Predicting the ride comfort of passenger cars for single-obstacle crossings through an innovative feature extraction method consisting of non-linear geometric approximations of wavelet-transformed acceleration data and airborne sound

Abstract

In this paper a method is presented to predict the ride comfort of passenger cars for single-obstacle crossings based on measured acceleration data and airborne sound. The method takes advantage of applying the continuous complex wavelet transform to the signals using specially adapted Gabor wavelets. Through an innovative approach, the comfort-relevant features are extracted to describe the geometric properties of the transformed data and to predict the ride comfort using artificial neural networks with a feedforward structure. In order to avoid overfitting, basic data division is applied to the available training data and the networks are trained using Bayesian regulation.