Development of a Fault Prognostics Model for a Non-linear Control System using the Extended Kalman Filter

Model-based approaches have been applied widely to linear systems for fault detection and
diagnosis. Unfortunately, the linearisation is not always an effective way of representing the original system. The inaccuracies are even higher when the system is contaminated by various noises such as operating uncertainties and experimental noise. An Extended Kalman Filter (EKF) has the potential to overcome these problems.
Developing the EKF-based approach for fault prognostics is difficult both in implementation and in understanding. In order to understand the characteristics of the EKF and its performance in fault detection, a numerical simulation is developed in the Simulink environment. Such software is a platform for applying the EKF-based approach on real non-linear control systems, which can help to determine its feasibility before applying it to real systems. For using the EKF-based approach, a prior knowledge of the system’s model is required to compute both the prediction of the state
estimate and its Jacobian matrixes. The non-linear control system modelled for this paper is a classical fluid level control system. Both healthy and commonly occurring fault conditions are implemented into the model by inducing faults in the simulated fluid level control system.