Zastosowanie sieci neuronowych w detekcji uszkodzeń silnika synchronicznego ze zwartym uzwojeniem wirnika

This paper describes an automatic recognition of synchronous motor faults based on spectral analysis of stator and field currents with the aid of neural networks. As the diagnostic object a specially designed and constructed 2-pair pole synchronous machine with two parallel branches containing two groups of coils on each stator phase was used. The ends of every group are accessible allowing different connections to be made thus simulation of various fault conditions to be done. For the purpose of the paper a nineteen different fault conditions were investigated, including alone breaks of one group of coils for each phase, simultaneous breaks of one group of coils for each pair of phases and for all three phases, alone shorts of one group of coils for each phase and combined short and break faults. The spectral analysis of the phase and field currents revealed the frequencies which are representative for a particular fault. The amplitudes relating to these frequencies were then used in automatic process of motor state recognition. The recognition was made by a two different neural networks - one for analyzing the phase currents and the other one for analyzing the field current.