (x)+ηF t (x)

[0046] Among them, F t (x) represents the model's predicted value after the t-th iteration, F t-1 (x) is the predicted value after the (t-1)th iteration;

[0047] Step S214: Repeat steps S211 to S213 a preset number of times.

[0048] In a preferred embodiment of the present invention, step S3 includes:

[0049] Step S31: Based on the optimal process parameters, the selected material model, the defined elastic modulus, Poisson's ratio and yield criterion, use finite element simulation to simulate the manufacturing of the longitudinal beam and output simulated stamping springback data including the maximum springback, residual stress distribution, springback angle change and stress distribution in the thickness direction.

[0050] Step S32: Use a variational autoencoder to learn the relationship between the mold shape and the stamping springback data, and generate a parameterized initial compensation shape;

[0051] Step S33: From the initial compensation shape generated by the variational autoencoder, sample the initial mold shape and use a genetic algorithm to optimize the initial mold shape to generate the mold compensation shape;

[0052] Step S34: Use the mold compensation shape to perform finite element simulation again to determine whether the springback error has decreased. If yes, the optimization is considered complete. If not, return to step S33 and use a genetic algorithm to optimize the mold compensation shape again.

[0053] In a preferred embodiment of the present invention, step S33 includes:

[0054] Step S331: Sample the initial mold shape from the initial compensation shape generated by the variational autoencoder;

[0055] Step S332: Calculate the fitness function;

[0056] Step S333: Select individuals with high fitness using a preset selection method to enter the next generation;

[0057] Step S334: Perform crossover and mutation processing;

[0058] Step S335: Repeat steps S332 to S334 until the preset number of iterations or convergence criteria are reached.

[0059] In a preferred embodiment of the present invention, step S4 includes:

[0060] Step S41: Calculate local compensation offset based on the optimized mold compensation shape:

[0061] Δh=h real -h compensated

[0062] Among them, h real h represents the uncompensated shape of the die after stamping. compensated To compensate for the ideal shape of the mold;

[0063] Step S42: Calculate the local stress distribution:

[0064] σ res (x,y)=σ plastic (x,y)-σ elastic (x,y)

[0065] Where, σ plastic (x,y) represents the plastic stress, σ elastic (x,y) represents elastic stress;

[0066] Step S43: Set the rebound critical threshold and select the area with the first preset condition as the local heating area:

[0067] A critical ={(x,y)|σ res (x,y)>σ crit}

[0068] Where, σ crit The critical stress for springback;

[0069] Step S44: Calculate the temperature field of the locally heated region using finite element heat conduction analysis.

[0070]

[0071] Where ρ is the material density, c p ρ is the specific heat capacity, k is the thermal conductivity, and Q is the heating source term;

[0072] Step S45: Calculate the local optimal heating temperature:

[0073]

[0074] Where σ0 is the room temperature yield stress, T0 is the initial temperature, and T m Where m is the melting point temperature, and m is a material parameter.

[0075] Step S46: Calculate the local optimal heating time:

[0076]

[0077] Where ΔT is the desired temperature rise and Q is the heating power.

[0078] In a preferred embodiment of the present invention, step S43 further includes:

[0079] Step S431: Select the region with the second preset condition as the ultrasonic vibration action area:

[0080] A UV ={(x,y)|σ res (x,y)>σ crit}

[0081] Step S432: Calculate the flow stress reduction caused by ultrasonic vibration:

[0082] σ eff =σ flow -αf US

[0083] Where, σ flow The material resistance without ultrasonic vibration, α is the ultrasonic softening coefficient, and f US For ultrasonic frequency; Step S433: Calculate the optimal ultrasonic parameters:

[0084] A US =k·σ max

[0085] Where k is an empirical coefficient, σ max This represents the local maximum offset.

[0086] This application also provides a system for predicting and compensating for springback during stamping of extended longitudinal beams in automobiles using the aforementioned method, comprising:

[0087] The data construction module is used to calculate the impact of preset process parameters on stamping springback and construct an initial dataset containing the initial prediction results.

[0088] The model optimization module is used to construct a springback prediction model using incremental plasticity theory, then use Bayesian optimization method to adjust process parameters and use gradient boosting decision tree method to train the springback prediction model to minimize springback error.

[0089] The simulation compensation module is used to simulate the manufacture of longitudinal beams through preset process parameters using finite element simulation, generate simulated stamping springback data, and optimize the mold compensation shape based on the simulated stamping springback data using a genetic algorithm and a variational autoencoder.

[0090] The real feedback module is used to calculate the longitudinal beam compensation data for auxiliary stamping based on the mold compensation shape, generate longitudinal beam compensation data, and use the longitudinal beam processing data as the real stamping process parameters of the longitudinal beam.

