A method for optimizing process parameters of a new energy power battery shell based on RBF

Abstract

The application provides a new energy power battery shell process parameter optimization method based on RBF, and relates to the technical field of metal plate stamping forming and process optimization. The method comprises the following steps: determining stamping speed, blank holder force, friction factor and die gap as process parameters to be optimized, adopting optimized Latin hypercube sampling to obtain an initial process parameter combination matrix; performing multi-pass thinning drawing simulation through finite element analysis, and extracting the maximum thinning rate and the maximum principal strain; constructing an RBF proxy model, and improving the model precision through precision acceptance and dynamic point adding; constructing a multi-objective constraint optimization model with the maximum thinning rate and the maximum principal strain as optimization objectives, obtaining a Pareto optimal process parameter solution set by adopting an NSGA-III algorithm with a reference point mechanism, and determining the optimal process parameter combination based on TOPSIS, so as to realize the optimization of the multi-pass thinning drawing process parameters of the new energy power battery shell.

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 process parameters of new energy power battery casings based on RBF. Background Technology

[0002] The external protective shell of the battery pack in new energy vehicles is typically a deep-cavity, thin-walled rectangular box. Due to the requirements of internal space arrangement and lightweighting, the shell wall thickness accuracy is critical. These shells are mostly made of 3003-H14 aluminum alloy, formed through multiple processes such as stamping, deep drawing, thinning, and trimming. They are characterized by thin material, numerous processes, and high precision, making them highly susceptible to quality defects such as wrinkling, cracking, and uneven wall thickness during the forming process. Numerical simulation analysis reveals that the shell forming quality is significantly affected by process parameters such as stamping speed, blank holder force, friction coefficient, and die clearance. Traditional process parameter optimization often relies on experience or a limited number of numerical simulations for trial and error, which is time-consuming, costly, and difficult to obtain globally optimal process parameter combinations. In recent years, some studies have adopted methods combining surrogate models and intelligent optimization algorithms for process parameter optimization, such as linear neural networks (LNN) combined with the MOPSO algorithm, response surface methodology (RSM) combined with particle swarm optimization, and orthogonal experimental design combined with the NSGA-II algorithm.

[0003] However, the aforementioned existing technologies still have significant shortcomings. Surrogate models such as LNN and RSM have limited nonlinear mapping capabilities, making it difficult to accurately capture the strong nonlinear coupling between process parameters and quality indicators during multi-pass thinning and drawing processes, especially with a significant drop in prediction accuracy at the edge regions of the design space. Furthermore, classic experimental design methods such as BBD, orthogonal experiments, and CCD concentrate sample points in the corners and central regions of the design space, resulting in insufficient uniformity in filling the hyperdimensional space, further affecting the global generalization accuracy of the surrogate models. Regarding optimization algorithms, NSGA-II relies on crowding distance to maintain population diversity, which can easily lead to local clustering of the solution set under complex Pareto front shapes. The PSO algorithm also carries the risk of premature convergence, making it difficult to guarantee complete coverage of the global optimal solution. In addition, existing technologies lack quantitative acceptance thresholds for surrogate model accuracy and dynamic point addition mechanisms, failing to automatically supplement samples and update the model when accuracy is insufficient. Moreover, most methods do not systematically introduce multi-attribute decision mechanisms to accurately lock the Pareto solution set to the optimal engineering parameters, resulting in a disconnect between optimization results and actual production applications. Therefore, how to provide a method for optimizing the stamping process parameters of battery casings that can overcome the above-mentioned defects has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] To address the above issues, this invention employs optimized Latin hypercube sampling to ensure uniform filling of samples across the entire design space. It constructs a high-precision surrogate model using a radial basis function neural network, sets a precision acceptance threshold and a dynamic point addition mechanism, introduces the NSGA-III algorithm combined with a reference point mechanism to maintain population diversity and overcome local optima problems, and uses TOPSIS multi-attribute decision-making to achieve engineering-based locking of optimal parameters. This system optimizes the deep-drawing process parameters for new energy power battery casings, improving stamping stability and casing forming quality.

[0005] According to an embodiment of the present invention, a method for optimizing the process parameters of new energy power battery casing based on RBF is provided.

