Fault detection and isolation based on hybrid modelling in an AC motor

Abstract

A hybrid neural network-first principles modelling scheme is used in this paper, to model an induction motor and to develop a fault detection and isolation (FDI) scheme. The hybrid model combines a partial first principles model, which incorporates the available prior knowledge about the process being modelled, with a neural network which serves as an estimator of unmeasured and unknown process parameters that are difficult to model from first principles. A fault detection and isolation scheme has been defined based on this hybrid model. This suitable model enables system faults to be simulated and the change in corresponding parameters to be predicted without physical experimentation. The detection scheme is based on the calculus of the residues as the difference between the real system and the hybrid model. The isolation scheme is based on neural networks. A three-phase induction motor was simulated under normal operation conditions using the hybrid methodology. Faults in some internal parameters and voltage imbalance between phases supply have been simulated and detected with the FDI scheme, with quite good results.