Online prediction of fatigue remaining useful life for alloy equipment based on EM algorithm and nonlinear Wiener process

Abstract Under prolonged fatigue loading, fatigue cracks tend to initiate in stress-concentration areas of alloy equipment. The early-stage crack propagation behavior, influenced by the coupling effects of internal stresses and external environmental factors, exhibits highly complex stochastic degradation characteristics. This process may ultimately lead to structural fracture failures, posing significant safety hazards. To prevent sudden failures and enhance operational safety and reliability, implementing precise fatigue remaining useful life (RUL) prediction during early degradation stages for proactive maintenance holds significant engineering importance. However, traditional failure-mechanism-based RUL prediction methods struggle to accurately characterize such intricate degradation processes and associated measurement uncertainties, resulting in limited prediction accuracy. To address this challenge, this study proposes a nonlinear Wiener process-based fatigue RUL method that considers various degradation scenarios while incorporating measurement errors, combining the flexibility of the Wiener stochastic process according to the crack degradation characteristics of alloy equipment. This approach effectively captures the complex stochasticity inherent in crack evolution. Firstly, based on the basic properties of the Wiener process and the concept of first hitting time (FHT), the probability density function (PDF) expressions of the lifetime and RUL corresponding to the degradation model are given in the paper. For precise parameter identification and dynamic updating, the Expectation-Maximization (EM) algorithm is introduced for parameter estimation, combined with Bayesian theory for online parameter updating. Finally, the feasibility of the proposed method is validated using actual fatigue crack degradation data from 2017-T4 aluminum alloy and platinum alloy equipment. Comparative results demonstrate the novelty and superiority of the presented fatigue RUL prediction approach over existing methods.