Automatic determination of stator short circuit fault and fault severity of permanent magnet synchronous motor

Stator faults are one of the types of faults that are common in electric machines. An electric machine with a short circuit fault in the stator can not provide the desired performance and efficiency. The short circuit between two windings in the stator of an electric machine results in a high temperature rise between those windings and in a short time it can cause a short circuit in the windings around. Thus, the short-circuit phenomenon that begins at small ratios can spread rapidly and affect the operation of that machine. The stator short-circuit detection at early-stage is critical and important because it may cause unexpected results, especially in permanent magnet synchronous motors (PMSM), commonly used in applications where precise speed and position control is required. In this thesis, two methods are proposed which enable the fault detection and fault severity estimation of a PMSM while the short-circuit fault between stator windings is still in its initial stage. The first method uses one stator phase current only to detect the fault and the severity of the fault from the amplitudes of the fundamental and 3rd harmonic components obtained from the current spectrum at a wide load and speed range. The other method uses the positive and negative stator short-circuit fault related harmonics in the current and the voltage space vectors, which are identified by the studies made in the thesis, to detect short circuit fault and fault severity at wide range of speed and load conditions. The space vectors are obtained from the stator 3-phase current and voltage signals of the PMSM by the Park transformation. In both of the methods, artificial neural networks were used to automatically detect the stator short-circuit fault and determine fault severity at high accuracy rate. For the detection of fault and determination of fault severity, PMSM's mathematical equations are used to simulate the faulty motor models created in Matlab / Simulink environment and the fault signatures used in the two methods are tested. Similar results were obtained using an experimental test bench. In this thesis, for the detection of fault and estimation of fault severity, patterns were formed by fault signatures obtained from experimental test bench and classification of fault severity was provided by machine learning algorithms. Various machine learning algorithms are trained and the results are compared with each other using fault signatures obtained from motors with different fault rates at a wide speed and load range conditions. Considering accuracy and success of the algorithms, the suitability of the algorithms for determination of stator short circuit fault were identified.