Abstract:  Aiming at the problem that early weak faults of rolling bearings were difficult to detect and the fault diagnosis rate was not high, a rolling bearings fault diagnosis method was proposed based on parameter optimization of variational mode decomposition and improved deep belief network. Firstly,in order to eliminate the influence of artificial selection of VMD parameters,the whale optimization algorithm was used to search for the best combination of the numbers of modal decomposition and the penalty factor of VMD algorithm. Then,the original rolling bearing vibration signals were decomposed by parameter optimized variational mode decomposition and a high-dimensional data set after decomposition was composed of the frequency spectrum of intrinsic mode functions. Finally, the highdimensional data set was entered directly deep belief network optimized by sparrow search algorithm for pattern recognition. The results indicate that for rolling bearing faults, the VMD algorithm fault recognition rate of the same pattern recognition method is 97.4% higher than that of EMD algorithm 96.5%. And under the same signal processing method, the fault diagnosis rate of DBN network is 98.7%, which is higher than that of SVM algorithm of 97.4%. The WOA-VMD-SSA-DBN algorithm fault diagnosis rate reaches 100%, and the fault diagnosis effect is further optimized.

Key words:  rolling bearing; fault diagnosis; variational mode decomposition(VMD); whale optimization algorithm(WOA); deep belief network(DBN); sparrow search algorithm

SHENG Xiao-wei, YU Lin-xin, BI Peng-fei, et al. Rolling bearing fault diagnosis based on parameter optimization VMD and improved DBN［J］.Journal of Mechanical & Electrical Engineering, 2021,38(9):1107-1116.