Acoustic Signature Based Intelligent Health Monitoring of Air Compressors with Selected Features

Abstract

Dimensionality reduction and identification of relevant features are important for the classification accuracy. Selecting large number of features increases computational complexity whereas selection of too few features may not contain sufficient information required for the classification. This paper presents the comparative performance of different feature selection techniques namely Principal Component Analysis (PCA), Independent Component Analysis (ICA), Mutual Information (MI) methods: MIFS, mRMR, NMIFS, MIFS-U, and Bhattacharyya Distance (BD) in order to select optimal feature set for attaining better classification accuracy. With the results of comparative performance analysis one can get valuable insight about the effectiveness of different feature selection techniques, which in turn allows us to use the most suitable feature selection technique for enhanced fault diagnosis using CBM of air compressor.