Elastic Net Representation-based Projections for Bearing Fault Classification

In this paper, a novel dimensionality reduction algorithm named elastic net representation-based projections (ENRP) has been proposed for bearing fault diagnosis, which effectively excavates useful features from highly-correlated and strongly-disturbed vibration signals. An elastic net model is integrated into the algorithm, which absorbs the merits and also lower the limitations of applying sparse representation (SR) or collaborative representation (CR) alone. Moreover, maximum margin criterion (MMC) is also incorporated to take the global structures of data into consideration. The effectiveness of the proposed method has been demonstrated with experiments on CWRU bearing data using l-Nearest-Neighbor (lNN) classifier. The ENRP method outperforms the other compared approaches with greater classification accuracy and stability, and shows its superiority especially when dealing with vibration signals interfered by strong background noise.