Abstract
Rolling pairs are widely used in high-value equipment, making remaining useful life (RUL) prediction essential for reliability and safety. However, RUL prediction remains challenging due to complex wear mechanisms, scarce data, and insufficient representation of operating history. A hybrid method combining a Gamma process and a backpropagation (BP) neural network is proposed. Monte Carlo samples generated from Gamma and inverse power-law models fitted to accelerated-test data are used to augment the training data, while a history matrix represents time-varying conditions. The proposed method is evaluated using wear data from an enclosed, low-speed, heavy-load rotating component in a nuclear power plant. The results show that the proposed method achieves high accuracy in wear prediction and produces a continuous RUL trajectory under time-varying operating conditions. A continuous RUL trajectory is also produced under time-varying operating conditions. The method provides a practical framework for rolling pairs with limited degradation data and variable operating conditions.