Expectation maximization approach to data-based fault diagnostics

The data-based fault detection and isolation (DBFDI) process becomes more potentially challenging if the faulty component of the system causes partial loss of data. In this paper, we present an iterative approach to DBFDI that is capable of recovering the model and detecting the fault pertaining to that particular cause of the model loss. The developed method is an expectation–maximization (EM) based on forward–backward Kalman filtering. We test the method on a rotational drive-based electro-hydraulic system using various fault scenarios. It is established that the developed method retrieves the critical information about presence or absence of a fault from partial data-model with minimum time-delay and provides accurate unfolding-in-time of the finer details of the fault, thereby completing the picture of fault detection and estimation of the system under test. This in turn is completed by the fault diagnostic model for fault isolation. The obtained experimental results indicate that the developed method is capable to correctly identify various faults, and then estimating the lost information.