Artificial intelligence in electrical machine condition monitoring

Electrical machine condition monitoring plays an important role in modern industries. Instead of allowing the machines to run until failure, it is preferred to gather more information about the machine condition before the machine is shut down, so that the machine downtime can be reduced due to repair. Also, it would be very useful to track the machine condition and predict the future machine condition so that maintenance plan can be scheduled in advance. In this thesis, artificial intelligence techniques are utilized for machine condition monitoring. The thesis consists of 3 parts. In the first part, Neural Network and Support Vector Machine models are built to classify different machine conditions. In the second part, time series prediction models are built with Support Vector Regression and wavelet packet decomposition to predict the future machine vibration. Support Vector Regression is applied again in the final part of the thesis to track the machine condition and determine if the machine has thermal sensitivity issue or not. In all 3 parts, experimental results are promising and they certainly can be used in practice in order to facilitate the machine condition monitoring process.