Use partly hidden Markov model to evaluate a future failure

The diagnosis of failures, if done properly and enabling early degradation detection, represents a means to optimise the production unit and to reduce the costs by avoiding failures. This challenge can be addressed through hidden Markov models HMMs that can estimate the probability of a future failure based on observation system. However, sudden changes in system behaviour due to either system malfunction or one of its components will affect the operation process. Thus, previous errors have an impact on the current system state and a regular HMM does not meet this requirement unlike partly hidden Markov models PHMMs, which combines the power of conditioning the state transition probability to the previous observation. In this paper and for the first time, we propose to use PHMM as a mechanism to identify a future failure of industrial furnace. The obtained results prove that using PHMM seems to be particularly effective, efficient and outperforms the HMM.