Rolling bearing fault diagnosis method based on variational mode decomposition and kernel-extreme learning machine optimized by subtraction-average-based optimizer

Aiming at the problems of difficultly for extracting the vibration signals’ features and low fault identification rate of aircraft engine’s faulted rolling bearings， a rolling bearing fault diagnosis method based on Subtraction-Average-Based Optimizer （SABO） to optimize Variational Mode Decomposition （VMD） and Kernel-Extreme Learning Machine （KELM） is proposed. Firstly， the minimum envelope entropy is used as the fitness function. And the VMD’s parameters are optimized by SABO to obtain the optimal parameter combinations of the decomposition modes number K and the penalty factor α. At that process， the index values of the minimum fitness’s IMF are obtained. And then， the optimal combinations of ［K，α］ and the index values corresponding to each fault are brought back to the VMD to obtain the best IMF of each sample， and calculate nine kinds of time-domain features of the best IMF. Furthermore， SABO and KELM are combined to optimize the kernel function of KELM， then KELM is reconstructed. Eventually， the features matrix is input into KELM to get the fault diagnosis results. Using the bearing data of CWRU， the data of a constructed rolling bearing simulation test and the data of a kind of dual-rotor aircraft engine’s intermediate bearing demonstrates the method’s effectiveness. The result shows： the SABO’s optimization effect based on the fitness function of minimum envelope entropy is better than the other four kinds of fitness function， and the VMD can decompose the faulted bearing vibration signals more effectively. The correct diagnosis rate of the optimized KELM for the three experiments respectively reaches：99.33%，98.67%， and 98.67%， which is a more significant improvement than the correct diagnosis rate of the unoptimized KELM. Based on experiment 3， the diagnostic accuracy of SABO-KELM was 99.2%， which was higher than that of PSO-KELM （98.3%） and GWO-KELM （97.5%）. Based on the above analysis， SABO-KELM model can be applied to simple and complex propagation paths of different bearing types， and it can be inferred that the model can be used as a rolling bearing fault diagnosis method for general mechanical systems.