Nominal state determination and its effect on remaining useful life prediction

Abstract

Abstract In machinery operation, a prolonged healthy or nominal state often lacks prognostic significance, causing challenges like data overload, biased predictions, and complex models. Moreover, many prediction methods utilize the complete history of monitoring data from the machine’s startup to its failure; however, prognostics mostly relies on data from the degradation stage. To address this, this study proposes a method to identify and exclude the prolonged period of the nominal state, thereby enhancing the prediction performance of remaining useful life (RUL). A health index (HI) is formulated by integrating acceleration signals from multiple time windows, with deviations computed as the disparity between the HI and its root mean squares. The identification of start and end times for the nominal state, determined by the intersection of consecutive deviation curves, leads to its exclusion from degradation behaviour modelling. The utilization of polynomial degradation trends from HI data after the nominal state’s end time, incorporating a positive slope constraint, aids in mitigating extrapolation uncertainty. The method’s efficiency is demonstrated in three defect cases, highlighting improved RUL predictions without the nominal state’s inclusion.