Online prognostics based on multiple dependent degradation processes

For long-lifetime industrial equipments, the online prognostics can predict the remaining useful life (RUL) of the systems in real time, which further offers effective references for the maintenance schedule. Most existing literatures with respect to the issue of prognostics only deal with one single degradation process or simply use the Copula functions to model the correlation structures for multi-component systems in a statistical sense. In this paper, we mainly focus on the systems with multiple dependent degradation processes from practical perspectives, and then establish a new kind of Wiener process based degradation model in which the dependencies can be directly described by a two-part diffusion coefficient matrix. Following this model, an online RUL prediction method is presented based on the Monte Carlo algorithm. In particular, once the new condition monitoring data are available, the unknown parameters in the model can be updated by utilizing an expectation maximization (EM) algorithm. Meanwhile, the reliability function as well as the distribution of the RUL at the current monitoring time can be derived by simulating the first passage time of the corresponding degradation processes, according to the predetermined failure thresholds. Finally, the usefulness of our proposed method is validated by a numerical example in regard to bivariate dependent degradation processes.