[0091] In some embodiments, this application also relates to a computer medium storing a computer program, which is executed by a processor to implement the method for predicting and compensating springback during stamping of extended longitudinal beams for automobiles.

[0092] In some embodiments, this application also relates to a computer, including the aforementioned computer medium.

[0093] The technical solution described in this application has the following advantages over the prior art:

[0094] 1. This application uses incremental plasticity theory to calculate springback error and combines it with finite element simulation to establish a high-precision springback prediction model, which effectively avoids the problem of large predicted springback error. At the same time, gradient boosting decision tree is used to model springback error, and Bayesian optimization method is used to adjust process parameters to improve the adaptability and generalization ability of springback prediction model. Finally, five-fold cross-validation is used to evaluate springback prediction model to ensure that springback error can be accurately predicted under different materials and different stamping process parameters.

[0095] 2. This application uses a variational autoencoder to learn the mapping relationship between the mold shape and the stamping springback data, and generates a parameterized initial compensation shape to improve the efficiency of compensation shape optimization. Then, a genetic algorithm is used to further optimize the mold compensation shape, avoid local optima, improve the accuracy of the mold compensation shape, and then verify the effectiveness of the mold compensation shape through finite element simulation to ensure that the shape of the longitudinal beam after compensation meets the design requirements.

[0096] 3. Based on the optimized mold compensation shape, calculate the local compensation offset, accurately adjust the mold shape, improve processing accuracy, and then use a combination of local heating and ultrasonic assisted stamping to calculate and adjust the assisted stamping compensation data, optimize the stamping springback control strategy, and ensure that the error between the dimensions of the longitudinal beam after forming and the target dimensions does not exceed 1%. Attached Figure Description

[0097] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0098] Figure 1 This is a flowchart of the method for predicting and compensating for springback during stamping of extended longitudinal beams in automobiles, provided in an embodiment of the present invention.

[0099] Figure 2 This is a flowchart of the method for constructing an initial dataset provided in an embodiment of the present invention.

[0100] Figure 3 This is a flowchart of the method for generating optimal process parameters provided in an embodiment of the present invention.

[0101] Figure 4 This is a flowchart of the method for optimizing mold compensation shape provided in an embodiment of the present invention.

[0102] Figure 5 This is a flowchart of the local heating-assisted stamping method provided in an embodiment of the present invention.

[0103] Figure 6 This is a flowchart of the ultrasonic vibration-assisted stamping method provided in an embodiment of the present invention.

[0104] Figure 7 This is a module connection diagram of the automotive extended longitudinal beam stamping springback prediction and compensation system provided in an embodiment of the present invention.

[0105] Explanation of reference numerals in the instruction manual:

[0106] 100. Data Construction Module; 101. Model Optimization Module; 102. Simulation Compensation Module; 103. Realistic Feedback Module. Detailed Implementation

[0107] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0108] refer to Figure 1 As shown, in some embodiments, a method for predicting and compensating for springback during stamping of extended longitudinal beams in automobiles includes the following steps:

[0109] Step S1: Based on the influence of preset process parameters on stamping springback, construct an initial dataset containing initial predicted values.

[0110] Specifically, refer to Figure 2 As shown, in step S1, the method includes:

[0111] Step S11: Select the material model and define the elastic modulus, Poisson's ratio, and yield criterion.

[0112] The material models include, but are not limited to, bilinear hardening models, multilinear isotropic hardening models, and Johnson-Cook Strength constitutive models of plastic materials. The yield criteria include, but are not limited to, Von Mises and Hill 48. Specifically, the bilinear hardening model is generally applicable to low-carbon steel, using yield stress and hardening modulus to simulate the elastoplastic behavior of the material. The linear isotropic hardening model is generally applicable to high-strength steel, allowing for multi-segment nonlinear stress-strain relationships. The Johnson-Cook Strength constitutive model of plastic materials is generally applicable to heat-sensitive materials under high strain rate conditions, such as aluminum alloys and stainless steel. Specifically, the Von Mises yield criterion is generally applicable to isotropic materials, and the Hill 48 yield criterion is generally applicable to anisotropic materials.

[0113] For example, if this application uses DP980 high-strength steel, then a multilinear isotropic hardening model and the VonMises yield criterion are selected.

[0114] Step S12: Mesh the material using four-node shell elements or eight-node solid elements.

[0115] The four-node shell unit is typically suitable for thin plate stamping, while the eight-node solid unit is typically suitable for complex deformation regions; the specific mesh size is set according to the size of the longitudinal beam workpiece.

[0116] Step S13: Set preset process parameters, which include, but are not limited to, stamping pressure, stamping time, stamping temperature and other parameters related to the stamping process.

[0117] Step S14: Use finite element simulation to simulate the longitudinal beam stamping process, including: applying stamping load to simulate the stress and deformation of the plate, calculating the stress field and strain field after deformation; applying boundary conditions to remove the stamping load, calculating the residual stress and the maximum springback amount and springback angle after deformation.

