摘要：AbstractPredicting the remaining useful life (RUL) of in-service commercial plants is closely linked with safety production and maintenance cost optimization. It is noteworthy that real-life degradation processes could be affected by their past states coupled with unknown disturbances. To make a comprehensive consideration of multisource dependencies and uncertainties, we develop a generalized non-stationary, nonlinear, and non-Markovian degradation model consisting of a Gaussian disturbed drift term and a sub-fractional Brownian motion (sub-FBM) based diffusion term. Both parts are state-dependent, and reflect two different types of memory effects. The main parameters and the approximate probability density function (PDF) of RUL can be solved on foundation of the Markovian transformation theories. A case study finally illustrates the effectiveness of the proposed scheme.
