Bayes error based feature selection: An electric motors fault detection case study

In the modern industrial sector there is a growing interest on electric motors safety, reliability and maintainability. In this context health monitoring and fault detection are crucial tasks to be performed. In this paper we introduce a univariate filter method based on Bayes error for feature selection in a fault detection scenario. The feature selection algorithm firstly estimates the probability density function of the data. At a second stage we compute the PDFs intersection area which is related to the Bayes error. Finally we choose the features with the minimum Bayes error. In order to properly test the proposed algorithm, a starter motor assembly line has been considered as a case study. Features extraction is performed on statistical time and frequency domain analysis while the quadratic classifier is used for the final fault detection. Performances of the proposed approach have been compared with those of Relieff and SFS algorithms on a 649 motors data set. Results show that our method outperforms the other two in terms of Area Under Curve - Receiver Operating Characteristic (AUC-ROC).