Prediction of vehicle sliding door system parameters using optimization techniques and neural networks

This work describes an application of using the artificial neural network and Bayesian optimization based multi-objective optimization to predict the design parameters of vehicle sliding door system. Artificial neural network is used to solve complex and uncertain models. It gives effective results in black box problems that have difficulties in solving analytically. It is also a suitable approach for solving problems that involve uncertainties but require costly physical test or long run simulations. Artificial neural network and Bayesian optimisation were used in the prediction and optimization of the design parameters, since the sliding door design, which is considered within the scope of the study, is a complex problem with high uncertainties. After performing explicit dynamic analyses with the finite element method, the analysis results for different input values of the design parameters were predict using artificial neural network and Bayesian optimisation. Regression, artificial neural network, and Bayesian optimisation results are compared for prediction performance. Then, the optimal solution of the genetic algorithm (GA) for the multi-objective optimization problem was obtained. By eliminating long analysis times, a more flexible and faster method is presented.