Prediction of quantitative in-situ local corrosion via deep learning

Abstract

A deep understanding of corrosion behavior is critical for improving steel durability and reliability. Typically, it requires significant experimental effort and extensive measurements over a long period of time. Therefore, exploring the time-dependent corrosion at an early stage to predict its progress at a later stage can effectively understand it. In this study, we quantitatively predicted later stages of local corrosion behavior using deep learning methods based on the early stage topographical information. Furthermore, we predicted and visualized the formation, growth, and accumulation of the particle-like oxides. Our proposed method can be extended to other types of corrosion-resistant electrochemical materials.