Remaining Useful Life Prediction Based on Forward Intensity

Remaining useful life (RUL) prediction is an essential tool for enhancing the reliability and resilience of systems. Most existing RUL prediction methods consider degradation-induced failures, and assume there exists a fixed failure threshold for degradation signals. In many real applications, however, it is difficult or even impossible to find such failure thresholds, making the RUL prediction a challenging task. This study addresses the challenge by exploiting the method of forward intensity, a novel tool introduced in econometrics. In particular, the historical degradation signals are treated as time-varying covariates for the forward intensity. The dynamics of unobserved future degradation signals are captured by a time-varying regression parameter in the forward intensity function, which are usually estimated using the maximum pseudolikelihood method in existing studies. To improve prediction accuracy, this study proposes a new estimation method based on smoothing splines. The proposed method not only allows RUL prediction in the absence of a fixed failure threshold, but also achieves satisfactory prediction accuracy with reasonable interpretability of covariate effects. Prediction intervals for the RUL are also developed. The effectiveness of the proposed method is validated through Monte Carlo simulations and three real-data examples. The results show that the proposed method outperforms existing methods in terms of prediction accuracy.