[0006] In a first aspect of the present invention, a method for optimizing process parameters of a new energy power battery casing based on RBF is provided. The method includes: Step S01: Determine the stamping speed, blank holder force, friction coefficient and die clearance as the process parameters to be optimized and construct the design space. Use optimized Latin hypercube sampling to obtain the initial process parameter combination matrix. Step S02: Input the initial process parameter combination matrix into the finite element analysis model to perform multi-pass thinning and deep drawing simulation, and extract the maximum thinning rate and the maximum principal strain; Step S03: Construct an RBF proxy model based on the process parameters and the corresponding maximum thinning rate and maximum principal strain, and obtain an RBF proxy model that meets the accuracy requirements through accuracy acceptance and dynamic point addition; Step S04: Using the maximum thinning rate and maximum principal strain as optimization objectives, and combining forming quality constraints and process parameter boundaries, a multi-objective constrained optimization model is constructed; Step S05: Based on the RBF surrogate model that meets the accuracy requirements, the NSGA-III algorithm with reference point mechanism is used to iteratively optimize the multi-objective constraint optimization model to obtain the Pareto optimal process parameter solution set; Step S06: Based on TOPSIS, perform a comprehensive decision on the Pareto optimal process parameter solution set to determine the optimal combination of process parameters.

[0007] Furthermore, in step S01, the design space is sampled using optimized Latin hypercube sampling, and the initial process parameter combination matrix is ​​obtained by optimizing the spatial distribution of the sample points.

[0008] Furthermore, in step S02, a multi-pass thinning and deep drawing numerical simulation is performed on each process parameter combination in the initial process parameter combination matrix using a finite element analysis model, and the maximum thinning rate and maximum principal strain corresponding to each process parameter combination are extracted from the simulation results after forming is completed.

[0009] Furthermore, the RBF proxy model described in step S03 takes the stamping speed, blank holder force, friction coefficient and die clearance as inputs, and the maximum thinning rate and maximum principal strain as outputs, and uses the coefficient of determination and preset acceptance threshold for accuracy acceptance.

[0010] Furthermore, when the accuracy of the RBF proxy model does not reach the preset acceptance threshold, dynamic points are added in the design space region corresponding to the prediction error to supplement finite element simulation samples and update the RBF proxy model until the RBF proxy model reaches the preset acceptance threshold.

[0011] Further, the multi-objective constraint optimization model mentioned in step S04 is specifically as follows: taking the maximum thinning rate and the maximum principal strain as dual optimization objectives, and minimizing the maximum thinning rate and the maximum principal strain respectively; the forming quality state constraint includes the maximum thinning rate not exceeding 20% ​​and the maximum thickness increase rate not exceeding 10%; the process parameter boundary constraint includes the stamping speed, blank holder force, friction coefficient and die clearance being within the corresponding preset value range.

[0012] Further, the specific steps of step S05 are as follows: the maximum thinning rate and maximum principal strain are predicted for the individuals in the NSGA-III algorithm population using the RBF surrogate model; offspring populations are generated through crossover and mutation; the parent and offspring populations are subjected to non-dominated sorting under constraint; when the number of individuals in the critical non-dominated layer exceeds the capacity of the next generation population, the individuals in the critical non-dominated layer are screened based on reference point association and niche counting, and the next generation population is formed according to the screening results; the iteration is terminated when the preset maximum number of iterations is reached or the preset population convergence condition is met, and the process parameter combination in the final non-dominated front is used as the Pareto optimal process parameter solution set; otherwise, the next generation iteration continues.

[0013] Furthermore, step S06 specifically includes: Establish a decision matrix for the Pareto optimal process parameter solution set; The decision matrix is ​​dimensionless to determine the positive and negative ideal solutions; Calculate the weighted distances between each candidate solution in the Pareto optimal process parameter solution set and the positive ideal solution and the negative ideal solution, respectively; The overall proximity of each candidate solution is calculated based on the weighted distance, and the candidate solution with the highest overall proximity is selected as the optimal combination of process parameters.

[0014] In a second 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 the first aspect of the invention.

[0015] In a third 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 the first aspect of the invention.

[0016] This invention employs optimized Latin hypercube sampling to ensure uniform filling of samples throughout the entire design space, constructs a high-precision surrogate model using radial basis function neural networks and sets accuracy acceptance thresholds and dynamic point addition mechanisms, introduces the NSGA-III algorithm combined with a reference point mechanism to maintain population diversity to overcome local optima problems, and achieves engineering locking of optimal parameters through TOPSIS multi-attribute decision-making. The system optimizes the deep drawing process parameters of new energy power battery shells, improving stamping stability and shell forming quality.

[0017] 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.

