Experimental Verification of Neural Network-Based Fault Types Classifier for Current Sensors in Induction Motor Drive

This paper proposes a method to detect and classifying the types of current sensor faults used in induction motor drives. The proposed fault classifier uses a multilayer perceptron (MLP) architecture with separate classifier paths for each phase of the motor. The classifier is trained using mathematical models of different types of sensor faults. The use of values of rotor flux angle from the DRFOC (direct rotor-fieldoriented control) structure allows simultaneous acquisition of one period of stator current and extraction of symptoms indicating the type of damage. The proposed classifier achieves a high classification accuracy in laboratory tests on 1.1 kW induction motor of 98.2% in phase A and 98.1% in phase B. The overall classification accuracy of 98.1% was determined by considering a wide range of motor operating points. The training process can be carried out without using measurements from the laboratory bench but including measurements from undamaged sensors in the training set can improve performance. The proposed classifier provides an effective method to detect and classifying stator current sensor faults, with potential applications in fault diagnosis and maintenance.