Lightweight Design Of Thin-wall Structure Based On Machine Learning And Intelligent Optimization

With the development of automobile technology and the increasingly strict regulations on energy conservation and emission reduction,the demand for lightweight in the era of electrification is more urgent.As an important way to achieve automotive lightweighting,traditional body structure optimization design mainly relies on intelligent optimization algorithm iterative optimization,which requires a large number of simulation calculations,with low efficiency and long design cycle.The surrogate modeling technique provides an effective solution to shorten the design cycle by collecting simulation sample points to construct an approximate mathematical model to fit the complex relationship between design variables and response.However,the surrogate models commonly used in the field of structural engineering have weak generalization ability and poor fitting for nonlinear,high-dimensional problems.Based on this,this paper uses machine learning to build the surrogate model.Even under the condition of small sample data set,machine learning can overcome the above problems and effectively improve the accuracy and reliability of the surrogate model.The lightweighting of thin-walled structures,typical of bodywork,is key to improving the performance and competitiveness of delivery vehicles.Therefore,this paper takes the thin-walled body structure as the research object,uses the machine learning algorithm to expand the model library of the surrogate model technology,and adopts the design mode of"machine learning+heuristic optimization algorithm"to realize the lightweight design of multi-dimensional size parameters of the thin-walled structure.The details of the work are as follows:(1)Aiming at the problems of weak generalization ability and low fitting accuracy of traditional surrogate model,the thin-walled surrogate model is constructed by machine learning algorithm,which can effectively improve the accuracy and interpretability of the surrogate model.First,the dataset is constructed by sampling technique and joint simulation;second,the automatic tuning of the hyperparameters of the machine learning model is achieved by Bayesian optimization to further improve the accuracy of the surrogate model.Then,the accuracy of the surrogate models was compared by 5-fold cross-validation,and it was found that for the three structures of frame,hood and cab,the powerful machine learning models(XGBoost,Light GBM and Deep Forest)achieved higher accuracy,with R~2 coefficients of determination over 0.94for mass,stiffness and first-order modes,proving that machine learning as a surrogate model has obvious advantages.Finally,the SHAP method was used to interpret the model using machine learning to explore the effect of each design variable on the mechanical properties.(2)A mathematical model for the lightweighting of thin-walled body structures is established based on the surrogate model,and the lightweighting problem is transformed into a single-objective optimization problem with constraints.To address the problems of traditional metaheuristic optimization algorithms with stagnation,suboptimal selectivity and behavioral consistency defects,a reinforcement learning mechanism is introduced and a reinforcement learning-based gray wolf optimization algorithm(RLGWO)is proposed.Then,the RLGWO algorithm is used to solve the mathematical model of thin-walled body structure lightweighting,and the convergence speed,optimization effect and stability of the RLGWO algorithm are proved to be optimal by comparing with the experimental results of four other metaheuristic algorithms.Finally,the dimension parameters optimized by RLGWO algorithm are brought into Abqus for verification.The results show that the stiffness and modal performance of the frame,hood and cab meet the corresponding constraint requirements,and the mass decreases by 0.47%,6.05%and 8.72%respectively compared with the base value.This proves the effectiveness of RLGWO algorithm in lightweight design of thin-wall body structure.(3)Based on Python and PyQt toolkit,the thin-wall structure parameter optimization system was designed and developed.The system consists of four functional modules:co-simulation model,data processing,predictive analysis and global optimization.And use the signal and slot mechanism in PyQt to develop the user interface of the system,and further improve the use efficiency of the software.