Multi-fault identification of rolling bearings using VMD and optimized MCKD

Abstract Aiming at the challenge of identifying multiple faults in rolling bearings caused by the interrelation and cross-interference of collected vibration signals coupled with environmental noise interference-which leads to overlapping multi-fault characteristic information and difficulty in fault type determination. This paper proposes a multi-fault identification method based on the combination of Variational Mode Decomposition (VMD) and parameter-optimized Maximum Correlated Kurtosis Deconvolution (MCKD). Firstly, the VMD method is employed for signal processing, where the kurtosis criterion is applied to select optimal Intrinsic Mode Functions (IMF) components for vibration signal reconstruction. Subsequently, the fitness function is constructed to determine the optimal parameter combination for MCKD, enabling separation of reconstructed signals to extract individual fault signals. Characteristic parameters are then extracted to form feature vector sets. Then, the Support Vector Machines (SVM) method is utilized for fault feature recognition and classification. Finally, the proposed method is validated through designed multi-fault injection experiments on car pully bearing. Experimental results demonstrate that the proposed method effectively separates multiple fault signals under environmental noise and achieves accurate identification of specific faults in car pully bearing, with a fault recognition rate reaching 97.8%. This research provides a valuable reference for multi-fault diagnosis in bearing systems, offering technical insights for condition monitoring and predictive maintenance in rotating machinery.