Machinery condition prognosis using multivariate analysis

Abstract

Condition prognosis is an essential element for predicting the health of physical assets. Prognosis enables a projection of asset condition progression from the past and current into the future, thus providing a significant input into the asset related decision making process. Of different prognosis technologies, signal or feature based methods make predictive decisions according to progressive condition information and have been one of the focuses of recent research in the area. The success of these methods largely relies on effective feature extraction and feature manipulation. This paper proposes the use of eigenvector analysis for machinery condition trending and prognosis which incorporates multiple variables. An eigenvector analysis method, Principal Component Analysis (PCA), is selected to handle the features extracted from both time and frequency domains of vibration signals. The derived composite features facilitate the visualization of condition progression from different dimensional views, hence assisting condition prognosis. Control limits of the composite features are also designed for triggering condition alarms. The proposed method is verified using data from pump condition monitoring during flow fluctuations.