[0018] The beneficial effects of this invention are: 1. By integrating experimental design, RBF surrogate model and NSGA-III optimization algorithm, a high-precision and highly generalizable process parameter-quality index surrogate model is established. This model can accurately capture the strong nonlinear coupling relationship between process parameters and quality index during multi-pass thinning and deep drawing, improve the prediction accuracy and generalization ability of the surrogate model in the entire design space, and solve the problems of limited nonlinear mapping ability and uneven sample filling in traditional surrogate models. 2. This paper uses the NSGA-III optimization algorithm to optimize process parameters, effectively maintaining population diversity and obtaining a uniformly distributed global optimal solution set. This avoids the problems of local clustering and premature convergence of solution sets in traditional algorithms, ensuring the integrity and global optimality of process parameter optimization. 3. The process parameters optimized by the method of this invention can effectively control the stability of shell deep drawing and the forming thinning rate, improve the stability of stamping process and the consistency of formed part quality, avoid defects such as wrinkling and cracking, and at the same time significantly reduce the number of tests and trial mold costs, and significantly improve processing quality and production efficiency. Attached Figure Description

[0019] 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: Figure 1 A flowchart illustrating a method for optimizing the process parameters of a new energy power battery casing based on RBF according to an embodiment of the present invention is shown. Figure 2A comparison chart of the predicted values ​​of the RBF proxy model and the verified values ​​of finite element simulation according to an embodiment of the present invention is shown; Figure 3 A schematic diagram of the Pareto front and decision points obtained by the multi-objective evolutionary algorithm according to an embodiment of the present invention is shown. Figure 4 A schematic diagram of an apparatus for optimizing process parameters of a new energy power battery casing based on RBF according to an embodiment of the present invention is shown. Detailed Implementation

[0020] 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.

[0021] According to an embodiment of the present invention, a method for optimizing the process parameters of new energy power battery casing based on RBF is proposed. By adopting optimized Latin hypercube sampling to ensure uniform filling of samples in the entire design space, a high-precision surrogate model is constructed using radial basis function neural network and a precision acceptance threshold and dynamic point addition mechanism are set. The NSGA-III algorithm is introduced in combination with a reference point mechanism to maintain population diversity to overcome the local optimum problem. The optimal parameters are locked in an engineering manner through TOPSIS multi-attribute decision-making. The system optimizes the deep drawing process parameters of new energy power battery casing, improving stamping stability and casing forming quality.

[0022] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.

[0023] Figure 1 This is a schematic flowchart of a method for optimizing the process parameters of a new energy power battery casing based on RBF according to an embodiment of the present invention. The method includes: Step S01: Determine the stamping speed, blank holder force, friction coefficient and die clearance as the process parameters to be optimized and construct the design space. Use optimized Latin hypercube sampling to obtain the initial process parameter combination matrix. Step S02: Input the initial process parameter combination matrix into the finite element analysis model to perform multi-pass thinning and deep drawing simulation, and extract the maximum thinning rate and the maximum principal strain; Step S03: Construct an RBF proxy model based on the process parameters and the corresponding maximum thinning rate and maximum principal strain, and obtain an RBF proxy model that meets the accuracy requirements through accuracy acceptance and dynamic point addition; Step S04: Using the maximum thinning rate and maximum principal strain as optimization objectives, and combining forming quality constraints and process parameter boundaries, a multi-objective constrained optimization model is constructed; Step S05: Based on the RBF surrogate model that meets the accuracy requirements, the NSGA-III algorithm with reference point mechanism is used to iteratively optimize the multi-objective constraint optimization model to obtain the Pareto optimal process parameter solution set; Step S06: Based on TOPSIS, perform a comprehensive decision on the Pareto optimal process parameter solution set to determine the optimal combination of process parameters.

[0024] 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.

[0025] To provide a clearer explanation of the above-mentioned method for optimizing the process parameters of new energy power battery casing based on RBF, a specific embodiment is described 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.

[0026] The following example will further illustrate the method for optimizing the process parameters of new energy power battery casings based on RBF.

[0027] Step S01: Determine the stamping speed, blank holder force, friction coefficient and die clearance as the process parameters to be optimized and construct the design space. Use optimized Latin hypercube sampling to obtain the initial process parameter combination matrix.

[0028] The process parameters to be optimized and their value ranges are determined, a multi-dimensional design space is constructed, and the stamping speed is established for the multi-pass thinning deep drawing process of new energy thin-walled power battery shells. Blank pressure coefficient of friction Die clearance The core design variables to be optimized are shown below, and the design variable space is as follows: Its spatial boundary is defined as: , in, mm / min; kN; ; mm.

[0029] To achieve globally uniform and edge-aligned sample acquisition in a four-dimensional hypervolume space, the Optimized Latin Hypercube Sampling (OLHS) algorithm is employed. This algorithm optimizes the spatial distribution by minimizing the enhanced entropy criterion or maximizing the minimum Euclidean distance between sample points (Maximin distance criterion). Its objective function is expressed as: , In the formula, For sample points and The Euclidean distance between them; To design variable dimensions, this optimized sampling is performed to efficiently obtain data in hyperdimensional space. The set of initial process parameter combination matrices possesses high spatial representativeness. .

