Defect depth classification for deep surface-opening defects in stainless steel plates based on eddy current testing

Abstract

As a non-destructive testing technique, eddy current testing (ECT) is widely applied in the industrial field, especially suitable for defect detection of non-magnetic metal components such as stainless steel. However, traditional ECT technology still has obvious limitations. This paper integrates deep learning with ECT technology, focusing on the depth classification task of defects in stainless steel plates. Taking ResNet as the backbone network, a multi-scale convolution module and channel attention mechanisms are embedded to construct an intelligent detection model adapted to 1D ECT signals. In the research, a standardization method is adopted to preprocess ECT signals, retaining key features related to defect depth. Combined with the cosine annealing learning rate scheduling strategy with warm restarts, the convergence performance of the model is guaranteed, and the effectiveness of the model is verified through ablation experiments. Experimental results on the dataset demonstrate that the proposed model achieves an average accuracy of 92.47% ± 0.65% over five independent runs, and an ensemble accuracy of 94.2%, significantly outperforming baseline methods and confirming its robustness and effectiveness for the three-class classification of deep surface-opening defects.