Quantitative Evaluation of Surface Defect Parameters in Eddy Current Testing Based on AUN-CNN-LSTM Algorithm

Eddy current testing (ECT) is a key nondestructive testing technology for surface defect detection in industrial production, yet liftoff interference between the probe and specimen severely degrades the accuracy of defect parameter quantitative evaluation. To solve this problem, a quantitative evaluation method for ECT surface defect parameters based on the AUN-CNN-LSTM algorithm is proposed in this study. A coaxial dual excitation-receiver (CDER) probe was designed to acquire voltage signals on a self-build ECT experimental platform, and the additive uniform noise (AUN) strategy was adopted to expand the signal dataset. Multi-dimensional features of the augmented signals were extracted and input into the CNN-LSTM hybrid model, where the CNN captures local feature patterns of ECT signals, and the LSTM learns multi-feature sequence evolution laws, thus decoupling the complex relationship between ECT signal features and defect parameters and suppressing liftoff interference. Experimental results demonstrate that the defect radius evaluation error is less than 0.1 mm and the depth evaluation error is less than 0.25 mm, and it maintains excellent anti-interference performance even under unknown liftoff conditions. This method provides a reliable technical solution for accurate ECT surface defect parameter measurement and has important practical value for industrial high-precision defect detection.