A New Generalised Approach for Feature Representation in Vibroacoustical Fault Diagnosis

Abstract

A new generic approach has been proposed for those cases in which oneor multidimensional Fourier transforms are used for health monitoring and pattern recognition. The approach consists of using two new monitoring features simultaneously: the real and the imaginary components of the Fourier transform. It was shown that the approach is more general than the power spectral density (PSD), phase spectrum, and Hartley approaches and is more effective than the PSD approach. This is in contrast to other condition monitoring and pattern recognition applications, including Fourier optics applications, where the PSD or phase spectrum are used. However, the approach has been proposed and investigated only for one particular case, without taking into account the covariance between new monitoring features. Taking covariance between monitoring features into account may improve the monitoring effectiveness. Therefore to generalise the proposed approach, we need to take into account the covariance between the features, i.e., Fourier components. Application of the generalised approach to a particular health monitoring task was not undertaken as well. The purposes of this technical note are: ! to generalise the approach and estimate the effect of generalisation; ! to estimate and compare the monitoring effectiveness of the generalised approach with the monitoring effectiveness of the PSD approach; and ! to employ the generalised approach for the acoustical monitoring of damping. 2. THEORETICAL ANALYSIS