Tuning fuzzy observers for residual generation

In this paper, we present an application study of a fuzzy output observer to fault detection on a laboratory three tanks system. The concept used here is based on fuzzy relational model-building and model-based residual generation technique. Several parameters are involved in the model-building procedure. We focus our attention on their influence on the behaviour of the residuals. We present the results obtained by the application of the observer to the detection of sensor failures. The influence of the design parameters, such as the norm operators, the definition of linguistic variables or the choice of the identification algorithm is also discussed in this context.