Key points are not available for this paper at this time.
Recent progress in data driven approaches is reshaping corrosion electrochemistry by enabling quantitative interpretation of complex electrochemical responses that involve interactions among environmental variables, alloy composition and surface chemistry. This review examines research published after 2010 that applies machine learning methods such as artificial neural networks, support vector regression and ensemble learning to predict corrosion rates from multivariate electrochemical and environmental datasets. These studies demonstrate that data driven models can reproduce current density evolution, mass loss behaviour and other electrochemical indicators with improved accuracy compared with traditional empirical relations while also revealing mechanistic patterns such as the contribution of alloying elements or the influence of moisture controlled wet and dry cycles. Parallel developments in corrosion type identification have used supervised learning and unsupervised learning to analyse electrochemical noise data, in situ images and microscopy based measurements, allowing automated recognition of uniform attack, pitting initiation and deposit controlled localized degradation. A central point of debate is whether models trained on limited electrochemical conditions can generalize to new regimes. This issue is addressed through hybrid approaches that incorporate physics informed parameters, electrochemical rate expressions or structural constraints that enhance interpretability and improve extrapolation beyond the original training domain. By integrating evidence from more than one hundred peer reviewed studies, this review clarifies the demonstrated capabilities of machine learning in electrochemical corrosion research and highlights the methodological and data related challenges that must be resolved for widespread adoption in scientific and industrial practice.
Guo et al. (Wed,) studied this question.