Advanced multidimensional spectral analysis and its application for early fault detection

Abstract

For the early detection and diagnosis of a machine fault, it is necessary to identify the source that contains a fault signal, and to estimate the magnitude of the fault signal in the case of multiple input systems. However, most conventional research is based on the fast Fourier transform (FFT) method, so it requires entire time data and cannot be used for fault signals occurring instantaneously. Fourier-based methods are not well suited to the analysis of nonlinear or non-stationary systems owing to their time-varying nature. Thus, in this paper, a wavelet packet based technique that calculates the time-varying coherence functions for input/ output relationships is developed. Advanced multidimensional spectral analysis (MDSA) is introduced, and the proposed method analyses the signal instantaneously in both time and frequency domains. Introducing the instantaneous ordinary coherence function (IOCF), which is obtained from the wavelet packet analysis, it shows the possibility of ‘early fault detection’ by analysing signals instantaneously in time. Thus, by examining the trend of coherence functions, it is possible to find which signal contains the major fault signal and the degree to which the system is damaged.