Research on Few-Shot Fault Diagnosis and Feature Extraction Mechanism Based on ADPCW-ELCNN

Aiming at the challenges in few-shot fault diagnosis of rotating machinery, such as weak features being easily masked by noise, the disconnection between parameter and feature adaptation, and the difficulty in balancing generalization ability with efficiency, an Adaptive Dual-Parameter Collaborative Wavelet Convolutional Neural Network is proposed. This model focuses on dual-parameter collaborative optimization, dynamic feature adaptation, and lightweight generalization enhancement. Through innovative designs including quantitative correlation between wavelet and convolutional kernel parameters, as well as global-local feature adaptation, it effectively integrates the advantages of time-frequency analysis with lightweight requirements. Experimental validation on bearing and gearbox datasets under complex working conditions shows that the model size is only 55 KB, with an average diagnostic accuracy of 99.98%, significantly outperforming traditional models. Even under strong noise and extreme few-shot scenarios with only 20 samples, it maintains excellent performance, providing an efficient and practical new solution for intelligent fault diagnosis of rotating machinery with limited samples.