Skip to Main Content
Automated Condition Monitoring of machines typically involves the detection and diagnosis of developing defects. However increased demand for reliability requires these systems to also predict the Remaining Useful life (RUL) of the machine in order to schedule timely maintenance and reduce machine downtime by preventing catastrophic faults. In this paper, an automated approach to degradation analysis is proposed that uses the acoustic noise signal from a rotating machine to determine the RUL. We incorporate a novel approach to Feature Subset Selection to extract relevant features for classification. This method is applied to the high dimensionality multivariate time series data extracted from the acoustic data acquired over the lifetime of the fan to reduce confounds from redundant features as well as improve computational efficiency. Our approach requires no a-priori information regarding the spectral location of defects or class label boundaries of the training data. Using such an approach, the RUL of the machine was determined with an accuracy of 98.7%.