Equipment Health Monitoring with Non-Parametric Statistics for Online Early Detection and Scoring of Degradation

This paper develops a health monitoring scheme to detect and trend degradation in dynamic systems that are characterised by multiple parameter time-series data. The presented scheme provides early detection of degradation and ability to score its significance in order to inform maintenance planning and consequently reduce disruption. Non-parametric statistics are proposed to provide this early detection and scoring. The non-parametric statistics approximate the data distribution for a sliding time window, with the change in distribution is indicated using the two-sample Kolmogorov-Smirnov test. Trending the changes to the signal distribution is shown to provide diagnostic capabilities, with deviations indicating the precursors to failure. The paper applies the equipment health monitoring scheme to address the growing concerns for future gas turbine fuel metering valve availability. The fuel metering unit within a gas turbine is a complex electro-mechanical system, failures of which can be a major source of airline disruption. The application is performed on data acquired from a series of industrial tests performed on large civil aero-engine fuel metering units subjected to varying levels of contaminant. The data exhibits characteristics of degradation, which are identified and trended by the equipment health monitoring scheme presented in this paper.