Visualization of electrochemical behavior in carbon steel assisted by machine learning

Identification of microstructures in steel has been extensively studied to improve the understanding of corrosion behavior. However, identification by expert eyes could be subjective, and most previous works on identification are solely based on morphological features. Furthermore, it is more difficult to identify local microstructures on a small scale-length. In this study, we developed a method for differentiating local microstructures on low carbon steel based on multiple physical properties at the nanoscale combined with machine learning techniques. Machine learning techniques were applied to the atomic force microscopy images of multiple physical properties, that is, not only of morphological features but also of the surface potential and capacitance gradient. Thereafter, we analyzed the corrosion behavior according to the concentration of the NaCl solution in which the samples were immersed, on the basis of the identified local microstructures as well as obtained physical properties. This study, which is based on these multiple physical properties, potentially provides a powerful tool for identifying and visualizing features of data. It could be further extended to electrochemical systems with more complex microstructures.