Rotor faults diagnosis in synchronous generators using feature selection and nearest neighbors rule

Feature selection and nearest neighbors rule technique is used to diagnose large-generator rotor faults. Thus, a specific experimental setup has been designed to perform the methodology of rotor faults detection. This experimental setup is a small scale prototype of a nuclear plant turbo-generator, which is actually DC excited synchronous machine. In this generator, electrical (turn-to-turn failures) and mechanical faults (eccentricities) can be carried out. Sixteen functional states have been performed for five operating points. Stator current, voltage and flux density in the air-gap have been recorded. A list of features is then extracted from these records. To reduce their number, the SBS algorithm is used and the classification is performed by using the k-NN rule. As a result, the classification accuracy is 77.4% and the rotor faults accuracy reaches 85.1% which prove the efficiency of the method.