Abstract
This paper reviews recent approaches to modelling and predicting the corrosion rate of metallic materials using machine learning methods. Corrosion remains a pressing issue for pipeline systems, industrial water circuits, structures exposed to atmospheric and marine environments, energy infrastructure, and protective coatings. Its development is associated not only with reduced material reliability and shorter service life of structures, but also with the risks of accidents, leaks, and adverse environmental impacts. The prediction of corrosion processes depends on the quality, completeness, and physicochemical relevance of the input data. In recent studies, such data are obtained through sensor-based monitoring, electrochemical measurements, atmospheric exposure tests, analysis of industrial water systems, and assessment of protective coating degradation. Model input parameters include temperature, relative humidity, pH, chloride concentration, electrical conductivity, corrosion current, exposure time, material composition, and the characteristics of the surface or protective layer. The paper summarises the applicability of regression and ensemble algorithms, neural network-based, hybrid, two-stage, and event-based approaches to predicting the corrosion behaviour of materials. Particular attention is given to models that account for the temporal nature of the corrosion process, the structure of the input data, and intermediate physicochemical indicators of degradation. It is shown that such approaches make it possible to move from assessing the current condition of a material to predictive monitoring of corrosion risks in engineered systems.