Simulation data-driven bearing fault diagnosis method under small sample conditions
WU Yanan
WU Jinhui
CHEN Yongchang
YANG Yu
TAO Yourui
Abstract：[Objective]Deep learning-based fault diagnosis methodologies typically require substantial volumes of high-quality training data.In response to the prevalent challenge of acquiring authentic bearing fault data,which significantly impedes diagnostic accuracy,a novel simulation data-driven bearing fault diagnosis framework was specifically proposed for small-sample scenarios.[Methods]The simulation data of bearings in different states was obtained by constructing a dynamic model,the simulation data was enhanced based on the conditional channel Wasserstein generative adversarial network with gradient penalty(CCWGAN-GP)model that introduced the channel attention mechanism,and synthetic data that was highly similar to the simulation data was generated to form the virtual data.Finally,the parameters learnt from the virtual data were migrated to the test data by using a migration learning model that was based on the loss of the validation set to make fine adjustments to achieve effective diagnosis of bearing faults under the conditions of small sample test data.[Results]By using Case Western Reserve University bearing data for tests,the effectiveness of the proposed method is verified.It is shown that the proposed method can achieve excellent fault diagnosis performance under small sample conditions.
Keywords：BearingDynamic modelingFault diagnosisSmall sampleData augmentationMigration learning
Publication Date：2026-07-15
Online Publishing Date：2026-09-12(First online date of this platform, not the publication date of the document)
Pages：12( 142-153 )