Comparison of machine learning methods for fault analysis in induction motors

The use of induction motors has increased considerably in recent years. Because of this increase, failure rates of induction motors have increased. The increase in induction motor errors causes disruption of many processes, loss of work, loss of equipment and even worse, loss of life. For all these reasons, it has become very significant to detect the failure beforehand. In this thesis, data sets were created with three different induction motors under six different loads, which are representing different operating conditions. While creating this data set, induction motors, which have created eleven types of errors deliberately, were operated under different loads and noise, vibration and current data were collected. This data set consists of seven channels of audio data, six channels of vibration data and three channels of current data. Nine kinds of time-based and five kinds of frequency-based features were extracted from the created data set. The features were used in Neural Network training and testing firstly separately and then combined. As a result, motor types and fault types in motors are easily distinguished.