Bearing fault diagnosis and prognosis using data fusion based feature extraction and feature selection

The extraction of significant features is essential for efficient fault diagnosis and prognosis of rolling element bearing. Data fusion is the predominant technology for extracting significant features by fusing several original features. In this paper, seventy-two original features are extracted from bearing vibration data using various signal processing techniques. The relevant features subset is selected from the extracted features using the Random Forest method. The selected features are fused by fourteen dimensionality reduction techniques to extract 2D fault features and health indicators, and a comparison is made between the techniques to identify the most efficient technique. The Bhattacharyya distance and Support vector machine are used to verify fault diagnosis accuracy. A new index is computed for selecting the suitable prognosis health indicator, and the Long short-term memory technique is used to predict the remaining useful life of bearing. Two real-world bearing datasets are utilized to validate the proposed methodology.