Method for optimizing stamping process parameters of thin-walled battery shell based on lnn prediction

Abstract

The present application relates to the technical field of sheet metal stamping forming and process optimization, and provides a method for optimizing stamping process parameters of a thin-walled battery shell based on LNN prediction, which comprises the following steps: establishing a new energy thin-walled battery shell geometric model, selecting three process parameters with the greatest impact on forming quality as optimization variables, and selecting an index for evaluating forming quality; generating a plurality of experimental combinations in finite element simulation calculation by adopting BBD design test scheme, extracting the maximum thinning rate and the maximum thickening rate corresponding to each test to form an initial sample data set; constructing and training an improved LNN prediction model; taking the trained LNN prediction model as a fitness evaluation function of MOPSO algorithm, predicting a Pareto optimal solution set, and selecting a unique optimal solution. The present application can capture the complex coupling relationship and nonlinear effect among multiple process parameters, and effectively realize the collaborative optimization of contradictory quality objectives.

Description

Technical Field

[0001] This invention relates to the field of metal sheet stamping and process optimization technology, and in particular to a method for optimizing the stamping process parameters of thin-walled battery casings based on LNN prediction. Background Technology

[0002] The external protective shell of the battery pack for new energy vehicles is a key component, and its manufacturing process faces numerous challenges. These shells are typically made into deep-cavity, thin-walled rectangular boxes, formed from 3003-H14 aluminum alloy through multiple processes including stamping, deep drawing, thinning, and trimming. They are characterized by thin material, numerous forming processes, and high precision requirements. Because the deformation of the metal sheet during stamping is difficult to control, and influenced by factors such as part shape, process design, and stamping dies, the shell is highly susceptible to forming quality defects such as wrinkles, cracks, scratches, and uneven wall thickness. This directly affects the safety and reliability of the battery pack.

[0003] In the deep drawing process of shells, the selection of process parameters is crucial, but existing optimization methods often rely on experience or a limited number of numerical simulations for trial and error. This approach struggles to systematically handle the complex coupling and nonlinear relationships between multiple process parameters, resulting in inefficient optimization processes and unreliable results. Traditional process parameter optimization methods lack scientific and systematic consideration and cannot adequately address the dynamic changes in highly nonlinear problems.

[0004] Specifically, traditional response surface methodology (RSM) has limited accuracy in approximating highly nonlinear problems, making it difficult to accurately capture the subtle interactions between process parameters. This limits its application in optimizing complex stamping processes, causing parameter adjustments to often remain at a local level and fail to obtain a globally optimal solution. Furthermore, when using neural networks as surrogate models, the generation of initial samples often lacks proper design, resulting in an unstable foundation for model training.

[0005] Furthermore, neural network models often fail to adequately address the dynamic response characteristics of time-varying or parameter perturbations, and the interaction effects between parameters are frequently overlooked. This can lead to optimized parameter combinations that are only locally optimal, resulting in poor adaptability in real-world production environments. When faced with dynamic factors such as material fluctuations and mold wear, this limitation can easily lead to batch quality instability and increase production risks.

[0006] Overall, existing technologies have systematic methodological defects in optimizing the stamping process of battery pack housings for new energy vehicles. They are unable to balance the complex relationships between multiple process parameters and cannot effectively cope with dynamic disturbances in production, thus restricting the stable improvement of housing forming quality. Summary of the Invention

[0007] To address the above issues, this invention integrates experimental design, liquid neural networks, and multi-objective optimization algorithms to construct a high-precision, highly generalizable process parameter-quality index proxy model. This model systematically captures the complex coupling relationships and nonlinear effects among multiple process parameters, effectively achieving the synergistic optimization of conflicting quality objectives.

[0008] According to embodiments of the present invention, a method for optimizing stamping process parameters of thin-walled battery casings based on LNN prediction is provided.

[0009] In a first aspect of the invention, a method for optimizing stamping process parameters of thin-walled battery casings based on LNN prediction is provided. The method includes:

[0010] Step S01: Use 3D modeling software to establish a geometric model of the thin-walled battery shell for new energy, select the three process parameters that have the greatest impact on the forming quality as optimization variables, and select the maximum forming thinning rate and the maximum forming thickening rate of the shell as indicators for evaluating the forming quality.

[0011] Step S02: Using BBD to design an experimental scheme, generate several sets of experimental combinations and perform simulation calculations in the finite element method. Extract the maximum thinning rate and maximum thickening rate corresponding to each set of experiments to form an initial sample dataset, which is then normalized and divided into a training set and a test set.

