Research On Lightweight Design Of Autobody Structure Using Robust And Reliability-based Design Optimization

Weight reduction of automobiles is a fundamental approach to realize fuel economyand environmental protection. The weight reduction in body structure plays a ratherimportant role in decreasing the weight of full vehicle, which results from the fact thatbody structure possesses about40%weight of full vehicle. In real engineering applications,the variation of gauge thicknesses and mechanical parameters of material (yield limit, ultratensile strength, Young’s modulus) caused by uncertainties including environmentalchange, manufacturing precision and individual difference of operators, affects directly thestructural performances of autobody structure. The lightweight design without consideringaforementioned variations may fail to meet the requirements of structural performances,which will lead to a lack of feasibility and reliability in a real engineering application. Thedissertation concentrates on the study of determining the parameters of the existingautobody structure topology, including sheet thicknesses and mechanical parameters ofmaterial, to achieve the lightweight design of autobody structure. Based on robust andreliability-based design optimization (RRBDO), the method for the lightweight design ofautobody structure, coupled with its engineering application, is investigated consideringthe variation of sheet thickness and mechanical parameters of material caused by variouskinds of uncertainties. The main research work and corresponding conclusions are listedas follows:(1) Comparative study of complex metamodeling techniques for approximatingcrashworthiness performance indicatorsThe lightweight design for autobody structure is generally considered as the designoptimization problem for a complex system, since autobody structure is characterized bystrong nonlinear structural performance responses and large number of random variables.Polynomial response surface frequently undergoes bad accuracy and efficiency when approximating the structural performance indicators. To overcome this difficulty, supportvector regression (SVR) that is a new technique in machine learning is introduced toapproximate the crashworthiness performance indicators of autobody structure. Thecomparative study of SVR with four commonly used complex metamodeling techniquesincluding moving least square (MLS), artificial neural network (ANN), Kriging (KG) andradial basis function (RBF) is performed. The comparison is investigated both inapproximation accuracy and computational efficiency. It is shown that SVR outperformsother metamodeling techniques considering both approximation accuracy and efficiency.Consequently, SVR will be used in the succeeding study work to approximate structuralperformance indicators of autobody structure.(2) Study on SVR-based adaptive metamodeling method for approximatingperformance indicators of autobody structureThe improvement in prediction accuracy of metamodels at the design optimum willhelp obtain a reliable and feasible design scheme in real engineering applications. Toachieve this goal, the adaptive metamodeling method is studied based on the idea ofverifying the design optimum by using the finite element analysis. The SVR-basedadaptive metamodeling method is proposed and the framework and steps of the methodincluding verification and criterion are introduced in detail. The study case of lightweightdesign for frontal autobody structure is given and the surrogates of crashworthinessperformance indicators are constructed by using the proposed SVR-based adaptivemetamodeling method. The optimum design verified by the finite element simulation isfinally obtained, which makes the lightweight scheme more reliable in the real engineeringapplication. In this method, the design optimum is verified by using finite elementsimulation and then the simulation results of the design optimum will be fed back as theinformation of a sampling point that will be used if necessary to reconstruct themetamodels of structural performance indicators. It is shown that the prediction accuracyof metamodels at the design optimum is effectively improved and guaranteed, whichprovides a high-fidelity metamodeling method for RRBDO of autobody structure.(3) Study on sequential optimization strategy of RRBDO based on MCSIn robust and reliability-based design optimization (RRBDO), traditional double loop(DL) strategy is frequently used but the weakness in solving efficiency and accuracy isusually aggravated for a large-scale engineering application. In order to tackle the problem, sequential optimization strategy is investigated based on decoupling the process ofoptimization and reliability analysis. The MCS-based sequential optimization strategy isproposed and the design steps are introduced in detail. The adaptive constraint shiftingmethod is created, which helps to realize the constraint boundary shifting for thedeterministic optimization process. The proposed strategy is verified by using classicalmathematical and engineering cases. By using the MCS-based sequential optimizationstrategy, both the unstability and multi-MPP problems of DL strategy are overcome. Whilethe accuracy for the reliability analysis is guaranteed and the solution efficiency is greatlyimproved. Consequently, the MCS-based sequential optimization strategy is suitable forbeing used in the RRBDO of complex system, which lays the foundation for thesucceeding study on the lightweight design method and corresponding engineeringapplications.(4) Study on lightweight design method of autobody structure based on RRBDOAccording to the particularity of the lightweight design for autobody structure, thelightweight design method of autobody structure based on RRBDO is investigated bycombining the SVR-based adaptive metamodeling method and MCS-based sequentialoptimization strategy. The framework of RRBDO-based method for autobody lightweightdesign is proposed and the detailed introduction to the design steps is given. Anapplication case of the lightweight design for full autobody structure is investigated andthe reliable lightweight scheme is obtained and applied in real engineering application,which verifies the feasibility of the proposed RRBDO-based method for autobodylightweight design. By using the RRBDO-based method for autobody lightweight design,the lightweight scheme is more reliable and the design efficiency is effectively improved.Furthermore, a software platform is developed based on the proposed method, which aimsat providing an operable and convenient tool for being used in automotive engineeringapplications.