A physics-guided deep learning framework for corrosion prediction using electromechanical impedance

Abstract

Corrosion monitoring plays a crucial role in ensuring the safety and reliability of steel structures. Among various methods, the electromechanical impedance (EMI) technique has been widely recognized as a promising corrosion detection approach due to its high sensitivity and nondestructive characteristics. In recent years, many scholars have attempted to combine deep learning models with EMI technology to achieve quantitative monitoring of corrosion damage. However, existing deep learning methods still face numerous challenges, such as insufficient physical interpretability, limited generalization capability, and strong dependence on large amounts of labeled data. To address these issues, this article proposes a physics-informed deep learning framework that incorporates system dynamics principles into the model training process by introducing carefully designed physical constraints into the network structure. While ensuring physical consistency, this method significantly improves prediction performance, with the coefficient of determination increasing from R² = 0.9316 in traditional deep learning models to R² = 0.9843. Furthermore, this method not only demonstrates excellent performance in prediction accuracy and robustness but also significantly enhances model interpretability, providing a practical solution for the application of steel structure corrosion monitoring in actual engineering projects.