[0012] Step S03: Constructing an improved LNN prediction model: Activation state of the liquid hidden layer It is governed by a linear first-order ordinary differential equation; and the LNN prediction model is trained using a training set and the prediction accuracy of the LNN prediction model is verified using a test set;

[0013] Step S04: Use the trained LNN prediction model as the fitness evaluation function of the MOPSO algorithm to predict the Pareto optimal solution set and select the unique optimal solution from it as the preferred combination of process parameters.

[0014] Furthermore, the three process parameters mentioned in step S01 are: blank holder force, friction coefficient, and die clearance.

[0015] Furthermore, the maximum thinning rate mentioned in step S01 Calculation formula:

[0016] ,

[0017] Maximum thickness increase Calculation formula:

[0018] ,

[0019] in, The initial thickness of the sheet metal. This is the minimum thickness after forming. This represents the maximum thickness after forming.

[0020] Furthermore, the input layer of the improved LNN described in step S03 contains 4 neurons. These correspond to the normalized process parameters: stamping speed, blank holder force, friction coefficient, and die clearance.

[0021] Furthermore, the output layer of the improved LNN described in step S03 contains two neurons, and the liquid layer is mapped using a linear mapping. The steady-state output at time step 1 is converted into a prediction quality metric:

[0022] ,

[0023] in, The weight matrix of the output layer. This is the bias vector for the output layer.

[0024] Furthermore, the liquid hidden layer state vector described in step S03 The specific steps to achieve physical feature inheritance between processes are as follows:

[0025] The residual stress tensor of the sheet metal extracted from the first-order deep drawing simulation Equivalent plastic strain The hardening parameters are defined as the physical legacy features from the previous pass, and are mapped to the initial state of the liquid hidden layer using a feature extraction operator. ,in Represents the current process sequence number;

[0026] Establish inter-process state transfer function , set the first End of the course The hidden layer state is Then the first Initial state of the path Defined as:

[0027] ,

[0028] in, These are the station transfer disturbance parameters between each pass;

[0029] No. The improved LNN model for each pass inherits the initial state and incorporates the process parameters of the current process. The state vector is driven by the controlled ODE equation. Towards evolution.

[0030] Furthermore, the specific steps of step S04 are as follows:

[0031] Set particle swarm size Maximum number of iterations Inertial weight Learning factor , The position of each particle represents a combination of process parameters. ;

[0032] The trained improved LNN prediction model is used as the fitness evaluation function of MOPSO, and the optimization objective is to simultaneously minimize the maximum thinning rate and the maximum thickening rate:

[0033] ,

[0034] The constraints are: if the thinning rate is less than the limit value, the shell forming will not crack; if the thickness increase rate is less than the limit value, the shell will not wrinkle.

[0035] Update the particle's velocity and position, call the LNN prediction model to predict and calculate the fitness value for each new position, update the individual optimal solution and the global optimal solution, and maintain an external archive to store the Pareto optimal solution set;

[0036] After the iteration, the Pareto front is obtained due to the conflict between the maximum thinning rate and the maximum thickening rate. Based on the quality assessment that the impact of thin-walled battery shell cracking is greater than that of wrinkling in actual production, a unique optimal solution is selected from the Pareto solution set as the preferred combination of process parameters using fuzzy membership function or weighted method.

[0037] In a second aspect of the invention, an apparatus for optimizing stamping process parameters of thin-walled battery casings based on LNN prediction is provided. The apparatus includes:

[0038] Process parameter selection module: Used to establish a geometric model of the thin-walled battery shell for new energy using 3D modeling software, select the three process parameters that have the greatest impact on forming quality as optimization variables, and select the maximum forming thinning rate and the maximum forming thickening rate of the shell as indicators for evaluating forming quality;

[0039] Dataset construction module: Used to generate several sets of experimental combinations using BBD design experimental schemes for simulation calculation in finite element method, extract the maximum thinning rate and maximum thickening rate corresponding to each set of experiments to form an initial sample dataset, and divide it into training set and test set after normalization;

[0040] LNN prediction model building module: used to build improved LNN prediction models: activation states of the liquid hidden layer. It is governed by a linear first-order ordinary differential equation; and the LNN prediction model is trained using a training set and the prediction accuracy of the LNN prediction model is verified using a test set;

[0041] Process parameter optimization module: It is used to use the trained LNN prediction model as the fitness evaluation function of the MOPSO algorithm to predict the Pareto optimal solution set and select the unique optimal solution as the preferred combination of process parameters.

