A modified gamma process for RUL prediction based on data with time-varying heavy-tailed distribution

Predicting remaining useful life (RUL) plays a critical role in condition-based maintenance (CBM). However, this task remains challenging as the collected data often has time-varying, non-homogenous properties (trend, variance) and exhibits non-Gaussian distributions. These aspects pose significant challenges to the classical approaches. To address these issues, we propose a modification of the gamma process (suitable for non-Gaussian data) to predict RUL for the degradation process with time-varying characteristics. Specifically, we use the Gompertz cumulative hazard function instead of the linear function used in the classical approach. The modified gamma process aims to outperform the classical one and other variants already used by exploiting its probabilistic nature to effectively manage the uncertainty in the degradation curve. In order to evaluate the effectiveness of the proposed approach, extensive experiments are performed on simulated data with both Gaussian and non-Gaussian distributions. In addition, the performance of the model is validated in real-world scenarios using two benchmark datasets. The results consistently demonstrate the effectiveness of the proposed approach for both simulated and real datasets.