PRINCIPAL COMPONENT ANALYSIS BASED APPROACH FOR HEALTH ASSESMENT OF ROLLING ELEMENT BEARING

An online rolling element bearing condition monitoring approach is developed using multivariate statistical techniques on vibration signal. The developed methodology uses multivariate analysis tool, Principal Component Analysis (PCA) as a data fusion technique for fusing Damage Identification Parameters (DIPs) into a single effective parameter. The DIPs from time domain, frequency domain and time-frequency domain are selected on the basis of monotonic behaviour with defect severity level. Two statistical analysis tools, K-Clustering technique & Z-statistic are modified and applied on identified relevant component for bearing diagnosis. Z-statistic approach based on Chebyshev's inequality is formulated considering bearing defects propagation and has been used to identify fault initiation and defect level change. K-clustering approach has been tested on principal component for fault level identification but has shown limitations due to the overlapping indicators on the cluster map. HFRT based DIPs are used for predicting leading defect type qualitatively. The developed tool has been tested and verified on simulated defect data and available seeded defect data. It has been applied to naturally progressed defect data generated using accelerated life test to check its effectiveness on real life situations.