[0042] In a third aspect of the invention, an electronic device is provided. The electronic device includes a memory and a processor, the memory storing a computer program, the processor executing the program to implement the method according to a first aspect of the invention.

[0043] In a fourth aspect of the invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method according to a first aspect of the invention.

[0044] This invention constructs a high-precision, highly generalized process parameter-quality index proxy model by integrating experimental design, liquid neural network and multi-objective optimization algorithm. It systematically captures the complex coupling relationship and nonlinear effect between multiple process parameters and effectively achieves the synergistic optimization of conflicting quality objectives.

[0045] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description.

[0046] The beneficial effects of this invention are:

[0047] 1. By integrating experimental design, liquid neural networks and multi-objective optimization algorithms, a high-precision and highly generalized process parameter-quality index proxy model was constructed. This model systematically captures the complex coupling relationships and nonlinear effects among multiple process parameters, effectively achieving the synergistic optimization of conflicting quality objectives.

[0048] 2. The stability of the stamping process is significantly improved, the consistency of the quality of the processed parts is significantly improved, and the number of tests and costs required by traditional trial and error methods are greatly reduced.

[0049] 3. By optimizing the combination of process parameters, not only has the processing quality and production efficiency been significantly improved, but reliable technical support has also been provided for dealing with dynamic factors such as material fluctuations and mold wear in actual production. It has shown outstanding performance in controlling the stability of the deep drawing and forming of the shell and the forming thinning rate, thereby improving the overall robustness and economy of the battery shell manufacturing process. Attached Figure Description

[0050] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Wherein:

[0051] Figure 1 A flowchart of a method for optimizing stamping process parameters of thin-walled battery casing based on LNN prediction according to an embodiment of the present invention is shown.

[0052] Figure 2 A schematic diagram of the mechanical model for thinning and deep drawing of a shell according to an embodiment of the present invention is shown;

[0053] Figure 3 A schematic diagram of a liquid neural network according to an embodiment of the present invention is shown;

[0054] Figure 4 A Pareto front analysis diagram according to an embodiment of the present invention is shown;

[0055] Figure 5 A block diagram of an apparatus for optimizing stamping process parameters of thin-walled battery casing based on LNN prediction according to an embodiment of the present invention is shown.

[0056] Figure 6 A schematic diagram of an apparatus for optimizing the stamping process parameters of thin-walled battery casings based on LNN prediction, according to an embodiment of the present invention, is shown. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] According to an embodiment of the present invention, a method for optimizing the stamping process parameters of thin-walled battery casing based on LNN prediction is proposed. By integrating experimental design, liquid neural network and multi-objective optimization algorithm, a high-precision and highly generalized process parameter-quality index surrogate model is constructed, which systematically captures the complex coupling relationship and nonlinear effect between multiple process parameters and effectively achieves the synergistic optimization of contradictory quality objectives.

[0059] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.

[0060] Figure 1 This is a schematic flowchart of a method for optimizing stamping process parameters of thin-walled battery casing based on LNN prediction, according to an embodiment of the present invention. The method includes:

[0061] Step S01: Use 3D modeling software to establish a geometric model of the thin-walled battery shell for new energy, select the three process parameters that have the greatest impact on the forming quality as optimization variables, and select the maximum forming thinning rate and the maximum forming thickening rate of the shell as indicators for evaluating the forming quality.

[0062] Step S02: Using BBD to design an experimental scheme, generate several sets of experimental combinations and perform simulation calculations in the finite element method. Extract the maximum thinning rate and maximum thickening rate corresponding to each set of experiments to form an initial sample dataset, which is then normalized and divided into a training set and a test set.

[0063] Step S03: Constructing an improved LNN prediction model: Activation state of the liquid hidden layer It is governed by a linear first-order ordinary differential equation; and the LNN prediction model is trained using a training set and the prediction accuracy of the LNN prediction model is verified using a test set;

[0064] Step S04: Use the trained LNN prediction model as the fitness evaluation function of the MOPSO algorithm to predict the Pareto optimal solution set and select the unique optimal solution from it as the preferred combination of process parameters.

[0065] It should be noted that although the operation of the method of the present invention has been described in a specific order in the above embodiments and figures, this does not require or imply that the operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0066] To provide a clearer explanation of the above-mentioned method for optimizing the stamping process parameters of thin-walled battery casings based on LNN prediction, a specific embodiment will be used for illustration below. However, it is worth noting that this embodiment is only for better illustrating the present invention and does not constitute an improper limitation of the present invention.

