Detection of Bias, Drift, Freeze and Abrupt Sensor Failure using Intelligent Dedicated Observer Based Fault Detection and Isolation for Three Interacting Tank Process

This paper presents a design of MANFIS (Multiple Adaptive Neuro Fuzzy Inference System) based sensor Fault Detection and Isolation (FDI) scheme for a three interacting tank system. Three pairs of dedicated observers are designed to estimate the three states of the system. The observers designed are fuzzy systems whose optimal membership functions and rule base are determined by neural networks. The difference between the estimated and measured value is called as residuals. Decision functions are determined from the residuals. These functions are compared to a threshold value, when the value of these functions exceed a particular threshold, the presence of fault is indicated. The FDI designed is implemented to detect sensor bias, abrupt sensor failure, sensor drift and sensor freeze types of sensor faults.