ModelAdaptive Bearing Fault Diagnosis in Small Sample Situations

Aiming at the challenges of high fault feature complexity and sample scarcity in real industrial scenarios, this study proposes a small-sample fault diagnosis framework based on feature sensitivity optimization, the convolutional meta-learning network (GCMAML), which deeply integrates the complex feature parsing and meta-learning mechanisms. First, the feature sensitivity enhancement module adopts Gramian Angular Difference Field (Gramian Angular Difference Field, GADF) to perform differential geometric transformations on vibration signals to construct tensor features with fault parameter sensitivity, and then designs the convolutional-meta-learning synergistic mechanism to realize cross-task knowledge migration through the gradient iteration strategy. Extreme small-sample scenarios are constructed based on the public bearing dataset to verify the effectiveness of the proposed method. The experimental results show that the proposed method has excellent diagnostic performance and generalization ability.