On a Prediction Method for Remaining Useful Life of Rolling Bearings via VMD-Based Dispersion Entropy and GAN
Cancan Yi, Shuhang Li, Tao Huang, Han Xiao, Yefeng Jiang
Abstract
Data-driven approaches to predicting the remaining useful life (RUL) of rolling bearings (RBs) depends on the construction of health indicators and prediction models. Aimed at the problems of fuzzy deterioration points of RB and low accuracy of prediction models, this article has proposed a data-driven method for predicting the RUL of RB. Specifically speaking, the dispersion entropy (DE) is employed to quantify the dynamic changes of intrinsic mode functions (IMFs) after variational mode decomposition (VMD). Then, the first principal component of the DE of IMFs is extracted as the new health indicator named V-DEF, which can describe the degradation state of RB. After that, the starting point of degradation state is determined according to V-DEF. Finally, the GRU-CNN-WGAN-GP (GC-WGAN-GP) model with deep adversarial learning features is used to predict the RUL of RB. The feature matrix used for model training and validation consists of traditional health indicators and V-DEF. To verify the effectiveness of V-DEF and GC-WGAN-GP, this article has used the bearing datasets of IMS and XJTU-SY for validation. Experimental results show that V-DEF has the capability to identify the starting points of degradation state earlier and the GC-WGAN-GP model can accurately predict the RUL of RB.