Effective Fault Detection and CBM Based on Oil Data Modeling and DPCA

Abstract

This chapter presents two methodologies for effective equipment condition monitoring and condition-based maintenance (CBM) decision-making. The first method is based on multivariate modeling of data obtained from condition monitoring (CM data), dimensionality reduction using dynamic principal component analysis (DPCA), and constructing and using on-line a multivariate statistical process control (MSPC) chart based on the DPCA. The second method is based on vector autoregressive (VAR) modeling of CM data, DPCA, and building a proportional hazards (PH) decision model using the retained principal components as covariates. These methodologies are illustrated by an example using real oil data histories obtained from spectrometric analysis of heavy-hauler truck transmission oil samples taken at regular sampling epochs. The performances of the MSPC chart-based policy and the PH model-based optimal control limit policy are evaluated and compared with the traditional age-based policy.