Diagonal recurrent neural network based on-line stator winding turn fault detection for induction motors

The main limitation of feed-forward neural network based modeling methods for stator winding turn fault detection is its poor dynamical processing capability. To solve this problem, a diagonal recurrent neural network based on-line turn fault detection approach for induction motors is presented in this paper. Two diagonal recurrent neural networks are employed to detect turn fault. One is used to estimate the fault severity, the other is used to determine the exact number of fault turns. In order to make the diagonal recurrent neural network model more simple and accurate, an adaptive dynamic back propagation algorithm is proposed to determine the optimum number of the hidden layer neurons. Experiments are carried out on a special rewound laboratory induction motor, the results show that the diagonal recurrent neural network based diagnosis model determines the shorted turns exactly, and is more effective than the forward neural network based diagnosis model under the condition of detecting a slowly developing turn fault.