[0067] The following example will further illustrate the method for optimizing the stamping process parameters of thin-walled battery casings based on LNN prediction.

[0068] In this embodiment, the aluminum shell of a 30194-type square battery is taken as an example. The object to be processed is a square shell of a power battery for a certain new energy vehicle, with dimensions of 194.3mm in length, 30.2mm in width, and 8.4mm in height, a long side wall thickness of 0.48mm, a short side wall thickness of 0.63mm, and the material is AL 3003-H14 with a sheet thickness of 1.0mm. The basic process requirements are: the maximum thinning rate must be controlled within 20% to prevent cracking, and the maximum thickening rate must be controlled within 10% to prevent mold jamming.

[0069] Step S01: Use 3D modeling software to establish a geometric model of the thin-walled battery shell for new energy, select the three process parameters that have the greatest impact on the forming quality as optimization variables, and select the maximum forming thinning rate and the maximum forming thickening rate of the shell as indicators for evaluating the forming quality.

[0070] A geometric model of a thin-walled battery casing for new energy was created using 3D modeling software and imported into finite element analysis software. The material of the sheet metal was defined, and the actual stress-strain curve and hardening index of the material were input based on the material's mechanical property test data. Thickness anisotropy coefficient Based on parameters such as blanking, deep drawing, shaping, and trimming, the stamping process flow is determined to be blanking, deep drawing, shaping, and trimming.

[0071] like Figure 2 As shown, based on the theory of metal plastic forming and pre-analysis, the three process parameters that have the greatest impact on forming quality are selected as optimization variables, denoted as vectors. .

[0072] Select the maximum forming thinning rate of the shell Maximum forming thickness increase As an indicator for evaluating forming quality.

[0073] Maximum thinning rate Calculation formula:

[0074] ,

[0075] Maximum thickness increase Calculation formula:

[0076] ,

[0077] in, The initial thickness of the sheet metal. This is the minimum thickness after forming. This represents the maximum thickness after forming.

[0078] In this embodiment, a geometric model of the thin-walled battery casing for new energy was established using UG_NX 3D modeling software and imported into Dynaform (a CAE software dedicated to sheet metal forming simulation) finite element analysis software in IGS format. The sheet metal material was defined as AL3003-H14. The blank holder force is expressed in kN. The coefficient of friction, This refers to the mold clearance, in mm.

[0079] Step S02: Using BBD to design an experimental scheme, several sets of experimental combinations are generated and simulated in the finite element method. The maximum thinning rate and maximum thickening rate corresponding to each set of experiments are extracted to form an initial sample dataset, which is then normalized and divided into a training set and a test set.

[0080] The experimental scheme was designed using Box-Behnken Design (BBD, an efficient second-order experimental design method for response surface methodology), based on process parameters. The range of values ​​is coded as low level, medium level, and high level.

[0081] Generated from BBD The test combinations were simulated in finite element software, and the maximum thinning rate corresponding to each test combination was extracted. and maximum thickness increase An initial sample dataset is formed. The sample data is normalized to eliminate the influence of units, and the sample data is mapped to the interval [0,1] and divided into training set and test set.

[0082] Step S03: Construct an improved LNN prediction model, train the LNN prediction model using the training set, and verify the prediction accuracy of the LNN prediction model using the test set.

[0083] To address the nonlinear and multi-physics coupling characteristics in the stamping process, this invention introduces a liquid neural network (LNN) as a surrogate model. The hidden layer neuron states of the LNN are described by ordinary differential equations, possessing continuous-time dynamics and inherent filtering characteristics, making it more sensitive to minute changes in process parameters and better able to capture their dynamic impact on quality indicators.

[0084] Taking the AL 3003-H14 thin-walled rectangular battery casing as the object, a multi-pass stamping finite element model was established using Dynaform software.

[0085] Among them, the stamping speed is selected. (mm / s), blank holder force (kN), friction coefficient Die clearance (mm) is used as the input variable to select the maximum forming thinning rate after shell forming. (%) and maximum forming thickness The percentage (%) was used as the output variable and generated using Latin hypercube sampling (LHS). Combine process parameters, run simulations, extract corresponding forming quality data, and construct a time-series dataset. .

[0086] like Figure 3 As shown, the LNN prediction model constructed in this invention includes an input layer, a liquid hidden layer, and an output layer.

