Calculating Remaining Useful Life in an Embedded System

Abstract

This paper presents a prognostic algorithm with low computational requirements that was implemented on an embedded system. The health of components such as shafts, bearings, and gears, is estimated based on a paradigm where condition indicators (CIs) are mapped into a health indicator (HI) for detecting and identifying faults based on vibration data. For estimating the time evolution of the HI along with remaining useful life (RUL) and its evolution, an alpha-beta and alpha-beta-gamma trackers are prosed. The estimator assumes that the plant noise is converging to a steady state over time. The advantage of the proposed filters is considerably less memory usage and computational load than a Kalman filter. The efficiency of the proposed method is demonstrated with a known fault case based on real-world data. The demonstration shows that the state observer is capable of tracking the evolution of the HI and estimates the RUL with sufficient robustness and performs favorably to previous results based on the Kalman filter. The method is well suited to computing on lower cost smart sensors.