Discrete wavelet transform–neural network crack-like damage identification of welds on chassis cross members

Abstract

This paper incorporates a discrete wavelet transform and a radial basis function neural network to implement a vibration-based crack-like damage identification approach to the simulated welds of chassis cross members. After finite element method modelling of a vehicle, eight acceleration sensors are placed on the chassis, and their responses are measured to the upward displacement on the excitation of two tyres while the other two tyres are fixed. Then, a discrete wavelet transform with a Daubechies 3 mother wavelet is applied to the responses of the sensors. All detail coefficients and approximation coefficients of the nine levels of the discrete wavelet transform are introduced into the radial basis function neural network, in order to extract the underlying features for assessment and localization of damage. This approach successfully assesses the existence of damage and the localization of various lengths of damage in numerically simulated vehicle chassis welds, with respect to the fact that the complete model of the vehicle is a highly complicated structure and its vibration responses are extremely non-linear and sophisticated.