[0030] Step S02: Input the initial process parameter combination matrix into the finite element analysis model to perform multi-pass thinning and deep drawing simulation, and extract the maximum thinning rate and the maximum principal strain.

[0031] The sampled The parameter matrix is ​​imported in batches into finite element analysis software for multi-pass thinning and deep drawing numerical simulation. For each sample point, the physical quantities of all nodes after forming are extracted through post-processing. This scheme focuses on the two most critical forming defect indicators of the power battery shell: Maximum thinning rate Used to characterize the deep drawing limit and fracture tendency of materials, its calculation formula is: , In the formula, This is the initial sheet thickness. For the shell first i Real-time calculated thickness of each grid node; Maximum principal strain Used to indirectly characterize the degree of instability and wrinkling in the unsupported suspended areas of thin-walled parts.

[0032] In existing related technical solutions, the characterization of wrinkling trend usually relies solely on the geometric indicator of maximum thickness gain. However, for the multi-pass thinning and deep drawing process of thin-walled battery shells in new energy, wrinkling is often not simply an accumulation in the thickness direction, but rather originates from the unsupported compressive instability of suspended areas under compressive stress. This invention abandons the traditional single thickness gain evaluation method and introduces the maximum principal strain. As a collaborative optimization objective, compared to the thickness increase rate, the principal strain quantifies the flow and accumulation trend of the material from the global perspective of the deformation field. It can capture the critical state of instability earlier and more sensitively. By decoupling the thickness reduction and the principal strain into dual objectives, it overcomes the problem of misjudgment of optimization objectives caused by single indicators or unclear physical mechanisms in existing technologies, and provides a more scientific physical label for subsequent high-precision surrogate models.

[0033] Step S03: Construct an RBF proxy model based on process parameters and the corresponding maximum thinning rate and maximum principal strain, and obtain an RBF proxy model that meets the accuracy requirements through accuracy acceptance and dynamic point addition.

[0034] Since explicit finite element methods take an extremely long time to calculate a single process cycle, this invention introduces a radial basis function (RBF) neural network to construct a nonlinear process response surface surrogate model. The topology of the RBF network includes an input layer, namely the stamping speed. Blank pressure coefficient of friction Die clearance Four input nodes, hidden layer, and output layer, representing the maximum thinning rate. and maximum principal strain Two output metrics.

[0035] For any input vector The prediction output function of the RBF surrogate model is expressed as: , In the formula: For the first The center vector of each training sample; These are the corresponding connection weight coefficients; This is the bias constant; The radial basis function represents the Euclidean distance between the point to be determined and the center point. Using the Gaussian function: , In the formula, To control the influence range of the basis functions, the deformation parameter (Width Parameter) is adaptively adjusted through global cross-validation.

[0036] To ensure the generalization accuracy of the surrogate model across the entire design space, a coefficient of determination is introduced. As the threshold for model acceptance: , In the formula: This represents the actual response value output by the finite element simulation. The predicted value for the RBF surrogate model; The arithmetic mean of the true values; The number of samples in the test set is [number]. The system automatically triggers a dynamic addition mechanism to recalculate until the accuracy meets the standard.

[0037] Existing surrogate model techniques suffer from two main shortcomings: Linear Neural Networks (LNNs), being essentially linear mappings, struggle to accurately fit the strongly nonlinear, multi-peak coupling relationship between blank holder force, friction coefficient, and thinning rate; Response Surface Methods (RSM) can only construct low-order polynomials and cannot characterize the dramatic abrupt changes in material flow during multi-pass thinning and drawing processes. To address these shortcomings, this invention introduces a Radial Basis Function (RBF) neural network. Leveraging its Gaussian kernel function's ability to balance local receptive fields and global smoothness, it theoretically enables the approximation of nonlinear continuous functions with arbitrary precision. Furthermore, an adaptive deformation parameter is added. The global cross-validation strategy enables the model to maintain stable interpolation and extrapolation capabilities even in sparse sample regions.

[0038] Traditional proxy models have a fixed structure after initial sampling; if the accuracy is insufficient, they need to be rebuilt, which is costly. This invention proposes a dynamic point-based acceptance mechanism for the first time in this field: setting a determination coefficient. To enforce the acceptance threshold, when the model accuracy fails to meet the standard, the system automatically refines sampling in the neighborhood of the design space with the largest prediction error, adds new sample points, and reconstructs the RBF model, iterating until the threshold requirement is met. This closed-loop mechanism overcomes the problems of insufficient model generalization ability and unreliable optimization results caused by sample fixation in LNN and RSM methods.

