Research On Offline-online Joint Optimization Algorithm Based On Surrogate And Its Application In Sheet Metal Forming

In the process of engineering research,many problems are often costly,such as time-consuming computational fluid dynamics simulation,problems based on practical experiments,large-scale finite element simulation and complex structure design,etc.Optimization involving these problems is called expensive optimization problems.The optimization of metal sheet forming parameters is a typical expensive optimization problem,appropriate forming parameters can avoid the occurrence of various defects in the forming process and improve the forming quality.However,due to the large deformation characteristics of the stamping process and the possibility of wrinkling,cracking,and springback,related experiments or simulations are very expensive.Traditional optimization algorithm is difficult to solve such expensive problems.Recent years,Surrogate-assisted evolutionary algorithms(SAEAs)have been widely used for expensive optimization.However,most SAEAs are still need to conduct thousands of simulations or experiments to collect data and actually difficult to solve high-dimensional and complex problems.This paper focuses on improving the efficiency and expanding the universality of algorithm,and proposes an offline-online joint optimization algorithm based on surrogate.This algorithm enhances the ability to solve complex high-dimensional problems,and combines offline optimization and online optimization.The first-stage optimization can be independently conducted based on offline data,and the secondstage optimization can be performed using online generated data,this offline-online combined optimization framework enables the algorithm to solve the problem that only historical data is available in engineering optimization.In the optimization process,the algorithm uses the offline updated global surrogate to optimize and narrow the search range,and then performs data-enhancement based on the first-stage results,and trains a more accurate local surrogate to perform the second-stage optimization to obtain the final solution.In order to make full use of limited offline data,in the first stage optimization,the offline global surrogate update strategy is proposed,which can dynamically delete low-value sample points to reduce the impact on the accuracy of the search area surrogate.In order to build a high-precision surrogate,in the second stage optimization,the promising area cluster infill strategy is proposed to generate a high-density local data set.Based on experiments,the proposed algorithm is analyzed and evaluated.Firstly,comparative experiment is conducted to verify the effectiveness of mechanisms in the algorithm.Then,the performance of the proposed algorithm is compared with the stateof-the-art SAEAs to prove its competitiveness among similar algorithms,and the experiment under the same optimization precision is conducted to prove that the proposed algorithm has huge advantages over traditional evolutionary algorithms in terms of computational cost.Finally,the algorithm is used to optimize the forming parameters of three actual automobile stamping parts,and the optimized parameters significantly improve the forming quality of the stamping parts.