Bearing Fault Diagnosis via Improved Collaborative Representation

Rolling bearing plays an important role in industrial applications, and the analysis of vibration is widely used in the bearing fault diagnosis. This paper presents a novel model for fault classification and detection of rolling bearing based on vibration data. Unlike commonly used models based on feature extraction, this model emphasizes the ability of a given sample to be represented by all training samples. We take advantage of the observation that the vibration signals of different conditions have similar amplitude of the frequency spectrum. On that premise, an improved collaborative representation based classification (CRC) algorithm via dictionary learning (DL) is used to intelligently determine the condition of rolling bearing. The experiments demonstrate the proposed model is effective and adaptive for bearing fault diagnosis, while it has significantly small computational complexity.