The use of artificial intelligence combiners for modeling steel pitting risk and corrosion rate

Highlights

• Artificial intelligence is used in modeling the pitting risk and corrosion rate of steel.

• This study examines diverse novel prediction methods.

• The hybrid metaheuristic regression model has superior prediction accuracy.

• A scientific methodology for predicting pitting risk/corrosion rate in steel rebar.

Abstract

Corrosion is a common deterioration that reduces the service life of concrete structures and steels. Particularly, corrosion behavior is a highly nonlinear problem influenced by complex characteristics. This study used advanced artificial intelligence (AI) techniques to predict pitting corrosion risk of steel reinforced concrete and marine corrosion rate of carbon steel. The AI-based models used for prediction included single and ensemble models constructed from four well-known machine learners including artificial neural networks (ANNs), support vector regression/machines (SVR/SVMs), classification and regression tree (CART), and linear regression (LR). Notably, a hybrid metaheuristic regression model was implemented by integrating a smart nature-inspired metaheuristic optimization algorithm (i.e., smart firefly algorithm) with a least squares SVR. Prediction accuracy was evaluated using two real-world datasets. According to the comparison results, the hybrid metaheuristic regression model was better than the single and ensemble models in predicting the pitting corrosion risk (mean absolute percentage error=5.6%) and the marine corrosion rate (mean absolute percentage error = 1.26%). The hybrid metaheuristic regression model is a promising and practical methodology for real-time tracking of corrosion in steel rebar. Civil engineers can use the hybrid model to schedule maintenance process that leads to risk reduction of structure failure and maintenance cost.

Introduction

Civil infrastructures are often subject to aggressive conditions and consequently serious deterioration may occur. In recent years, the maintenance of concrete structures has become important with respect to the sustainability of infrastructure. Corrosion of reinforcing steel-bar (rebar) is a typical deterioration process that reduces the service lives of reinforced concrete structures and steel structures. To evaluate the performance of concrete members and structures, the corrosion of steel rebar must be predicted as early as possible.

For the industrial sectors in the USA, the repairing cost of corrosion damage is equal annually to 3.1% of the gross national product, among which the utilities, transportation and infrastructures contribute the most (Shaw and Kelly, 2006). Approximately 70% of corrosion occurs in a localized area (Nimmo and Hinds, 2003) and localized corrosion is more dangerous than uniform corrosion because it is more difficult to detect. The most common localized corrosion type is pitting corrosion, which may decrease steel strength as a result of a loss in rebar thickness and initiate early fatigue crack (Nimmo and Hinds, 2003). Worsening fatigue cracks can eventually cause a catastrophic failure of a structure. Risks are endemic in practically every aspect of our lives, and they always cause a great deal of potential damage (Wu and Birge, 2016). Thus, pitting corrosion must be identified before it reaches a dangerous level.

The rapid growth in the number of offshore structures urgently demand a model for predicting corrosion rate in order to reduce potential structural failures (Caines et al., 2013). However, corrosion is a highly nonlinear problem influenced by complex characteristics and models for predicting the corrosion rate of steel currently lack a theoretical basis. Researchers have yet to reach a consensus on the best model to predict corrosion rate or pitting risk due to the lack of good understanding of factors that affect the corrosion process. Many of the better known factors that affect corrosion of steel structure, like the electrochemical measurement of pitting corrosion, are not included in existing models. Therefore, an accurate model for predicting corrosion based on electrochemical data is needed.

Machine learning (ML) and artificial intelligence (AI)-based approaches have attracted a great deal of scientific attention (Chou and Ngo, 2016a, Wu et al., 2015) and have successfully used in civil engineering (Chou et al., 2015, Chou et al., 2016, Goel and Pal, 2009, Wen et al., 2009) such as modeling of pier scour (Pal et al., 2011). For instance, Wen et al. (2009) applied a single support vector regression (SVR) model for predicting the corrosion rate of 3C steel in five different seawater environments (Wen et al., 2009). Single artificial neural networks (ANNs) has been applied to predict pitting corrosion in steel reinforced concrete (Shi et al., 2011, Wen et al., 2009). To the best knowledge of the authors, only single AI models are proposed in literature to modeling the corrosion related issues.

Although the single AI-based models have proven moderately effective for solving prediction problems, one of the critical problems is how to select an appropriate model and fine-tune the model parameters, which plays an important role in good generalization performance and prediction accuracy for future use (Chou and Ngo, 2016b, Jiménez-Come et al., 2013, Yang et al., 2011). To overcome the drawback and enhance the accuracy, ensemble and metaheuristic approaches appear to be promising solutions for the above situations. Hybrid approaches that combine multiple AI models and optimization algorithms have been proposed to enhance the prediction accuracy of single AI models (Kazem et al., 2013).

This study investigated the applicability of four single AI models, four meta ensembles (i.e., voting, bagging, stacking, and tiering), and a hybrid metaheuristic regression in solving the non-linear problem of estimating pitting risk or corrosion rate of steel rebar. The single AI models are ANNs, support vector machines and regression (SVMs/SVR), classification and regression tree (CART), and linear regression (LR). These single AI models are the most commonly used techniques in related works (Chou and Pham, 2013). Moreover, four ensemble AI models are combined from the above single AI models. The ensemble AI models consists of voting, bagging, stacking, and tiering methods. The hybrid metaheuristic regression model integrates the nature-inspired metaheuristic optimization (i.e., smart firefly algorithm (SFA)) and least squares support vector regression (LSSVR).

The first originality of this study was a comprehensive investigation of AI models for modeling pitting corrosion risk and marine corrosion rate. The combination of those AI models has not been proposed yet in modeling steel corrosion. Secondly, the hybrid metaheuristic regression was presented to predict steel corrosion, in which the SFA was used to automatically optimize the hyperparameters of the LSSVR for improving prediction accuracy. Lastly, the findings of this study provide civil engineers with a promising and practical methodology for tracking of steel corrosion. Two real-world datasets were used to evaluate the applicability of various AI models. The performance of the investigated models is compared in terms of mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), and synthesis index (SI).

The rest of this paper is organized as follows. Section 2 briefly reviews the relevant literature on corrosion behavior of steel rebar in reinforced concrete structures as well as steel structures. The machine learners, meta ensembles, metaheuristic regression, and the method used for model performance evaluation are introduced in Section 3. Section 4 presents the results of numerical experiments using two real-world datasets. Finally, Section 5 concludes with remarks and discussions.