[0087] Input Layer: Contains 4 neurons, each corresponding to a normalized process parameter. Although the process parameters are constant at the setpoint, they are treated as signals that are continuously input over time t in the model to drive the dynamic evolution of the system.

[0088] Liquid Layer: The core of the simulation of the stamping physics process. This layer contains N fully connected recurrent neurons. Unlike traditional RNNs, the activation states of neurons in this layer... Controlled by linear first-order ordinary differential equations (ODEs), simulating the "missing integral" characteristic in physical systems:

[0089] ,

[0090] in, For the hidden state vector, It is a time constant. For the input vector, and These are the input weight matrix and the recursive weight matrix, respectively. For bias vectors, It is a non-linear activation function.

[0091] state It is used to characterize the instantaneous physical field state of a material, such as equivalent plastic strain and residual stress.

[0092] time constant For learnable parameter vectors, smaller The transient response of the simulated material to rapid impacts such as the instant of stamping contact is relatively large. By simulating the work hardening history and other long-term memory effects during the deep drawing process of materials, this design enables the network to simultaneously capture local abrupt changes and global trends during the stamping process.

[0093] Output Layer: Contains 2 neurons, which linearly map the liquid layer... The steady-state output at time step 1 is converted into a prediction quality metric:

[0094] ,

[0095] in, The weight matrix of the output layer. This is the bias vector for the output layer.

[0096] The LNN prediction model is trained using the backpropagation algorithm over time combined with the Adam optimizer, enabling it to accurately predict given process parameters. and Construct the weighted mean squared error (MSE) loss function:

[0097] ,

[0098] in, , These are the weighting coefficients. To determine the coefficients, during the forward propagation of training, the differential equations are discretized using either the explicit Euler method or the Runge-Kutta method.

[0099] The prediction accuracy of the improved LNN prediction model was validated using a test set, employing root mean square error and coefficient of determination Rm. 2 As an evaluation criterion, if R² > 0.95, it ensures that it can be used as a reliable surrogate model to replace time-consuming finite element simulation.

[0100] Specifically, considering the multi-stage stamping characteristics of thin-walled battery casings in new energy applications, this invention utilizes the liquid hidden layer state vector of an LNN. The specific steps to achieve physical feature inheritance between processes are as follows:

[0101] The residual stress tensor of the sheet metal extracted in the first pass, i.e., the initial deep drawing simulation. Equivalent plastic strain The hardening parameters are defined as the physical legacy features from the previous pass, and are mapped to the initial state of the liquid hidden layer using a feature extraction operator. ,in This represents the current process number.

[0102] Unlike traditional neural networks that reset the initial state to zero during each prediction pass, this invention establishes an inter-process state transfer function. , set the first End of the course The hidden layer state is Then the first Initial state of the path Defined as:

[0103] ,

[0104] in, For the station transfer disturbance parameters between each pass, such as ambient temperature fluctuations and lubrication attenuation coefficients, this mechanism makes the first pass... The improved LNN prediction model for each pass has the work hardening history and geometric deformation memory caused by previous processing at the initial moment.

[0105] No. The LNN prediction model for each pass inherits the initial state and then incorporates the process parameters of the current process. The state vector is driven by the controlled ODE (Ordinary Differential Equation) equation. Towards This evolution enabled the prediction of the final forming quality under the cumulative effect of multiple passes.

[0106] In this embodiment, the initial parameter range is set as: stamping speed Blank pressure coefficient of friction mold gap 80 sets of samples were generated using LHS sampling for training the LNN, and 20 sets were used for testing. The number of hidden layer nodes... The activation function chosen is tanh, and the ODE solver step size is... After training for 500 epochs, the root mean square error (RMSE) on the test set decreased to 0.0042, and the prediction time was only 0.05 seconds per prediction, which is significantly more efficient than the 2 hours per prediction time of finite element simulation.

[0107] Step S04: Use the trained LNN prediction model as the fitness evaluation function of the MOPSO algorithm to predict the Pareto optimal solution set and select the unique optimal solution from it as the preferred combination of process parameters.

[0108] Set particle swarm size Maximum number of iterations Inertial weight Learning factor , The position of each particle represents a combination of process parameters. .

[0109] The trained LNN prediction model is used as the fitness evaluation function of MOPSO (Multi-Objective Particle Swarm Optimization), and the optimization objective is to simultaneously minimize the maximum thinning rate and the maximum thickening rate.