[0118] The simulation process is as follows:

[0119] Apply a stamping load, calculate the stress and strain field, and obtain the stress distribution of the workpiece under maximum deformation.

[0120] Apply boundary conditions, release the stamping load, and calculate the residual stress and the maximum springback after deformation;

[0121] Calculate the springback angle from the residual bending moment.

[0122] The maximum springback amount is obtained by subtracting the target workpiece length from the unloaded workpiece length; the springback angle is:

[0123]

[0124] Among them, M res denoted as residual bending moment, L as plate length, E as elastic modulus, h as plate thickness, and b as plate width.

[0125] Step S15: Extract the maximum rebound amount, residual stress distribution, and rebound angle change under the preset process parameters to generate initial predicted values.

[0126] Step S2: Construct a springback prediction model using incremental plasticity theory, then train the springback prediction model using gradient boosting decision tree, and adjust the process parameters using Bayesian optimization method to generate the optimal process parameters.

[0127] Specifically, a brief introduction to the incremental plasticity theory is provided below:

[0128] During the stamping process, the material undergoes a loading-unloading process:

[0129] During the loading stage, the impact force causes plastic deformation, and the stress state changes.

[0130] During the unloading phase, the sheet material springs back, and some of the plastic strain cannot be recovered, resulting in residual stress and shape errors.

[0131] Therefore, the fundamental equation of incremental plasticity theory is:

[0132] dε=dε e +dε p

[0133] Where dε is the total strain increment, dε e dε is the elastic strain increment. p This represents the plastic strain increment.

[0134] During the unloading phase, the material mainly exhibits elastic recovery, and its plastic strain no longer increases. Therefore, the incremental plasticity theory is used to construct a springback prediction model:

[0135]

[0136] Where Δε is the springback error, E is the material's elastic modulus, and σ res σ represents the residual stress after stamping.y This represents the yield stress of the material.

[0137] Specifically, refer to Figure 3 As shown, in step S2, the method further includes:

[0138] Step S21: Using preset process parameters as input features, train the rebound prediction model using a gradient boosting decision tree:

[0139]

[0140] Among them, F t (x) represents the decision tree in the t-th round, T is the number of decision trees, and R is the number of decision trees. ptredicted This represents the rebound error predicted by the model.

[0141] Step S22: Define the Bayesian optimization objective function:

[0142] f(x) = min|R real -R ptredicted (x)|

[0143] Among them, R real The rebound error is measured experimentally, and x is a preset process parameter.

[0144] The optimization objective is to adjust the preset process parameters so that the predicted springback error is close to the actual error.

[0145] Step S23: Optimize the preset process parameters using the Gaussian process regression method:

[0146] p(y|x)=N(μ(x),k(x,x'))

[0147] Where p(y|x) is the probability distribution of the rebound error y given the parameter x, μ(x) is the mean function, representing the estimate of the rebound error, and k(x,x') is the covariance function, representing the correlation between the parameters.

[0148] The covariance function (kernel function) mentioned above references the radial basis function kernel function:

[0149]

[0150] Where, σ 2 Let be the kernel function variance, to control the prediction range; l is the length scale, to control the similarity between different inputs; thus, the optimization process of the Gaussian process regression method is: predict μ(x) and variance k(x,x') based on the current rebound error, and then select the optimal parameter x to minimize the objective function.

[0151] Step S24: Use five-fold cross-validation to evaluate the generalization ability of the rebound prediction model.

[0152] Five-fold cross-validation involves dividing the dataset into five parts, using four parts for training and one part for testing each time, repeating this process five times. Then, based on the mean squared error and the coefficient of determination, the model's generalization ability is ensured, and overfitting is prevented.

[0153] Step S25: Iterate through steps S21 to S24 for a preset number of times until the preset requirements are met and the optimal process parameters are output.

[0154] The preset number of times in this application is set by the designer according to actual needs. The preset requirement refers to the rebound error converging to a set threshold, such as |R real -R ptredicted |<0.05mm.

[0155] In some examples of this application, references Figure 3 As shown, in step S21, the method further includes:

[0156] Step S210: Initialize the predicted rebound error value:

[0157]

[0158] Where F0(x) is the mean of the rebound error, N is the number of training samples, and R0 is the mean of the rebound error. real,i Let be the rebound error of the i-th sample; since the gradient boosting decision tree improves the prediction accuracy by learning the residuals, the initial prediction value is the mean of the rebound errors of all samples as the benchmark.

[0159] Step S211: Calculate the deviation between the rebound error and the predicted value:

[0160]

[0161] in, F represents the bias of the i-th sample in the t-th iteration. t-1 (x i Let be the prediction bounce error for the i-th sample in the (t-1)-th iteration; this deviation is used to measure the error between the current model prediction and the true value.

