Dynamic monitoring using hidden Markov regression model for predicting remaining useful life

<p dir="ltr">This paper presents the remaining useful life (RUL) prediction problem in civil engineering applications using a hidden Markov regression model (HMRM), as a promising approach for model-based degradation. Unlike self-transition hidden Markov models for mass-produced components, where prior lifetime signals are available to estimate state information, the proposed HMRM formulates the conditional probability of RUL in terms of the estimated regressor parameters, after temporally fitting the damage model. The discrete property of state in HMRM makes it possible to handle heterogeneities in the degradation process. The HMRM can also synthesise multiple signals by adopting a decision-level fusion. An adaptive closed-form solution for RUL prediction is presented. The performance of HMRM is demonstrated on synthetic measurements and compared with a Bayesian extended Kalman filter (EKF) updating technique.</p>