Brushless Synchronous Machine Field Winding Interturn Fault Severity Estimation Through Deep Neural Networks

Interturn faults (ITF) at the main machine field winding in brushless synchronous machines (BSM) are not easy to detect as the scarcity of measurements in the rotor. This paper presents a new field winding ITF severity estimation technique based on deep learning algorithms. The theoretical exciter field current is calculated through an artificial neural network, from the machine output values, and then compared to the measured exciter field current. The estimation and the comparison are carried out without need of other than the machine electrical outputs and the exciter field current measurements. To verify the proposed technique, numerous healthy and faulty condition tests were performed on a special laboratory testbench, with more than 7 million measurements in healthy and more than 1 million in different faulty conditions. So as firstly to train the neural network with healthy operation datasets, and secondly to test its capability to estimate the fault severity level. The use of deep neural networks has proven to enhance the accuracy of the exciter field current estimation with respect to previous theoretical model-based methods, enabling to increase the sensitivity and the reliability of the fault detection and fault severity calculation.