Remaining Useful Life Prediction for Hybrid Stochastic Degradation Processes with Finite Time Delay

Advanced prognostic technologies gradually become one of the most powerful tools in the field of process monitoring and control to ensure the operational security of modern industrial systems like large blast furnaces and generator sets. Degradation processes of long-lifetime equipment generally possess slow time-varying characteristics, leading to difficulties in analyzing memory effects among the entire life-cycle. Under the influence of multisource stochastic factors including external environments, internal conductions, and information transmissions, there may exist certain time delay in the condition monitoring data. To overcome these limitations of traditional data-driven prognostic methods, we construct a hybrid degradation model by introducing two layers of state dependencies as well as finite time delay. An additive combination of Markovian and non-Markovian diffusions is also applied to extend the scope of applications in complex processes. The unknown parameters are estimated by a wavelet estimator and a maximum likelihood (ML) algorithm, and then the distribution of remaining useful life (RUL) can be derived analytically on the basis of a mild transformation. A simulation study fully examines the advantages of the proposed theoretical scheme.