[0162] Step S212: Deviation between the fitting rebound error and the predicted value:

[0163]

[0164] Among them, F t (x)' is the decision tree trained in the t-th round, used to fit the residuals, F(x) iLet be the predicted value of the trained decision tree model for the i-th sample, and argmin be the optimization objective that minimizes the sum of squared biases; that is, to train a new decision tree to predict biases in order to find the relationship between the input x and the residuals so that the decision tree can minimize the error.

[0165] Step S213: Update the predicted values ​​and control the step size with the learning rate η:

[0166] F t (x)=F t-1 (x)+ηF t (x)

[0167] Among them, F t (x) represents the model's predicted value after the t-th iteration, F t-1 (x) is the predicted value after the (t-1)th iteration; the learning rate η is usually greater than 0 and less than or equal to 1, controlling the contribution of the new decision tree to the final predicted value.

[0168] Step S214: Repeat steps S211 to S213 a preset number of times.

[0169] The preset number of times in this application is set by the designer according to actual needs.

[0170] Step S3: Based on the optimal process parameters, use finite element simulation to simulate the manufacturing of the longitudinal beam, generate simulated stamping springback data, and optimize the mold compensation shape using a genetic algorithm and variational autoencoder based on the simulated stamping springback data.

[0171] Specifically, refer to Figure 4 As shown, in step S3, the method includes:

[0172] Step S31: Based on the optimal process parameters, the selected material model, the defined elastic modulus, Poisson's ratio, and yield criterion, use finite element simulation to simulate the manufacturing of the longitudinal beam and output simulated stamping springback data including the maximum springback, residual stress distribution, springback angle variation, and stress distribution in the thickness direction.

[0173] Therefore, stamping simulation is performed based on the optimal process parameters to extract springback data after stamping.

[0174] Step S32: Use a variational autoencoder to learn the relationship between the mold shape and the stamping springback data, and generate a parameterized initial compensation shape.

[0175] Specifically, by learning the relationship between the die shape M and the stamping springback data, a possible compensation shape distribution is generated; the variational autoencoder structure includes an encoder and a decoder, wherein the encoder is:

[0176]

[0177] The input is the initial mold shape M, and the output is the mean μ of the latent variable z. M and variance The decoder is:

[0178]

[0179] Where the input is the latent variable z, and the output is the reconstructed mold shape.

[0180] Step S33: Based on the initial compensation shape generated by the variational autoencoder, the initial mold shape is sampled and optimized using a genetic algorithm to generate the mold compensation shape.

[0181] Step S34: Use the mold compensation shape to perform finite element simulation again to determine whether the springback error has decreased. If yes, the optimization is considered complete. If not, return to step S33 and use a genetic algorithm to optimize the mold compensation shape again.

[0182] In this process, the optimized die compensation shape is used to perform finite element simulation of stamping again to verify whether the springback error has been reduced. If |R| is satisfied... new -R real If |≤∈, then the optimization is considered successful; otherwise, continue optimization; where R new To compensate for the springback error of the optimized mold shape, ∈ represents the error tolerance.

[0183] In some examples of this application, references Figure 4 As shown, in step S33, the method further includes:

[0184] Step S331: Sample the initial mold shape M0 from the latent distribution p(M|z) generated by the variational autoencoder.

[0185] Step S332: Calculate the fitness function:

[0186] F fitness (M)=-|R real -R ptredicted (M)|

[0187] Step S333: Select individuals with high fitness using a preset selection method to enter the next generation.

[0188] The preset selection method includes, but is not limited to, at least one of the following methods in genetic algorithms: roulette wheel selection, tournament selection, Monte Carlo selection, probability selection, random traversal selection, Boltzmann selection, etc., in order to select individuals with high fitness to enter the next generation.

[0189] Step S334: Perform crossover:

[0190] M new =αM1+(1+α)M2

[0191] Where M1 and M2 are the two selected parent mold shapes, and α is the crossover rate;

[0192] Perform mutation processing:

[0193] M mut =M+βN(0,σ)

[0194] Where β is the mutation rate and N(0,σ) is a random disturbance that follows a normal distribution.

[0195] Step S335: Repeat steps S332 to S334 until the preset number of iterations or convergence criteria are reached.

[0196] The preset number of iterations or preset convergence criteria are set by the designer according to actual needs.

[0197] Step S4: Calculate the longitudinal beam compensation data for auxiliary stamping based on the mold compensation shape, generate longitudinal beam compensation data, and use the longitudinal beam processing data as the actual stamping process parameters of the longitudinal beam.

[0198] Specifically, refer to Figure 5 As shown, in step S4, the method includes:

[0199] Step S41: Calculate local compensation offset based on the optimized mold compensation shape:

[0200] Δh=h real -h compensated

[0201] Among them, h real h represents the uncompensated shape of the die after stamping. compensated To compensate for the ideal shape of the mold.

