A Multilayer Perceptron-Based Approach for Stator Fault Detection in Permanent Magnet Wind Generators

The permanent magnet synchronous-based generator (PMSG) is a common type of wind generator that recurrently interrupts its operation due to stator faults occurrences. These faults are difficult to detect and can softly lead to machine damages since they occur between turns (turn-to-turn fault), or between turn and machine housing (turn-to-ground fault). Thus, these generators must be constantly monitored so that these faults can be detected in their initial stage. In fact, early detection reduces the cost of maintenance, while also decreases the inactivity time of the turbines. This work proposes a strategy to detect stator faults at the first moments by means of a classifier module that analyzes the stator electrical current patterns. This classifier is based on a multilayer perceptron (MLP) neural network, which was trained using a dataset of instances created by means of a mathematical model of the PMSG. The results show the MLP classifier is able to identify the problem with more than 94 % of accuracy, considering both turn-to-turn and turn-to-ground faults. Moreover, the detection was achieved in a very initial stage (1 % of faulty turns), contributing to the low-cost maintenance and correct operation of the turbines.