An Age-Dependent Hidden Semi-Markov Model With Multiple State-Dependent Aging Factors for Remaining Useful Life Prognosis

This paper addresses remaining useful life (RUL) prognosis for inhomogeneously deteriorating components whose degradation rates vary with both health state and age. To capture age-dependent degradation, an extension of the traditional hidden semi-Markov model (HSMM), referred to as the age-dependent HSMM with a global aging factor, has been introduced in prior work. However, a single global factor cannot adequately represent state-varying degradation rates, which are commonly observed in real-world datasets. To overcome this limitation, we extend the age-dependent HSMM by introducing multiple state-dependent aging factors. Each factor characterizes the degradation rate associated with a specific state, providing greater flexibility for modeling degradation processes that are non-homogeneous with respect to both age and state and enabling more accurate RUL prognosis. The proposed model is first demonstrated on simulation data generated via a Wiener process with distinct degradation patterns, and then validated on the XJTU-SY bearing dataset. Comparative studies against the hidden Markov model and the age-dependent HSMM with a global aging factor consistently show the superior performance of the proposed approach in both simulated and real-world scenarios.

Keywords: Age-dependent HSMM, state-dependent aging factors, remaining useful lifetime, non-homogeneous degradation, Wiener process, bearing dataset.