Detection of induction motor faults using vibration, current and acoustic data

Early diagnostics of incipient faults in induction motors is an important aspect of preventive maintenance strategies. In this thesis, frequently encountered induction motor fault types are detected and classified using stator current, vibration and acoustic data. Current, vibration and acoustic that data are acquired from the experiments realized under different loading conditions of induction motors on which different fault types created synthetically are used for feature extraction by means of different signal processing techniques including Wavelet Packet Decomposition, 2D Wavelet Transform and Local Binary Patterns. Conversion of one-dimensional data signals into two-dimensional grayscale images whose sizes are arranged due to their autocorrelation value provide the opportunity of utilization of texture based methods for feature extraction. Novel feature vectors are proposed for fault classification and their performances are tested with Neural Network and Bayesian based classifiers. Besides, a remarkable benchmark database is constructed consisting of stator current, vibration and acoustic data acquired under many operating conditions. This database is expected to be used as medium for future works of fault diagnosis related to preventive maintenance strategies of the induction motors.