[0202] Step S42: Calculate the local stress distribution:

[0203] σ res (x,y)=σ plastic (x,y)-σ elastic (x,y)

[0204] Where, σ plastic (x,y) represents the plastic stress, σ elastic (x,y) represents elastic stress;

[0205] Step S43: Set the rebound critical threshold and select the area with the first preset condition as the local heating area:

[0206] A critical ={(x,y)|σ res (x,y)>σ crit}

[0207] Where, σ crit The critical stress for rebound; the region of the first preset condition is σ. res (x,y)>σ crit The area.

[0208] Step S44: Calculate the temperature field of the locally heated region using finite element heat conduction analysis.

[0209]

[0210] Where ρ is the material density, c p is the specific heat capacity, k is the thermal conductivity, and Q is the heating source term.

[0211] Step S45: Calculate the local optimal heating temperature:

[0212]

[0213] Where σ0 is the room temperature yield stress, T0 is the initial temperature, and T m denoted as melting point temperature, and m as a material parameter.

[0214] Step S46: Calculate the local optimal heating time:

[0215]

[0216] Where ΔT is the desired temperature rise and Q is the heating power.

[0217] Therefore, based on the local heating area, the temperature field of the local heating area, the local optimal heating temperature, and the local optimal heating time, longitudinal beam processing data is generated and used as the actual stamping process parameters for the longitudinal beam.

[0218] Therefore, in the above example, localized heating of key deformation areas (such as bending or protruding areas) improves material ductility and reduces springback. Specific heating methods can include resistance heating, induction heating, or laser heating, which locally heats the metal before or during stamping. A suitable temperature range is selected based on the actual longitudinal beam workpiece material; for example, DP980 and Q&P steel can be heated between 350°C and 600°C. This ensures that the longitudinal beam workpiece material is stamped in a softened state to achieve optimal plastic deformation. Furthermore, after stamping, the material can be rapidly cooled immediately using liquid nitrogen cooling spray, compressed air cooling, or spray cooling equipment to lock the deformation in the mold and reduce springback.

[0219] Based on the above-mentioned local heating-assisted stamping technology solution, an example is given:

[0220] Stress data at different locations after stamping were extracted using FEM simulation and experimental measurements. The stress data for a key bending region were then obtained as follows:

[0221] Position (x, y) <![CDATA[σ plastic (MPa)]]> <![CDATA[σ elastic (MPa)]]> <![CDATA[σ res (MPa)]]> (10,20) 450 200 250 (15,25) 420 190 230 (20,30) 380 200 180 (25,35) 300 150 150

[0222] Among them, through experimental and simulation analysis, σ is set crit =200MPa, therefore, σ is selected. res The region (x,y)>200MPa, i.e., A critical ={(10,20),(15,25)}, the residual stress at these two points is relatively high, so local heating is required to assist in stamping; since the existing parameters of DP980 high-strength steel are used, the heating temperature is obtained as 400℃ through thermo-mechanical coupled finite element analysis, and the heating time is obtained as 10 seconds.

[0223] In some embodiments of this application, electromagnetic assisted stamping can also be used, by applying a dynamic magnetic field or pulsed current to optimize metal flow characteristics and reduce stamping springback.

[0224] Specifically, the dynamic magnetic field is applied by arranging coils on a mold or stamping platform, applying an alternating magnetic field with a frequency of 10–100 kHz and a magnetic field strength of 0.5–2 T. This induces eddy currents inside the metal through the alternating magnetic field, causing local temperature increases, softening the material and improving its plastic deformation capacity. The magnetic field is also used to change the grain orientation and reduce springback error.

[0225] Specifically, the pulsed current is applied as follows: before or during stamping, a high-frequency pulsed current is applied to the material to reduce the yield strength of the material. The high-frequency pulsed current is set to 1-10kA. The current generates a Joule heating effect through the metal, causing the material temperature to rise locally, softening the metal, improving the plastic deformation capacity, and using the current to affect the movement of dislocations inside the metal, optimizing the stress distribution and reducing springback error.

[0226] By adopting the above technical solution, non-contact control can be used to avoid affecting the mold structure, and it is compatible with existing stamping equipment, thereby improving forming accuracy.

[0227] In some embodiments of this application, reference is made to Figure 6 As shown, in step S43, the method further includes:

[0228] Step S431: Select the region with the second preset condition as the ultrasonic vibration action area:

[0229] A UV ={(x,y)|σres (x,y)>σ crit}

[0230] The second preset condition is |σ res (x,y)|>σ crit The area.