[0110] ,

[0111] The constraints are: if the thinning rate is less than the limit value, the shell forming will not crack; if the thickness increase rate is less than the limit value, the shell will not wrinkle.

[0112] Update the particle's velocity and position, call the LNN prediction model to predict and calculate the fitness value for each new position, update the individual optimal solution (Pbest) and the global optimal solution (Gbest), and maintain the external archive to store non-dominated solutions (Pareto optimal solution set).

[0113] After the iteration, the Pareto front is obtained due to the conflict between the maximum thinning rate and the maximum thickening rate. Based on the quality assessment that the impact of cracking on thin-walled battery casing is greater than that of wrinkling in actual production, the unique optimal solution X_opt is selected from the Pareto solution set as the preferred combination of process parameters using fuzzy membership function or weighted method.

[0114] In this embodiment, the trained LNN prediction model is embedded into the MOPSO algorithm, with a particle swarm size of 50 and 100 iterations. After the algorithm runs, a set of Pareto optimal solutions is obtained. Based on the high priority requirement for crack prevention in the production site, a set of process parameters biased towards low thinning rate is selected from these solutions. The optimized combination of process parameters is as follows: .

[0115] Before optimization, simulation results showed that the maximum thinning rate was 26.4%, which posed a risk of breakage, and the maximum thickening rate was 12.1%. After optimization, substituting the optimized parameters back into Dynaform, the maximum thinning rate was 16.8% and the maximum thickening rate was 8.5%. Based on these parameters, actual stamping production was carried out, and 500 pieces were produced continuously without any breakage. The wall thickness distribution was uniform and the surface quality was good.

[0116] The present invention also includes the verification of the optimization results. The obtained preferred process parameter combination X_opt is input into the finite element model for verification again. The predicted values ​​are compared with the simulated values, and the relative error is calculated.

[0117] The optimal process parameter combination X_opt was applied to the actual mold trial. The pressure force of the hydraulic press was adjusted, the friction coefficient was controlled by applying lubricating oil, and the mold clearance was adjusted to carry out small-batch trial production.

[0118] The prototype battery casing was subjected to 3D scanning and ultrasonic thickness measurement to detect its actual thinning rate and surface quality. If the actual product had no wrinkles or cracks and the uniformity of wall thickness distribution was significantly improved compared with that before optimization, the optimization method was deemed effective, and the process parameters were solidified for mass production.

[0119] To verify the prediction accuracy of the LNN surrogate model used in this invention, this embodiment selected 5 sets of validation samples and compared their prediction results with those of the traditional RSM model and experimental measured values, as shown in Table 1 below:

[0120] Table 1

[0121]

[0122] This table selects 5 representative verification samples, including boundary extreme points and optimized compromise solutions, with the comparison index being the maximum thinning rate Y1 (%).

[0123] At extreme points like Samples 2 and 4, which are at the edge of the process window, the equivalent plastic strain inside the sheet metal changes drastically. Due to the limitation of the second-order polynomial structure, RSM cannot capture this highly nonlinear abrupt change, resulting in an error exceeding 10%. LNN, through the dynamic evolution characteristics of the ODE in the hidden layer, can simulate the instantaneous hardening characteristics of the material during the stamping process and control the error within 2%.

[0124] In the optimal solution of sample 5 At this point, the LNN predicted value deviated from the measured value by only 0.71%, indicating that the surrogate model built based on LNN has extremely high fidelity across the entire domain. The generated Pareto front is closer to the physical reality than traditional methods. The average relative error of RSM is about 9.15%, while the average relative error of LNN is about 1.09%. The accuracy of the LNN prediction model in predicting the stamping quality of thin-walled battery casings is nearly 8 times higher than that of traditional RSM, ensuring the industrial applicability of the optimization results.

[0125] like Figure 4 As shown, this embodiment of the invention obtains the relevant Pareto front plot based on the LNN prediction model and the MOPSO algorithm. The plot shows the LNN dynamic search domain, the LNN-Pareto front, and the preferred process point. The preferred process point is selected based on engineering preferences, indicating that the maximum thinning rate and the maximum thickening rate have a significant conflicting relationship. The Pareto front clarifies the trade-off boundary between the two. The preferred process point prioritizes the control of the fracture risk and achieves quality synergistic optimization.

[0126] This embodiment demonstrates that by using a liquid neural network (LNN) to capture the dynamic time-varying characteristics during the stamping process, it is possible to accurately establish the mapping relationship between process parameters and quality indicators. Combined with the MOPSO algorithm, it effectively solves the problems of difficult forming of thin-walled battery casings and high scrap rate.

