A Data Fusion-Based Methodology of Constructing Health Indicators for Anomaly Detection and Prognostics

Abstract

Anomaly detection and Remaining Useful Life (RUL) prediction are the most significant components of Prognostics and Health Management. A typical workflow is to extract features or construct health indicators by sensor fusion of the equipment and then perform anomaly detection and RUL prediction according to the health indicators. In this paper, we present a data-fusion based methodology for constructing two composite health indicators through integrating multiple run-to-failure sensor data towards to the anomaly detection and the RUL prediction, respectively. A novel optimization methodology termed as the Genetic Algorithm is proposed for constructing the composite health indicators, which are capable of presenting a better reflection of the health condition of systems. This optimization methodology makes the fusion of multiple sensor data no longer limited in linear fusion. For this algorithm, the property of a health indicator is the fitness and the different fusion methods to produce health indicators are individuals of a population. Our methodology was verified by applying to a degradation dataset of an aircraft gas turbine engine that was generated by the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS). The result shows that the extracted feature have better performance than using original data in the anomaly detection and the RUL prediction.