[0231] Step S432: Calculate the flow stress reduction caused by ultrasonic vibration:

[0232] σ eff =σ flow -αf US

[0233] Where, σ flow The material resistance without ultrasonic vibration, α is the ultrasonic softening coefficient, and f US Ultrasonic frequency;

[0234] Step S433: Calculate the optimal ultrasound parameters:

[0235] A US =k·σ max

[0236] Where k is an empirical coefficient, σ max This represents the local maximum offset.

[0237] Therefore, applying high-frequency ultrasonic vibration to the surface of the mold or stamping sheet reduces deformation resistance and improves forming accuracy. The ultrasonic vibrator can be installed on the mold, punch, fixture or longitudinal beam workpiece sheet and the ultrasonic vibration frequency can be set in the range of 20 to 60 kHz and the amplitude in the range of 5 to 20 μm. The ultrasonic vibration can be used to destroy the sliding resistance inside the metal and reduce the friction between the metal and the mold, making the material easier to deform and reducing springback error.

[0238] Based on the above ultrasonic vibration-assisted stamping technology solution, an example is given:

[0239] Deformation error in some areas was obtained using FEM simulation and experimental measurements:

[0240] Position (x, y) <![CDATA[h real (mm)]]> <![CDATA[h compensated (mm)]]> Δh(mm) (10,20) 12.5 12.0 0.5 (15,25) 13.8 13.0 0.8 (20,30) 12.2 12.0 0.2

[0241] The error threshold is set to 0.4 mm, A UV ={(10,20),(15,25)}, meaning that ultrasonic vibration needs to be applied at positions (10,20) and (15,25); then the ultrasonic softening coefficient α = 0.2 MPa / kHz is calculated, using 20 kHz ultrasound: σ eff =σ flow -0.2×20, if the original flow stress σ of the material flow =400MPa, then σ eff=400-4=396MPa, the softening effect can reduce flow stress by 1% and reduce rebound; therefore, the ultrasonic amplitude A is calculated. US =0.2×0.8=0.16mm, that is, ultrasonic vibration with an amplitude of 0.16mm is applied at positions (10,20) and (15,25).

[0242] In this application, designers are also able to combine incremental plasticity theory with crystal plasticity, starting from the microscale, considering grain orientation, twin deformation and dislocation evolution, to construct a high-precision material constitutive model and combine it with finite element simulation to optimize springback prediction;

[0243] Specifically, at the microscale, a polycrystalline microstructure model can be defined using the three-dimensional crystal plasticity finite element method, assuming that each grain has a random orientation. Then, based on the orientation distribution function, the initial grain orientation is defined, and its Schmid factor is calculated to analyze the influence of different grain orientations on plastic deformation behavior. Subsequently, different stamping process parameters (stamping speed, temperature, stamping time, etc.) are applied to analyze the deformation, orientation evolution, and twinning behavior of each grain. Then, molecular dynamics methods are used to simulate the microscopic deformation mechanism of metals during the stamping process, and the dislocation motion, dislocation climb, and dislocation cross-slip processes of the material are studied under different strain rates, temperatures, and pressures. Furthermore, based on the evolution of dislocation density during stress relaxation, the Orowan equation is used to characterize microscopic plastic flow. In this way, data on the evolution of dislocation structure under different stamping process parameters can be obtained through simulation, providing microscopic mechanism support for constitutive equations. Therefore, based on the results obtained using the three-dimensional crystal plasticity finite element method and molecular dynamics method, the material hardening characteristic equation is established by combining the Taylor model and the Voce hardening model. Then, the stress-strain relationship of the material is determined by uniaxial tensile test, springback test and nanoindentation test. The parameters of the Taylor-Voce hardening model are calibrated by numerical optimization methods (such as least squares fitting or Bayesian optimization) to improve the prediction accuracy of the constitutive model. In addition, the micro-strain distribution is monitored by combining digital image correlation technology to verify the influence of grain orientation and twinning behavior on the rheological properties of the material, so as to predict the dynamic yielding behavior of metal sheets under various stamping process parameters.

[0244] Secondly, at the macroscopic scale, the constitutive equation of incremental plasticity theory is used to predict the springback behavior of stamping, and the springback prediction model is optimized through finite element simulation. Then, a variational autoencoder is used to learn the mapping relationship between the die shape and the stamping springback data to generate a parameterized compensation shape. With the minimum springback error as the fitness target, a genetic algorithm is used to optimize the compensation shape to ensure that the final shape can effectively reduce the springback error. Then, finite element simulation is used to verify the optimized die compensation shape and compare the springback error of the longitudinal beam before and after compensation. If the accuracy requirements are still not met, the optimization process is returned to continue adjusting the parameters until the accuracy requirements are met. Finally, the generated die compensation shape and the longitudinal beam compensation data for auxiliary stamping are used as the actual stamping process parameters of the longitudinal beam.