[0127] Based on the same inventive concept, this invention also proposes an apparatus for optimizing the stamping process parameters of thin-walled battery casings based on LNN prediction. The implementation of this apparatus can be found in the implementation of the method described above; repeated details will not be repeated. Figure 5 As shown, the device 100 includes:

[0128] Process parameter selection module 101: Used to establish a geometric model of a thin-walled battery shell for new energy using 3D modeling software, select the three process parameters that have the greatest impact on forming quality as optimization variables, and select the maximum forming thinning rate and the maximum forming thickening rate of the shell as indicators for evaluating forming quality.

[0129] Dataset construction module 102: Used to generate several sets of experimental combinations using BBD design experimental schemes, perform simulation calculations in finite element method, extract the maximum thinning rate and maximum thickening rate corresponding to each set of experiments to form an initial sample dataset, and divide it into training set and test set after normalization;

[0130] LNN Prediction Model Building Module 103: Used to build an improved LNN prediction model: activation state of the liquid hidden layer. It is governed by a linear first-order ordinary differential equation; and the LNN prediction model is trained using a training set and the prediction accuracy of the LNN prediction model is verified using a test set;

[0131] Process parameter optimization module 104: Used to use the trained LNN prediction model as the fitness evaluation function of the MOPSO algorithm to predict the Pareto optimal solution set and select the unique optimal solution as the preferred combination of process parameters.

[0132] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0133] like Figure 6 As shown, the device includes a central processing unit (CPU), which can perform various appropriate actions and processes based on computer program instructions stored in read-only memory (ROM) or loaded from storage units into random access memory (RAM). The RAM can also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0134] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0135] The processing unit executes the various methods and processes described above, such as method steps S01 to S04. For example, in some embodiments, method steps S01 to S04 may be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of method steps S01 to S04 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute method steps S01 to S04 by any other suitable means (e.g., by means of firmware).

[0136] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload programmable logic devices (CPLDs), and so on.

[0137] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0138] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0139] Furthermore, although the operations are described in a specific order, this should be understood as requiring that such operations be performed in the specific order shown or in sequential order, or requiring that all illustrated operations be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the invention. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0140] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for optimizing stamping process parameters of thin-walled battery casings based on LNN prediction, characterized in that, The method includes: Step S01: Use 3D modeling software to establish a geometric model of the thin-walled battery shell for new energy, select the three process parameters that have the greatest impact on the forming quality as optimization variables, and select the maximum forming thinning rate and the maximum forming thickening rate of the shell as indicators for evaluating the forming quality. Step S02: Using BBD to design an experimental scheme, several sets of experimental combinations are generated and simulated in the finite element method. The maximum thinning rate and maximum thickening rate corresponding to each set of experiments are extracted to form an initial sample dataset, which is then normalized and divided into a training set and a test set. Step S03: Constructing an improved LNN prediction model: Activation state of the liquid hidden layer It is governed by a linear first-order ordinary differential equation; and the LNN prediction model is trained using a training set and the prediction accuracy of the LNN prediction model is verified using a test set; The liquid hidden layer state vector The specific steps to achieve physical feature inheritance between processes are as follows: The residual stress tensor of the sheet metal extracted from the first-order deep drawing simulation Equivalent plastic strain The hardening parameters are defined as the physical legacy features from the previous pass, and are mapped to the initial state of the liquid hidden layer using a feature extraction operator. ,in Represents the current process sequence number; Establish inter-process state transfer function , set the first End of the course The hidden layer state is Then the first Initial state of the path Defined as: ， in, These are the station transfer disturbance parameters between each pass; No. The improved LNN model for each pass inherits the initial state and incorporates the process parameters of the current process. The state vector is driven by the controlled ODE equation. Towards evolution; Step S04: Use the trained LNN prediction model as the fitness evaluation function of the MOPSO algorithm to predict the Pareto optimal solution set and select the unique optimal solution from it as the preferred combination of process parameters. The specific steps are as follows: Set particle swarm size Maximum number of iterations Inertial weight Learning factor , The position of each particle represents a combination of process parameters. ; The trained improved LNN prediction model is used as the fitness evaluation function of MOPSO, and the optimization objective is to simultaneously minimize the maximum thinning rate and the maximum thickening rate: ， The constraints are: thinning rate < limit value, thickness increase rate < limit value; Update the particle's velocity and position, call the LNN prediction model to predict and calculate the fitness value for each new position, update the individual optimal solution and the global optimal solution, and maintain an external archive to store the Pareto optimal solution set; After the iteration, the Pareto front is obtained due to the conflict between the maximum thinning rate and the maximum thickening rate. Based on the quality assessment that the impact of thin-walled battery shell cracking is greater than that of wrinkling in actual production, a unique optimal solution is selected from the Pareto solution set as the preferred combination of process parameters using fuzzy membership function or weighted method.

