Research on Eddy Current Testing of Conveyor Belt Joint Twitch Based on MixUp-CS-CNN

Detecting twitch in conveyor belt steel wire rope joints using eddy current testing faces challenges such as insufficient training signals, low classification accuracy, and slow recognition. To address these issues, this paper designs a differential probe and proposes a MixUp-CS-CNN method. The finite element simulation model is validated by comparing simulated and experimental signals under different twitch distances. The MixUp method augments simulation samples, constructing a dataset of 600 groups covering six twitch distances. Fault features are extracted from time, frequency, and time-frequency domains, and the Chi-Squared test reduces dimensionality to obtain an optimal feature subset. This subset is then input into a CNN classifier. Results show a diagnostic accuracy of 98% for joint twitch. The proposed method overcomes the problems of insufficient sample data and poor generalization, while balancing detection accuracy, efficiency, and on-site adaptability through the optimized differential probe and feature reduction.