A generalized approach for remaining useful life prediction considering the multi-source uncertainty of mechanical systems

Mechanical system remaining useful life (RUL) prediction accuracy is often influenced by uncertain factors such as individual differences, degradation volatility, and failure threshold (FT) randomness.Current age-and state-dependent Wiener process models (WPMs) are implemented by considering only the constant diffusion coefficient and fixed FT, which are difficult to apply in practical engineering.To address this problem, a generalized approach is proposed for RUL prediction by considering the multi-source uncertainty of mechanical systems.In this approach, an improved age-and state-dependent WPM is proposed by introducing a unit-to-unit variability-included diffusion coefficient, and the model parameters are estimated by a modified artificial bee colony algorithm-based maximum likelihood estimation.After that, the two unit-to-unit variability parameters are synchronously updated by using an adaptive Kalman filter.And then, the probability density function of RUL is approximated by the degradation process simulation method under the condition that the random FT follows a truncated normal distribution.The effectiveness of the proposed method is verified experimentally, and the results show better prediction accuracy than commonly used models, which offers an alternative solution for RUL prediction of mechanical systems.