Reduced Model based Fault Detection and Isolation with Residual Threshold Determination

Model based fault diagnosis approaches are greatly beneficial for detecting and isolating faults accurately. However, when used on large scale systems, i.e. high order models, several problems can arise, especially when considering real-time applications. This paper presents a two step procedure for an accurate reduced order model based fault diagnosis. The first step consists of obtaining a reduced size model, in an optimal way, that takes the faults into consideration. The second step implements a fault detection and isolation method based on the reduced model. Reduced order observers and residuals are obtained in this step and possible thresholds are calculated. The procedure is demonstrated through an application example showing that it is able to accurately detect and isolate the faults.