Detecting faults in nonlinear dynamic systems using static neuro-fuzzy models

A fault detection method which uses a static neuro-fuzzy model to describe the correct operation of the plant is presented. It is shown that a static neuro-fuzzy model can be used to detect faults in nonlinear dynamic systems. Abnormal operation of the plant is detected when the residuals exceed a variable threshold value which accounts for modelling errors. The fault detection scheme is used to detect two types of faults in the cooling coil subsystem an air-conditioning plant. Results are presented which demonstrate that the scheme can detect abnormal operation, without generating false alarms, even if the fault-free and faulty behaviour are similar at some operating points