Corrosion Prediction of Weathered Galvanised Structures Using Machine Learning Techniques.

Area of Science:

• Materials Science
• Environmental Science
• Engineering

Background:

• Galvanized steel atmospheric corrosion is a global issue impacting structures and equipment.
• International Organization of Standardization (ISO) standards necessitate pollutant deposition data for corrosion prediction.
• Existing methods for corrosion loss prediction often lack accuracy or rely on inaccessible parameters.

Purpose of the Study:

• To develop predictive models for estimating galvanized steel corrosion loss.
• To utilize easily accessible global parameters for corrosion prediction.
• To improve the accuracy and reliability of corrosion loss estimations for various atmospheric conditions.

Main Methods:

• Utilized experimental data from internationally validated studies for data mining.
• Employed Self-Organising Maps (SOMs) with supervised and unsupervised layers for prediction.
• Developed a formula optimized with Newton's method for long-term corrosion extrapolation.

Main Results:

• Successfully predicted first-year corrosion loss, corrosivity categories, and uncertainty ranges.
• Achieved high prediction performance validated by Euclidean distance comparisons.
• Demonstrated an average saving of up to 16% in coating applications through accurate predictions.

Conclusions:

• The developed models offer accurate predictions of material loss for galvanized steel structures.
• These models reduce structural over-dimensioning and enhance efficiency and sustainability.
• Implementation leads to significant cost reductions in infrastructure maintenance and protection.