[0039] Step S04: Using the maximum thinning rate and maximum principal strain as optimization objectives, and combining forming quality constraints and process parameter boundaries, a multi-objective constraint optimization model is constructed.

[0040] To effectively prevent cracking and wrinkling of thin-walled power battery casings during multi-stage stamping, this step transforms the engineering problem into a standard multi-objective constrained optimization mathematical model based on the physical mechanism of the process. The specific construction process is as follows.

[0041] A dual-objective fitness function is established to minimize the cracking risk and wrinkling tendency during the forming process: minimizing the maximum thinning rate. and minimizing the maximum principal strain The overall objective function vector expression is as follows: .

[0042] Based on the basic process specifications and equipment limits of the aluminum casing for prismatic batteries made of AL 3003-H14 material, state constraints and boundary constraints are set as follows: Condition constraints: To prevent cracking, the maximum thinning rate must not exceed 20%; to prevent material accumulation from causing mold jamming, the maximum thickness increase rate... It is strictly forbidden to exceed 10%; Boundary constraints: The upper and lower limits of the parameters for standard optimization.

[0043] By combining the above variables, objectives, and constraints, a complete mathematical optimization model is constructed as follows: , Due to the two established optimization objectives, the maximum thinning rate With maximum principal strain There are inherent conflicts in the physical mechanisms; for example, while increasing the blank holder force is beneficial for suppressing wrinkling strain, However, it will exacerbate the thinning rate. Since there is no single optimal solution for this optimization problem that makes all objectives reach their absolute minimum simultaneously, the theoretical optimal solution of the mathematical model built in this step is not a single data point, but a set of Pareto optimal solutions that balances the conflicts among the objectives.

[0044] The Pareto dominance criterion is: for any two feasible process parameter solutions... and ,like Pareto Domination Then, if and only if the following conditions are met simultaneously: , That is, solution Non-inferior to the solution on all objectives And is strictly superior to the solution in at least one objective. The set of all decision variables that are not dominated by any other solution in the feasible region together constitutes the Pareto optimal front of this multi-objective optimization problem.

[0045] Since the RBF surrogate model is a pure numerical simulation mapping, the Pareto solution set cannot be directly obtained through analytical differentiation. Therefore, in the subsequent step S05, this invention introduces a non-dominated sorting genetic algorithm (NSGA-III) with a reference point mechanism to perform efficient swarm intelligence parallel search on the surrogate model to approximate the complete distribution of the Pareto front of the theory. Subsequently, in step S06, based on actual engineering preferences, the TOPSIS multi-attribute decision method is used to lock the unique optimal compromise solution from the solution set. Thus, the complete three-stage technical path of this invention, from mathematical model building, intelligent solution set search, and engineering decision locking, is completed.

[0046] Step S05: Based on the RBF surrogate model, the NSGA-III algorithm with reference point mechanism is used to iteratively optimize the multi-objective constraint optimization model to obtain the Pareto optimal process parameter solution set.

[0047] For the nonlinear, multi-constraint optimization model established above, the objective function in step S04 exhibits typical conflict characteristics, namely, increasing the blank holder force. Reduces wrinkling strain However, it will drastically increase the thinning rate. To avoid getting trapped in local optima, this step introduces the Non-Dominated Sorting Genetic Algorithm with Reference Point (NSGA-III) to perform a solution set search in the hyperdimensional space on the pre-constructed RBF surrogate model. The specific execution logic is as follows.

[0048] Setting the core control hyperparameter of the NSGA-III algorithm: population size Maximum number of iterations Crossover probability Probability of mutation .

[0049] On the normalized hyperplane of the two objectives, uniformly distributed reference points are generated using the Das and Dennis systematic methods. The calculation formula is: , In the formula, The number of divisions on each objective axis. These reference points will guide the population to converge uniformly toward the Pareto Front, maintaining population diversity.

[0050] After completing algorithm initialization and reference point generation, the main iteration loop of NSGA-III begins. During each generation of evolution, for each individual in the population... Its fitness value is not calculated directly using time-consuming finite element simulation software, but rather by directly calling the accuracy verification already completed in step S03. The RBF neural network surrogate model is used for fast mapping; specifically, the individual... The four-dimensional process parameters are input into the RBF proxy model. The model uses a Gaussian kernel function to perform a weighted summation and outputs the predicted values ​​of the two forming quality indicators in milliseconds. , , In the formula, For the first j The center vectors of the training samples , For the connection weights of the trained RBF network, , This is the bias constant.

