Condition monitoring for the endurance test of automotive light assemblies

Abstract

In this paper, we describe a condition classification technique designed to detect fault occurrence in an automotive light assembly during endurance testing. Inputs to the classifier are features extracted from vibration measurement data. They contain time domain parameters and frequency band energy parameters calculated using wavelet packet transforms. A support vector machine with Gaussian radial basis function kernel is designed for multiclass classification. A multiplex parameter estimation is achieved by searching for a minimum bound of the support vector count to achieve structural risk minimization. Through experiments, we show that the combination of effective feature extraction and classification with good generalization capability allows the proposed condition-monitoring system to be accurate and reliable. Additionally, acoustic signals known to have low signal to noise ratio are used as tests. We show that with the proposed methodology, acoustic signals can be used with increased sensitivity and accuracy for condition-monitoring purposes.