Study Of Data Driven Dominated Crashworthiness Analysis Method For Rail Vehicle Energy Absorbing Structures

Limited by the time cost,economic cost and complexity,the traditional vehicle structure crashworthiness design and analysis methods have gradually failed to meet the current requirements of shortening the product development cycle,tracking product information in real time,and quickly providing high-precision solutions.This study proposes a datadriven approach to the development of energy-absorbing structures,forming an energy absorbing structure design,analysis and optimization framework with machine learning as the core,including PerformanceIdentified Structure Concept Design,Indicator quantificated Initial Scheme Design,Process Visualized Response Quick Solve,and Unstable Deformation Cleared Accuracy Improvement Optimization.The universal domain data construction,energy absorption indicators prediction,process responses prediction and unstable deformation cleared optimization of the train radial deformation energy-absorbing structure instance are carried out.The main research contents are as follows:(1)A universal domain data construction method based on the characterization of mechanical properties of energy absorbing structures is proposed.The geometric and energy absorption evaluation indicators of the radially deformed energy absorbing circular tube instance are defined,and experimentally validated data-generating finite element model is established.The effects of four parameters,wall thickness t,expansion/shrink angle α,die radius r and friction coefficient u on the energy absorption and deformation of the radially deformed circular tube are characterized under a fixed scale condition,and then the effective range of each parameter is determined.The universal dimensionless geometric parameters,material parameters and contact parameters are further defined,and the universal design domain is constructed through the parameter range mapping under the fixed scale.Finally,sampling in the universal design domain and submitting to the finite element calculation can generate the universal domain data set of the radial deformed circular tube.(2)A prediction and explicit representation method of structural energy absorption indicators based on machine learning is proposed.The machine learning method is used to predict the three energy absorption indicators including 1 categorical response(deformation mode)and 2numerical responses(platform mean force and specific energy absorption)for a radial deformed circular tube instance.The importance order of variables for each response is determined,based on which explicit representation methods for categorical and numerical responses are proposed respectively,and then the explicit expression of the deformation mode boundary hyperplane and the empirical formulas of platform mean force and specific energy absorption are established.(3)Prediction models for two process responses,namely,deformation and force-displacement,are established based on the Point Net self-encoder and LSTM self-encoder.First,two different deep learning auto-encoders are used to establish the reconstruction frameworks based on point cloud data and time series data for deformation and force-displacement,respectively.Then the prediction models of the two process responses are obtained by establishing a neural network model between the input variables and the intermediate latent variables generated in the reconstruction frameworks.The shrink tube example verifies that the two of process response prediction models have high accuracy,and the speed of process responses solution and visualization is greatly improved compared with the finite element simulation,which can improve the efficiency of structural crashworthiness analysis.(4)A machine learning-based optimization accuracy improvement method for energy-absorbing structures is constructed.Aiming at the problem that the existing optimization method cannot deal with the categorical response causesing unstable deformation modes to appear in the optimization solution,a machine learning-based unstable deformation cleared optimization method is established.The method adds the categorical response,i.e.,deformation mode,as the constraint condition,so that restrictsing the algorithm to search only in the effective design space,which is obviously different from the existing numerical optimization method that can only deal with numerical responses and cannot accurately define the effective design domain.Two cases of two-dimensional shrink tube and three-dimensional circumferential corrugated tube are used to verify the effectiveness of this optimization method.The results confirm that the constructed optimization method can remove the unstable deformation mode in the optimization solution and effectively improve the optimization accuracy of energy-absorbing structures.(5)The application and verification of the proposed data-driven energy-absorbing structure research and development method is carried out.The proposed wide-area four-step design and analysis method of energyabsorbing structure is applied to the research and development of the energy-absorbing structure of train coupler,and an expansion-shrink combined energy-absorbing structure that meets the requirements of the coupler and has better energy-absorbing efficiency is designed.The established machine learning prediction models are used efficiently and accurately complete the initial parameter determination,the energy absorption parametric study and the energy absorption multi-objective optimization of the new structure,which significantly improves the efficiency and accuracy of product design.This thesis proposes a set of data driven crashworthiness analysis methods for energy absorbing structures from data construction,response prediction,optimization and other aspects,realizing the improvement of the efficiency and accuracy of energy absorbing structure product design and analysis.