[0051] In existing technologies, while LNNs can be used as the fitness function for MOPSO to achieve rapid evaluation, their linear nature means they cannot accurately fit the strong nonlinear coupling relationship between blank holder force, friction coefficient, and thinning rate. In contrast, this invention uses an RBF neural network, whose Gaussian kernel function has the ability to infinitely approximate nonlinearity, in conjunction with step S03. The precision acceptance threshold and dynamic point addition mechanism ensure that the fitness evaluation of each individual during population iteration is based on a high-fidelity surrogate model. As a result, the evaluation time for a single iteration is drastically reduced from hours in finite element simulation to milliseconds, enabling the NSGA-III algorithm to... The Pareto front can be fully searched during generational evolution without incurring any additional finite element computation costs.

[0052] Based on the current parent population Based on this, simulated binary crossover (SBX) and polynomial mutation are used to generate offspring populations. When performing crossover and mutation operations, the decision variables for all individuals are real process parameters in physical space, namely stamping speed. v Blank pressure F h coefficient of friction μ Die clearance C To ensure that the offspring produced by genetic manipulation do not deviate from the actual physical boundaries, the variable values ​​after crossover and mutation must be strictly clamped to the design space boundary X defined in step S01. min and Within that timeframe, after completing the above operations, the parent population will be... With offspring population Merging to form a scale of Mixed population .

[0053] For merged populations For each individual in the equation, the objective function value is... and All of them are directly derived from the predicted values ​​of the RBF surrogate model trained in step S03, that is and Instead of finite element simulation, a fast non-dominated sorting is performed based on this: For each individual in the population, the degree of constraint violation is determined. The process constraints of this invention include two categories: Thinning rate constraint: The maximum thinning rate predicted by the RBF model is directly determined. Thickness increase rate constraint: Among them, the thickening rate The output is also mapped through the RBF proxy model.

[0054] This invention uses the constraint-dominance principle to divide the non-dominant hierarchy: Individual Constraint and Domination of Individuals The conditions are: 1. If Feasible (satisfies all constraints) 1. Not feasible; 2. Or both are not feasible. The degree of violation of the constraints is smaller; 3, or both are feasible, Pareto dominance on dual objectives .

[0055] Based on the above criteria, mixed populations will be... Layer by layer, it is divided into non-dominated frontiers ,in Let be the optimal non-dominated solution set in the current generation.

[0056] The existing MOPSO algorithm relies on external archives to maintain non-dominated solutions, and its particle velocity update strategy is prone to generating out-of-bounds invalid solutions when dealing with discrete / continuous mixed variable spaces of process parameters. In contrast, this invention uses SBX crossover and polynomial mutation, which is specifically designed for continuous process parameter spaces. Combined with the millisecond-level fitness response of the RBF surrogate model, it can perform a complete constrained non-dominated sorting of all individuals in each generation of population iteration, ensuring that the optimal process parameters evolve stably along the true Pareto front direction within the physically feasible domain.

[0057] When the critical layer When the number of individuals exceeds the remaining population quota, not all of them can enter the next generation of the population. In this step, NSGA-III does not use the crowding distance of the traditional NSGA-II because it is prone to local clustering of the solution set in the high-dimensional objective space. Therefore, a reference point association and niche preservation strategy are introduced to screen individuals in the critical layer. All objective function values ​​in this step are calculated based on the predicted output of the RBF surrogate model in step S03. The specific selection process is as follows: In the current mixed population In the process, the minimum values ​​in two target directions are found based on the RBF prediction values, which constitute the ideal point for the current generation. Then, the target vector of all individuals in the population is subtracted from the ideal point and the coordinates are translated so that the origin of the target space is moved to the vicinity of the current optimal frontier.

[0058] The achievement scalarization function (ASF) is used to calculate the extreme points along each coordinate axis, constructing an intercept hyperplane to normalize the target vectors of all individuals into a unified dimension space with an intercept of 1. The purpose of this normalization process is to eliminate the deviation in subsequent distance calculations caused by the different dimensions of the thinning rate (unit: %) and the principal strain (dimensionless).

[0059] Will A uniform set of reference points radiates outward from the origin of the normalized hyperplane, forming a reference line covering the entire target space. For each normalized individual target vector in the population... Calculate its vertical Euclidean distance to each reference line: , In the formula, For the first The target vector of each individual after RBF prediction and normalization Given a reference line, the direction vector is used to associate each individual with the nearest reference line.

[0060] Count the number of individuals currently associated with each reference line. That is, niche counting, from the critical layer Priority selection The individual corresponding to the reference line with the smallest value is selected for the next generation, and this reference line is updated after each selected individual. This mechanism ensures that even in regions with complex or locally sparse Pareto front shapes, combinations of process parameters are preserved, thus preventing the population from prematurely converging to a local tradeoff region.

