Feature engineering in fault diagnosis of induction motor

The use of data driven intelligent system is gaining importance in the area of condition monitoring of electrical equipment. However, irrelevant and redundant input features make the system bulky, computation intensive and provides poor classification accuracy. Data mining and feature selection techniques play an important role to reduce these problems. Not only the feature selection techniques but also the clustering quality of the selected features actually guides the system engineer to pick up the best features for developing an intelligent system for real time applications. This paper have proposed an effort to investigate the “goodness” of the selected features yields by the various feature selection techniques in the area of fault diagnosis of induction motor. Both vibration and stator current based approach have been considered. Among several validity indices DB index has been used to measure the compactness and separated features of the dataset in the selected feature space. The total feature engineering guideline in the area of fault diagnosis of induction motor with some experimental verification has been demonstrated here.