Research On Surrogate Model Method And Its Application In Lightweight High-Speed EMU Carbody

The design and development of high-speed rail should be supported by abundant data,and multiple performances,e.g.mass,static strength,stiffness,modal,fatigue strength and collision,should be taken into consideration in the structural design of EMU.If we analyze the structural design with only traditional technologies like CAD and CAE,there will not only be high calculation costs,but it will also be difficult to meet the structural design requirements because of the performances that restrict and influence each other.Against such backdrop,Computer Aided Optimization(CAO)came into being.Integrating a variety of strategies and methods to meet the multidisciplinary requirements of product design,CAO technologies are applicable to many fields,and surrogate model is one of the technologies.The surrogate model replaces the finite element model to reduce the computational expense and improve the design efficiency,but there are still problems like difficult sampling and insufficient fitting accuracy in engineering practice.In this paper,based on DOE of sampling,the construction methods of different surrogate models are studied and applied to the lightweight design of the head car body of EMU.This paper can be roughly divided into the following aspects:(1)Through investigating relevant literature,this paper clarifies the function and significance of optimized structural design and studies the application of commonly used surrogate models in optimized design based on relevant basic principles,laying the foundation for the follow-up research.(2)This paper studies the modeling principles,methods and commonly used error analysis indexes of three types of single surrogate models,and then with four functions as test examples,analyzes and discusses the performance indexes and applicability of the three surrogate models under different nonlinearity degrees,different dimensions and different number of samples.Secondly,in view of the insufficient accuracy of the single model,this paper explores the prediction ability,stability and applicability of three linear weighted combined surrogate models based on cross-validation mean square error.Besides,the accuracy error caused by the limitations of cross-validation is analyzed,and a comparative analysis is made between the three single surrogate models and the three combined surrogate models,providing methodological guidance for subsequent design of lightweight car body.(3)The car body of TC01 standardized EMU is properly simplified to establish the finite element model of the car body.According to the EN 12663 standard,24 kinds of combined working conditions are applied to analyze the structure of car body in terms of strength,stiffness,modal and fatigue,thereby analyzing relevant performances of the car body.On the basis of meeting various performance indexes,this study provides data support for the followup research on lightweight car body.(4)On the premise of meeting the performance requirements,with the thickness of each structural section of the passenger chamber of TC01 train as the design variable,this paper determines six responses,namely the bending stiffness,torsional stiffness,maximum Von.Mises stress,first-order vertical bending modal frequency,the first principal stress of the fatigue position and the mass of the car body.After analyzing the sensitivity of each variable to each response and the interaction effect between each variable,then analyze the degree of nonlinearity between the above 6 responses and factors,and establish a surrogate y model suitable for each response.The NSGA-II algorithm is applied to conduct multi-objective optimization on the structure of the car body,and great results are obtained.The optimization results gained by the surrogate models are substituted into the finite element model,and the final errors are all less than 10%,which verifies the feasibility of the surrogate model.