[0061] If the current generation Or, the optimal frontier of the population has not been significantly updated within 50 consecutive generations, meaning the average movement distance of the frontier is less than a threshold. If the iteration terminates, the non-dominated frontier of the final generation is output. The entire set of individuals included constitutes the complete Pareto optimal solution set searched by this invention, which is used for the TOPSIS decision in subsequent step S06; otherwise, let Return to continue iterating.

[0062] Existing technologies employ NSGA-II, which relies on crowding distance to maintain population diversity. This mechanism tends to cause the solution set to exhibit a centrally clustered and sparsely distributed distribution in the target space when dealing with complex Pareto fronts. The reference point mechanism introduced in this invention, based on the OLHS sample space coverage in step S01, further ensures the uniform coverage of the optimal solution set on the Pareto front from an algorithmic perspective, thus enabling the TOPSIS decision in step S06 to have a richer range of trade-off candidate schemes.

[0063] When the number of iterations reaches When the optimization process terminates, the algorithm finally outputs a Pareto optimal set that converges and is uniformly distributed on the boundary of the feasible solution space. No combination of parameters in this set is absolutely superior or inferior, representing the best theoretical boundary of the physical trade-off between the risk of cracking and the risk of wrinkling of the battery casing, providing complete high-dimensional data support for the engineering decision in the subsequent step S06.

[0064] Step S06: Based on TOPSIS, perform a comprehensive decision on the Pareto optimal process parameter solution set to determine the optimal combination of process parameters.

[0065] Faced with the Pareto front solution set, a ranking technique approximating the ideal solution (TOPSIS) is used to achieve optimal parameter locking for practical engineering applications. Establish a decision matrix ,in Let be the number of non-dominated solutions. The number of objective functions; Perform dimensionless normalization on the decision matrix: ; Determine the positive ideal solution vector (All indicators reach their minimum) and negative ideal solution vector (All indicators have reached their maximum); Calculate the weighted distance from each alternative solution to the positive and negative ideal solutions. and And calculate the overall similarity. : , Select Proximity The highest solution is taken as the optimal combination of process parameters that is ultimately locked globally.

[0066] In this embodiment, the object being 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 wall thickness of 0.48mm on the long side, a wall thickness of 0.63mm on the short side, and is made of AL 3003-H14 with a sheet thickness of 1.0mm.

[0067] Material properties: AL 3003-H14 aluminum alloy, specifically designed for lightweight applications in new energy sources, is selected. The initial sheet thickness is 1.0 mm. The elastic modulus of this material is... Yield strength ,tensile strength Anisotropy index , , .

[0068] The core control hyperparameter of the NSGA-III algorithm: population size Maximum number of iterations Crossover probability Probability of mutation .

[0069] Table 1 shows representative partial design points (DoE) generated by sampling within the design space using OLHS and the corresponding finite element simulation response output data: Table 1

[0070] After training on 80 sets of global samples, the RBF surrogate model demonstrated excellent fitting performance on the independent test set, with its coefficient of determination reaching [value missing]. The error histogram distribution closely approximates the perfect prediction line, such as... Figure 2 As shown, this directly proves the accuracy and high generalization performance of the surrogate model when replacing time-consuming finite element calculations.

[0071] Relying on NSGA-III to perform global iteration, TOPSIS is introduced into the Pareto solution cluster generated by it to achieve solution set convergence for the trade-off strategy. Finally, the value of the globally optimal process parameter with the highest closeness is locked as: stamping speed. Blank pressure coefficient of friction mold gap .

[0072] like Figure 3 As shown, under this optimized parameter combination, the system output shell process quality response is as follows: maximum thinning rate Maximum principal strain Compared with the original unoptimized empirical parameters (sample 3), the maximum thinning rate was significantly reduced from 13.82% to 8.3%, a decrease of 39.9%. At the same time, the maximum principal strain remained in the safe and stable zone with no wrinkling and high ductility. The formed battery casing had no local cracks, and the consistency of the thickness of the long and short sides was improved by 32.5%. This embodiment proves that by using the RBF proxy model, the mapping relationship between process parameters and quality indicators can be accurately established. Combined with the NSGA-III optimization algorithm, the problems of difficult forming of thin-walled battery casing and high scrap rate are effectively solved.

[0073] like Figure 4As 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.

[0074] 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.

[0075] The processing unit executes the various methods and processes described above, such as method steps S01 to S06. For example, in some embodiments, method steps S01 to S06 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 S06 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute method steps S01 to S06 by any other suitable means (e.g., by means of firmware).

