Online health status monitoring with adaptive first prediction time determination for bearing degradation trend prognosis

Online health status monitoring of rolling bearings is paramount for ensuring the uninterrupted reliability of industrial systems. A pivotal challenge in this process is the accurate and automatic determination of the First Prediction Time (FPT), especially in the presence of complex noise. This paper proposes an online health status monitoring method with adaptive FPT determination for bearing degradation prediction. First, to address data anomalies that occur during the healthy stage and may interfere with FPT judgment, a dual-layer sliding window structure is developed. This structure detects spurious outliers and replaces them with predictions from an AutoRegressive Moving Average (ARMA) model built on local data, ensuring a clean signal. Second, an adaptive cross-validation FPT detection mechanism is introduced to enable the accurate identification of the initial degradation point from the monitored data stream. Moreover, a warm-up strategy is implemented for the online AutoRegressive Integrated Moving Average (ARIMA) algorithm, making it more suitable for degradation trend prognosis in industrial applications. The effectiveness and generalizability of the proposed method are rigorously validated using the IEEE PHM 2012 prognostic challenge dataset and the XJTU-SY bearing dataset.