Bearing Fault Detection Based on Multiresolution Permutation Entropy

Fault detection can provide constructive guidance for the predictive maintenance of rolling element bearings. For detecting different faults, permutation entropy (PE) has served as an effective tool due to its ability to measure the dynamic complexity of fault signals. Unfortunately, the existing PE-based methods only consider the one-fold oscillatory pattern, which limits the feature extraction ability. To address the above issue, this paper proposed multiresolution permutation entropy (MRPE) to match different oscillatory patterns by multiple wavelets. Based on MRPE, a fault detection framework has been developed to identify the different faults, and the Laplacian score (LS) is adopted to further select the most sensitive features in MRPE. The simulation results show that MRPE can characterize more comprehensive fault information hidden over the entire frequency band than existing PE-based methods. The experimental results indicate that the proposed fault diagnosis framework surpasses available PE-based methods in detecting different mechanical faults of bearings.