An experimental comparison of model-based fault detection and isolation techniques in a multi-tank system

Fault detection and isolation (FDI) systems ensure processes’ safety, reliability, and secure operation. Many techniques proposed in the literature work well in simulation scenarios, but perform poorly in realistic systems with significant noise. This work presents a hybrid-based technique that uses an extended Kalman filter to smooth out measurements and parameter estimation of a nonlinear second-order model for quantifying leakage faults in an interconnected multi-tank system. The cost function for parameter estimation minimizes the squared error between predictions and measurements. The experimental results demonstrate that the proposed hybrid technique yielded a better estimate than the baseline FDI techniques. The proposed technique improved the prediction of tank dynamics and fault estimation quality by about $\mathbf{3 2 \%}$.