Synchronous Machines Field Winding Turn-to-Turn Fault Severity Estimation Through Machine Learning Regression Algorithms

Interturn field windings faults are quite common in synchronous machines, particularly in turbogenerators. The synchronous machine can operate with a certain interturn fault severity level. This paper presents a new field winding interturn fault severity estimation method based on machine learning regression algorithms. The theoretical excitation current is estimated by artificial intelligent. For this purpose, it is necessary the use of numerous healthy operational data to train the algorithm. Afterward, the algorithm estimates the field current, which is compared to the real current measured. The fault severity level is calculated from this comparison. The use of machine learning implies an improvement on the sensitivity to previous method based on the estimation of the theoretical excitation current by traditional synchronous machines models as Potier or ASA. The measurement errors increase the minimum fault severity level that can be detected. The proposed algorithm has been verified with more than 1800 experimental results in a special laboratory synchronous machine, obtaining a better estimation on the fault severity level than traditional methods.