[0076] 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.

[0077] 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.

[0078] 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.

[0079] 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.

[0080] 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 process parameters of new energy power battery casing based on RBF, characterized in that, The method includes: Step S01: Determine the stamping speed, blank holder force, friction coefficient and die clearance as the process parameters to be optimized and construct the design space. Use optimized Latin hypercube sampling to obtain the initial process parameter combination matrix. Step S02: Input the initial process parameter combination matrix into the finite element analysis model to perform multi-pass thinning and deep drawing simulation, and extract the maximum thinning rate and the maximum principal strain; Step S03: Construct an RBF proxy model based on the process parameters and the corresponding maximum thinning rate and maximum principal strain, and obtain an RBF proxy model that meets the accuracy requirements through accuracy acceptance and dynamic point addition; Step S04: Using the maximum thinning rate and maximum principal strain as optimization objectives, and combining forming quality constraints and process parameter boundaries, a multi-objective constrained optimization model is constructed; Step S05: Based on the RBF surrogate model that meets the accuracy requirements, the NSGA-III algorithm with reference point mechanism is used to iteratively optimize the multi-objective constraint optimization model to obtain the Pareto optimal process parameter solution set; Step S06: Based on TOPSIS, perform a comprehensive decision on the Pareto optimal process parameter solution set to determine the optimal combination of process parameters.

2. The method for optimizing process parameters of new energy power battery casing based on RBF according to claim 1, characterized in that, In step S01, the design space is sampled using optimized Latin hypercube sampling, and the initial process parameter combination matrix is ​​obtained by optimizing the spatial distribution of the sample points.

3. The method for optimizing process parameters of new energy power battery casing based on RBF according to claim 1, characterized in that, In step S02, the finite element analysis model is used to perform multi-pass thinning and deep drawing numerical simulations on each process parameter combination in the initial process parameter combination matrix, and the maximum thinning rate and maximum principal strain corresponding to each process parameter combination are extracted from the simulation results after forming is completed.

4. The method for optimizing the process parameters of a new energy power battery casing based on RBF according to claim 1, characterized in that, The RBF proxy model described in step S03 takes the stamping speed, blank holder force, friction coefficient and die clearance as inputs, and the maximum thinning rate and maximum principal strain as outputs, and uses the coefficient of determination and preset acceptance threshold for accuracy acceptance.

5. The method for optimizing process parameters of new energy power battery casing based on RBF according to claim 4, characterized in that, When the accuracy of the RBF proxy model does not reach the preset acceptance threshold, dynamic points are added in the design space region corresponding to the prediction error to supplement finite element simulation samples and update the RBF proxy model until the RBF proxy model reaches the preset acceptance threshold.

6. The method for optimizing process parameters of new energy power battery casing based on RBF according to claim 1, characterized in that, The multi-objective constraint optimization model mentioned in step S04 is as follows: taking the maximum thinning rate and the maximum principal strain as dual optimization objectives, and minimizing the maximum thinning rate and the maximum principal strain respectively; the forming quality state constraints include the maximum thinning rate not exceeding 20% ​​and the maximum thickness increase rate not exceeding 10%; the process parameter boundary constraints include the stamping speed, blank holder force, friction coefficient and die clearance being within their respective preset value ranges.

7. The method for optimizing process parameters of new energy power battery casing based on RBF according to claim 1, characterized in that, The specific steps of step S05 are as follows: the maximum thinning rate and maximum principal strain are predicted for the individuals in the NSGA-III algorithm population using the RBF surrogate model; offspring populations are generated through crossover and mutation; the parent and offspring populations are subjected to non-dominated sorting under constraint; when the number of individuals in the critical non-dominated layer exceeds the capacity of the next generation population, individuals in the critical non-dominated layer are screened based on reference point association and niche counting, and the next generation population is formed according to the screening results; the iteration terminates when the preset maximum number of iterations is reached or the preset population convergence condition is met, and the process parameter combination in the final non-dominated front is used as the Pareto optimal process parameter solution set; otherwise, the next generation iteration continues.

8. The method for optimizing process parameters of new energy power battery casing based on RBF according to claim 1, characterized in that, Step S06 specifically includes: Establish a decision matrix for the Pareto optimal process parameter solution set; The decision matrix is ​​dimensionless to determine the positive and negative ideal solutions; Calculate the weighted distances between each candidate solution in the Pareto optimal process parameter solution set and the positive ideal solution and the negative ideal solution, respectively; The overall proximity of each candidate solution is calculated based on the weighted distance, and the candidate solution with the highest overall proximity is selected as the optimal combination of process parameters.

9. 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 8.

10. 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 8.

Images