A two-stage wiener process-based method for bearing remaining useful life prediction considering measurement uncertainty

Abstract To address the issue of significant prediction bias in remaining useful life (RUL) estimation caused by insufficient characterization of measurement uncertainty and stage-dependent degradation mechanisms in rolling bearings, this study proposes a two-stage Wiener process-based method incorporating measurement uncertainty. First, a two-stage Wiener process state-space model with change points is established, where stage-specific measurement noise terms are explicitly introduced in the observation equation to distinguish between degradation randomness and measurement uncertainty. Then, under the concept of the first passage time (FPT), the conditional probability density function and reliability function of RUL are derived for the two-stage model given the current latent state. On this basis, a parameter estimation and latent state updating strategy is developed by integrating change-point detection, maximum likelihood estimation, and Kalman filtering to jointly identify model parameters and the true degradation trajectory. The estimated results are subsequently substituted into the analytical FPT expressions to obtain the RUL distribution. Validation using the XJTU-SY bearing dataset demonstrates that the proposed approach achieves significantly higher prediction accuracy and robustness compared to single-stage models and methods that ignore measurement error. These results confirm its capability to deliver high-precision, low-uncertainty, and robust predictions under noisy conditions, providing an interpretable probabilistic modeling framework for RUL prediction in bearing health management.