[0245] Specifically, the above micro-modeling process is based on Schmid's law:

[0246]

[0247] Where, τ ω For the shear stress ω on the slip system, σ ij For stress tensor, Let be the direction cosine of the slip system; its slip rate is:

[0248]

[0249] in, Let ω be the shear strain rate on the slip system. For reference strain rate, Here, is the critical shear stress, and n is the strain rate sensitivity index. This describes the activation of different grain slip systems in sheet metal during the stamping process, thus affecting the overall plastic deformation characteristics.

[0250] In some embodiments of this application, when the stamping equipment performs stamping processing based on the longitudinal beam compensation data, laser measurement or three-dimensional vision inspection technology can also be used to monitor the springback of the longitudinal beam workpiece after stamping in real time, thereby dynamically adjusting the stamping process parameters (such as pressure, speed, local heating temperature, local heating time, local heating position, ultrasonic assisted stamping power, ultrasonic assisted stamping time, ultrasonic assisted stamping position, etc.) for feedback compensation; or real-time stamping prediction and compensation can be performed through 3D scanning and digital twin technology.

[0251] refer to Figure 7 As shown, in some embodiments, the involved automotive extended longitudinal beam stamping springback prediction and compensation system includes:

[0252] The data construction module 100 is used to calculate the influence of preset process parameters on stamping springback and construct an initial dataset containing the initial prediction results.

[0253] The model optimization module 101 is used to construct a springback prediction model using incremental plasticity theory, then use Bayesian optimization method to adjust process parameters and use gradient boosting decision tree method to train the springback prediction model to minimize springback error.

[0254] The simulation compensation module 102 is used to simulate the manufacture of longitudinal beams by preset process parameters using finite element simulation, generate simulated stamping springback data, and optimize the mold compensation shape by using genetic algorithm and variational autoencoder based on the simulated stamping springback data.

[0255] The real feedback module 103 is used to calculate the longitudinal beam compensation data for auxiliary stamping based on the mold compensation shape, generate longitudinal beam compensation data, and use the longitudinal beam processing data as the real stamping process parameters of the longitudinal beam.

[0256] In some embodiments, this application also relates to a computer medium storing a computer program, which is executed by a processor to implement the method for predicting and compensating springback during stamping of extended longitudinal beams for automobiles.

[0257] In some embodiments, this application also relates to a computer, including the aforementioned computer medium.

[0258] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0259] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting and compensating springback in stamping of an automobile elongated longitudinal beam, characterized by, The method comprises the following steps: Step S1: constructing an initial dataset containing initial prediction values according to the influence of preset process parameters on stamping springback; Step S2: constructing a springback prediction model using incremental plasticity theory, and then training the springback prediction model using gradient boosting decision trees and adjusting process parameters using a Bayesian optimization method to generate optimal process parameters; Step S3: simulating the manufacture of a longitudinal beam according to the optimal process parameters using finite element simulation, generating simulation stamping springback data, and optimizing the die compensation shape using a genetic algorithm and a variational autoencoder according to the simulation stamping springback data; In step S3, the method comprises: Step S31: simulating the manufacture of a longitudinal beam using finite element simulation according to the optimal process parameters, a selected material model, defined elastic modulus, Poisson's ratio, and yield criterion, and outputting simulation stamping springback data containing maximum springback, residual stress distribution, springback angle change, and stress distribution in the plate thickness direction; Step S32: learning the relationship between the die shape and the stamping springback data using a variational autoencoder to generate a parameterized initial compensation shape; Step S33: sampling the initial die shape from the initial compensation shape generated by the variational autoencoder and optimizing the initial die shape using a genetic algorithm to generate a die compensation shape; Step S34: using the die compensation shape to perform finite element simulation again to determine whether the springback error is reduced, and if so, determining that the optimization is complete, and if not, returning to step S33 to optimize the die compensation shape again using a genetic algorithm; Step S4: calculating longitudinal beam compensation data for auxiliary stamping according to the die compensation shape, generating longitudinal beam compensation data, and using the longitudinal beam processing data as the real stamping process parameters for the longitudinal beam.

2. The method for predicting and compensating springback of a stamping of a lengthened longitudinal beam of an automobile according to claim 1, characterized in that, In step S1, the method comprises: Step S11: selecting a material model and defining elastic modulus, Poisson's ratio, and yield criterion; Step S12: performing mesh division using four-node shell elements or eight-node solid elements; Step S13: setting preset process parameters; Step S14: simulating the stamping process of a longitudinal beam using finite element simulation, including: applying a stamping load to simulate the stress deformation of a plate, calculating the stress field and strain field after deformation; applying boundary conditions to remove the stamping load, and calculating the residual stress and maximum springback amount and springback angle after deformation; Step S15: extracting the maximum springback amount, residual stress distribution, and springback angle change under the preset process parameters to generate initial prediction values.

