Corrosion Prediction of Weathered Galvanised Structures Using Machine Learning Techniques

Abstract

Galvanised steel atmospheric corrosion is a complex multifactorial phenomenon that globally affects many structures, equipment, and sectors. Moreover, the International Organization of Standardization (ISO) standards require specific pollutant depositions values for any atmosphere classification or corrosion loss prediction result. The aim of this research is to develop predictive models to estimate corrosion loss based on easily worldwide available parameters. Experimental data from internationally validated studies were used for the data mining process, basing their characterisation on seven globally accessible qualitative and quantitative variables. Self-Organising Maps including both supervised and unsupervised layers were used to predict first-year corrosion loss, its corrosivity categories, and an uncertainty range. Additionally, a formula optimised with Newton’s method has been proposed for extrapolating these results to long-term results. The predictions obtained were compared with real values using Euclidean distances to know its similarity degree, offering high prediction performance. Specifically, evaluation results showed an average saving of up to 16% in coatings using these predictions. Therefore, using the proposed models reduces the uncertainty of the final structures state by predicting their material loss, avoiding initial over-dimensioning of structures, and meeting the principles of efficiency and sustainability, thus reducing costs.