Alternative Multivariate Methods for State Estimation, Anomaly Detection, and Prognostics

Abstract

To secure operational readiness of components, equipment, subsystems and systems and to assure successful job completion, appropriate monitoring, inspection and preventive maintenance, repair and replacement strategies are needed. Such requires suitable sensors and measurement approaches serving continuous monitoring of key operational parameters, aiming at discovering anomalies and assessing degradation levels, State of Health (SoH) and Remaining Useful Life (RUL) of any critical component involved. Serving this purpose, multivariate methods are important tools to analyzing multiple data sequences, providing means to compare actual measurement data against data representing a healthy system and making qualified assessments, typically based on measuring the distance between the actual system and the healthy system. The Multivariate State Estimation Technique (MSET) uses the least squares approach, the Auto-Associative Kernel Regression (AAKR) method uses the nonparametric Kernel estimation procedure, while the usage of the Mahalanobis distance is based on the covariance matrix of the different measured parameters. These methods are all based on specially selected distance definitions. In this paper, several extensions and variants of these procedures, yielding alternative measures, are introduced, analyzed and examined with focus on their advantages and disadvantages. Possible application areas are also outlined.