Research on the Evaluation of Metal Surface Defect Using CSE‐LSTMNet and Eddy Current Testing

The performance of metal workpieces is crucial in industry, and the evaluation of their surface defects is a key factor in ensuring the safety of industrial production. Eddy Current Testing (ECT) offers a non‐contact, non‐invasive method for detecting metal surface defects efficiently and accurately. However, the evaluation of metal surface defects often heavily relies on the experience of operators, which is not only inefficient but also susceptible to human factors. To effectively address the above issues, this paper proposes an evaluation method using Deep Learning (DL) technology and designs a CSE‐LSTMNet model, which introduces the dilated convolutional layer on top of the traditional CNN‐LSTM model, enabling faster training speeds and higher accuracy. Meanwhile, a Squeeze and Excitation Networks (SENet) module is added after each residual block to enable the model to automatically acquire the significance of each feature channel, thus enhancing useful channel features while suppressing unimportant ones. The CSE‐LSTMNet model has been tested on the publicly available MDDEECT dataset. The simulation results show that the proposed CSE‐LSTMNet model exhibits excellent performance in the metal surface defect classification task. Compared with the traditional CNN‐LSTM benchmark model, the model achieves significant performance improvement, with the classification accuracy increasing from 86.67% to 92.78% and the F1 score increasing from 86.69% to 92.79%. This indicates that the CSE‐LSTMNet model has higher performance and significant advantages in metal defect evaluation.