**1. Introduction**

Multiple metallic structures and equipment operate in outdoor conditions [1]. In such cases, one of the main problems related to their stability and durability is corrosion [2,3]. World Corrosion Organization (WCO) estimates the world direct cost of corrosion to be between 1.3 and 1.4 trillion EUR, 3.1% to 3.5% of a nation's GDP annually [4].

Corrosion is a very complex phenomenon based on the degradation of a material or its properties due to its reaction with the environment [5]. Multiple factors [6], particles [7], and variables [8,9] are involved. The character of the attack and the corrosion rate are consequences of the system formed by metallic materials, atmospheric environment, technical parameters, and operating conditions [10]. Corrective factors are introduced in the design phases to guarantee the structure's integrity during its useful life [11]. However, the difficulty of quantifying the material loss causes unnecessary over-dimensioning, leading to superfluous costs and resources consumption [12]. Proper management of this complex multifactorial phenomenon is key to sustainable development [13].

To ensure the integrity of the outer layer, structures are designed with physical protection. Historically, metallic zinc has provided excellent corrosion protection of steel structures [14]. Unfortunately, corrosion damage also occurs in such systems [15]. Since corrosion leads to a mass loss, an excess thickness is often considered to ensure service life. This not only increases manufacturing cost but also does not satisfy the principles of sustainable engineering efficiency [16]. Therefore, lacking an automated monitoring system or predictive model, routine thickness monitoring would be required [17]. These

**Citation:** Terrados-Cristos, M.; Ortega-Fernández, F.; Alonso-Iglesias, G.; Díaz-Piloneta, M.; Fernández-Iglesias, A. Corrosion Prediction of Weathered Galvanised Structures Using Machine Learning Techniques. *Materials* **2021**, *14*, 3906. https:// doi.org/10.3390/ma14143906

Academic Editor: Marián Palcut

Received: 10 June 2021 Accepted: 12 July 2021 Published: 13 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

phenomena have drawn increasing attention in recent decades due to the resulting catastrophic accidents [18] and the growing demand for sustainable designs [19]. For an optimal selection of materials, atmospheric aggressiveness must be considered. Depending on this, coating needs can be set.

The current regulation regarding galvanised metallic structures (ISO 9223:2012 [20]) groups the corrosivity level of an atmosphere into six categories. After studying the effect of corrosion on standard samples during 1 year of weathering exposure, the level of corrosion rates achieved can be established by measuring weight losses for different materials. This material's loss due to corrosion is commonly used as an initial measure for determining coating requirements. However, material loss margins are allowed within these categories, and coating thickness designs based on them are not fixed. These margins imply variability in the amount of material that can be translated into increased costs.

According to [20], two methods are proposed to classify the corrosivity of atmospheric environments, depending on the availability of experimental data. When experimental data are available, dose–response functions can be used. However, when no experimental data are available, corrosivity category estimation using the informative procedure is recommended, and as stated in the norm, it is based on the comparison of local environmental conditions with the description of typical atmospheric environments, which may cause misinterpretations [21]. Finding the optimum point between efficiency and competitive price, while remaining within limits, is therefore challenging given the lack of characterisation of the specific construction site.

The objective of this work is to develop machine learning models that, by analysing real cases, predict corrosion mass loss of zinc coatings over time. The aim is to characterise an environment without requiring long testing periods and sampling and generalising it to any location worldwide, with the data available from international studies. This considerably increases the existing knowledge about coated steel structure corrosion and extends it to the full diversity of atmospheres, thereby reducing the uncertainty of its final state.

This paper starts with a state-of-the-art analysis. Then, it explains the creation of the database through the characterisation of each sample. Next, the applied methodology is explained, and modelling and evaluation techniques are defined. Finally, results are discussed, and the conclusions obtained in this research are proposed.
