Applying the design of experiment (DoF) to optimise the NN architecture in the car body design system

Neural Network (NN) architecture is very important part to establish the best performance of NN. As a consequence, a lot of investments have been done in this research area. This paper is going to show how the design of experiment (Taguchi method) selects the neural network parameters in car body design system. NN architectures included dealing ways with number of neurons, number of hidden layers, transfer functions, learning algorithms and factors interaction. The paper employed Genetic algorithm (GA) which is built in function of software to adjust learning rate, momentum, additive, multiplicative and smoothing. Finally, the NN modules will be used in car body design system to provide the information of external aerodynamic noise, aerodynamic vibration and fuel consumption factors for the user or car body designer.