Weight optimization and reliability prediction of an automobile torque arm subjected to cyclic loading

The goal of this thesis is to obtain the optimum shape of an automobile torque arm and predict its reliability. Torque arm, part of the rear suspension, is subjected to cyclic loading. For optimzation and reliability prediction, finite element analyses which have high accuracy level are used. Fatigue life is quite sensitive some parameters such as material properties, loadings etc. Therefore finite element analyses utilized for this study need to be verified by fatigue tests. First of all this verification is carried out by fatigue tests performed in TOBB ETU Technology Centre. Results of these tests show that the fatigue life of torque arm is between the low cycle and high cycle assumptions. Therefore for the solution method of finite element analysis both strain-life and stress-life method are investigated. Fatigue analyses show that the strain-life method gives better solution to torque arm problem. In light of the results of fatigue analyses which are performed by using ANSYS Workbench Fatigue Module, it can be said that fatigue test results and finite element analyses are consistent. Second part of this thesis is the weight optimization of the torque arm. Operating conditions and fatigue test conditions of torque arm are not same in this study. Optimization is performed under operating conditions. Surrogate models are used for optimization to reduce computational cost. Response surface method and Kriging model are investigated. Optimumn design of Response Surface model is lighter than Kriging model. Therefore optimum design is obtained by using response surface model. At the end of optimization, torque arm weigth is reduced by 28.6 percent. The last part of this study is the reliability prediction of optimum design. Reliability prediction is performed by using Monte Carlo simulations and tail modelling. Tail modeling uses generalized Pareto distribution for relibility prediction. The parameters of this distribution are found by using maximum likelihood estimation and least square regression. With Monte Carlo simulation reliability of optimum design is calculated as 0.9991. The reliability prediction of tail modeling with maximum likelihood estimation is 0.9995 and with least square regression is 0.9990.