2. The method for optimizing the stamping process parameters of thin-walled battery casing based on LNN prediction according to claim 1, characterized in that, The three process parameters mentioned in step S01 are: blank holder force, friction coefficient, and die clearance.

3. The method for optimizing the stamping process parameters of thin-walled battery casing based on LNN prediction according to claim 1, characterized in that, The maximum thinning rate mentioned in step S01 Calculation formula: ， Maximum thickness increase Calculation formula: ， in, The initial thickness of the sheet metal. This is the minimum thickness after forming. This represents the maximum thickness after forming.

4. The method for optimizing the stamping process parameters of thin-walled battery casing based on LNN prediction according to claim 1, characterized in that, The input layer of the improved LNN described in step S03 contains 4 neurons. These correspond to the normalized process parameters: stamping speed, blank holder force, friction coefficient, and die clearance.

5. The method for optimizing the stamping process parameters of thin-walled battery casing based on LNN prediction according to claim 1, characterized in that, The output layer of the improved LNN described in step S03 contains two neurons, and the liquid layer is mapped using a linear mapping. The steady-state output at time step 1 is converted into a prediction quality metric: ， in, The weight matrix of the output layer. This is the bias vector for the output layer.

6. An apparatus for optimizing stamping process parameters of thin-walled battery casings based on LNN prediction, characterized in that, The device implements the method as described in any one of claims 1 to 5, comprising: Process parameter selection module: Used to establish a geometric model of the thin-walled battery shell for new energy using 3D modeling software, select the three process parameters that have the greatest impact on forming quality as optimization variables, and select the maximum forming thinning rate and the maximum forming thickening rate of the shell as indicators for evaluating forming quality; Dataset construction module: Used to generate several sets of experimental combinations using BBD design experimental schemes for simulation calculation in finite element method, extract the maximum thinning rate and maximum thickening rate corresponding to each set of experiments to form an initial sample dataset, and divide it into training set and test set after normalization; LNN prediction model building module: used to build improved LNN prediction models: activation states of the liquid hidden layer. It is governed by a linear first-order ordinary differential equation; and the LNN prediction model is trained using a training set and the prediction accuracy of the LNN prediction model is verified using a test set; The liquid hidden layer state vector The specific steps to achieve physical feature inheritance between processes are as follows: The residual stress tensor of the sheet metal extracted from the first-order deep drawing simulation Equivalent plastic strain The hardening parameters are defined as the physical legacy features from the previous pass, and are mapped to the initial state of the liquid hidden layer using a feature extraction operator. ,in Represents the current process sequence number; Establish inter-process state transfer function , set the first End of the course The hidden layer state is Then the first Initial state of the path Defined as: ， in, These are the station transfer disturbance parameters between each pass; No. The improved LNN model for each pass inherits the initial state and incorporates the process parameters of the current process. The state vector is driven by the controlled ODE equation. Towards evolution; Process parameter optimization module: This module uses the trained LNN prediction model as the fitness evaluation function of the MOPSO algorithm to predict the Pareto optimal solution set and select the unique optimal solution as the preferred combination of process parameters. The specific steps are as follows: Set particle swarm size Maximum number of iterations Inertial weight Learning factor , The position of each particle represents a combination of process parameters. ; The trained improved LNN prediction model is used as the fitness evaluation function of MOPSO, and the optimization objective is to simultaneously minimize the maximum thinning rate and the maximum thickening rate: ， The constraints are: thinning rate < limit value, thickness increase rate < limit value; Update the particle's velocity and position, call the LNN prediction model to predict and calculate the fitness value for each new position, update the individual optimal solution and the global optimal solution, and maintain an external archive to store the Pareto optimal solution set; After the iteration, the Pareto front is obtained due to the conflict between the maximum thinning rate and the maximum thickening rate. Based on the quality assessment that the impact of thin-walled battery shell cracking is greater than that of wrinkling in actual production, a unique optimal solution is selected from the Pareto solution set as the preferred combination of process parameters using fuzzy membership function or weighted method.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 5.

Images