3. The method according to claim 1 or 2, wherein, In step S2, the method comprises: Constructing a springback prediction model using incremental plasticity theory: where Δε is the predicted springback error, E is the material elastic modulus, is the residual stress after stamping, is the material yield stress.

4. The method of claim 3, wherein the method further comprises: In step S2, the method further comprises: Step S21: using gradient boosting decision trees to train the springback prediction model with preset process parameters as input features: wherein, is the tth decision tree, T is the number of decision trees, is the model predicted springback error; Step S22: defining a Bayesian optimization objective function: wherein, is the experimentally measured springback error, and x is the preset process parameter. Step S23: optimizing the preset process parameters using a Gaussian process regression method: wherein, is the probability distribution of the springback error y for a given parameter x, is the mean function representing the estimate of the springback error, is the covariance function representing the correlation between parameters; Step S24: evaluating the generalization ability of the springback prediction model using five-fold cross-validation; Step S25: iterating steps S21 to S24 for a preset number of times until the preset requirements are met and outputting the optimal process parameters.

5. The method for predicting and compensating springback of stamping of a lengthened longitudinal beam of an automobile according to claim 1 or 4, characterized in that, In step S21, the method comprises: Step S210: initialize the springback error prediction value: wherein, is the mean of the springback error, N is the number of training samples, is the springback error of the i-th sample; Step S211: calculate the deviation of the springback error and the prediction value: wherein, is the bias for the i-th sample in the t-th iteration, is the predicted rebound error for the i-th sample at the t-1-th iteration; Step S212: fit the deviation of the springback error and the prediction value: wherein, is the decision tree for the t-th round of training for fitting the residual, is the predicted value of the i-th sample by the trained decision tree model, and argmin is the optimization objective of minimizing the sum of squared deviations. Step S213: update the prediction value and control the step size with the learning rate η: wherein, is the predicted value of the model after the tth iteration, is the predicted value after the t-1th iteration; Step S214: loop steps S211 to S213 for a preset number of times.

6. The method of predicting and compensating springback of a stamping of a lengthened side rail of an automobile according to claim 1, wherein In step S33, the method comprises: Step S331: sampling the initial die compensation shape from the initial compensation shape generated by the variational autoencoder to obtain the initial die shape; Step S332: calculate the fitness function; Step S333: select individuals with high fitness into the next generation using a preset selection method; Step S334: perform crossover and mutation processing; Step S335: repeat steps S332 to S334 until a preset number of iterations or a convergence criterion is reached.

7. The method for predicting and compensating springback of stamping of a lengthened longitudinal beam of an automobile according to claim 1 or 6, characterized in that, In step S4, the method comprises: Step S41: calculate the local compensation offset based on the optimized die compensation shape: wherein, is the shape of the die after the stamping without compensation, is the ideal shape of the die after compensation; Step S42: calculate the local stress distribution: wherein, is the plastic stress, is the elastic stress; Step S43: set the springback critical threshold and select the region of the first preset condition as the local heating region: wherein is the critical stress for springback; Step S44: use finite element heat conduction analysis to calculate the temperature field of the local heating region: where p is the material density, Cp is the specific heat capacity, k is the thermal conductivity, and Q is the heat source term. Step S45: calculate the local optimal heating temperature: wherein, is the room temperature yield stress, is the initial temperature, is the melting temperature, m is a material parameter; Step S46: calculate the local optimal heating time: where ΔT is the desired temperature rise and Q is the heating power.

8. The method of predicting and compensating springback of a stamping of a lengthened side rail of an automobile according to claim 7, wherein In step S43, the method further comprises: Step S431: select the region of the second preset condition as the ultrasonic vibration acting region: Step S432: calculate the reduced flow stress under ultrasonic vibration: wherein is the material resistance without the action of ultrasonic vibrations, and a is the ultrasonic softening coefficient, is the ultrasonic frequency; Step S433: calculate the optimal ultrasonic parameters: where k is an empirical coefficient, is the local maximum offset.

9. A system for springback prediction and compensation of a stamping of an automobile longeron using the method of any one of claims 1-8, characterized in that, Comprising: A data construction module for calculating the influence of preset process parameters on stamping springback, and constructing an initial data set containing initial prediction results; A model optimization module for constructing a springback prediction model using incremental plasticity theory, and then adjusting process parameters using a Bayesian optimization method and training the springback prediction model using a gradient boosting decision tree method to minimize springback error; A simulation compensation module for simulating the manufacture of a longitudinal beam through preset process parameters using finite element simulation, generating simulation stamping springback data, and optimizing the die compensation shape using a genetic algorithm and a variational autoencoder based on the simulation stamping springback data; A real feedback module for calculating longitudinal beam compensation data for auxiliary stamping based on the die compensation shape, generating longitudinal beam compensation data, and using the longitudinal beam processing data as real stamping process parameters for the longitudinal beam.

Images