Abstract

Accurate anticipation of the remaining useful life (RUL) of a machine is becoming mandatory for efficient exploitation of the asset and avoiding the unplanned downtimes. This should be achieved without extra investments in additional sensors and processing power. In this paper we present an approach to the RUL prediction of a shot blasting machine by using recordings from inexpensive vibrational sensors. The key idea consists in (i) employing generalised Jensen-Rényi divergence (JRD) as a measure of change in the vibrational pattern and (ii) associating JRD with the abrasive wear in rotor blades. It is essential to note that these two show monotonic relationship. Hereupon, a simple hidden Markov model with stochastic inputs and JRD as output is proposed. The hidden states of the model are updated on-line bymeans of Kalman filter. Prediction of the remaining useful life is done by executing Monte Carlo simulations on the updated model and evaluation of the first passage time of the JRD. The approach is successfully validated experimentally by running the machine up to failure, hence allowing for naturally evolving wear progression and breakdown.