Elsevier

Measurement

Volume 112, December 2017, Pages 141-149
Measurement

Prediction of chloride diffusion in cement mortar using Multi-Gene Genetic Programming and Multivariate Adaptive Regression Splines

https://doi.org/10.1016/j.measurement.2017.08.031Get rights and content

Highlights

  • Machine learning (ML) models for chloride diffusion prediction are proposed.

  • A data set of mortar specimens is collected to construct ML models.

  • ML based equations can accurately predict the chloride ion diffusion.

  • The best model achieves RMSE = 0.70 and R2 = 0.91.

Abstract

Chloride-induced damage of coastal concrete structure leads to serious structural deterioration. Thus, chloride content in concrete is a crucial parameter for determining the corrosion state. This study aims at establishing machine learning models for chloride diffusion prediction with the utilizations of the Multi-Gene Genetic Programming (MGGP) and Multivariate Adaptive Regression Splines (MARS). MGGP and MARS are well-established methods to construct predictive modeling equations from experimental data. These modeling equations can be used to express the relationship between the chloride ion diffusion in concrete and its influencing factors. Moreover, a data set, which contains 132 cement mortar specimens, has been collected for this study to train and verify the machine learning approaches. The prediction results of MGGP and MARS are compared with those of the Artificial Neural Network and Least Squares Support Vector Regression. Notably, MARS demonstrates the best prediction performance with the Root Mean Squared Error (RMSE) = 0.70 and the coefficient of determination (R2) = 0.91.

Introduction

The durability of concrete structure has always been a critical concern in port and marine engineering [1]. Among all the factors affecting the reinforced concrete durability, corrosion of reinforcement is often considered as the most influential factor [2], [3], [4]. For marine concrete, chloride, which is dissolved in the surrounding environment, gradually penetrates into the structure. Accordingly, steel reinforcements in the structure are subject to corrosion when the chloride content reaches a sufficiently high level [5], [6].

In the case of chloride ingression, if no timely maintenance measure is carried out, the diffusion of chloride ion can cause serious consequences for the strength and esthetics of the structure, resulting in the reduction of the service life of structure [7], [8]. Noticeably, chloride-induced damage may trigger critical failure of the structure within a relatively short amount of time. Therefore, the study on chloride ion ingression in concrete is of practical need for better ensuring the durability of reinforced concrete structure.

Notably, the ability of ensuring the durability and service life of reinforced concrete structure in marine environment depends upon the accuracy in predicting chloride diffusion in concrete [9]. This prediction of chloride diffusion can help to formulate predictive concrete deterioration model. Based on that, cost-effective strategy can be made regarding the appropriate time of repairing or replacing the degraded structural elements [8].

Conventional prediction approach based on Fick’s second law of diffusion is commonly used to estimate the chloride diffusion process [10]. Nevertheless, this traditional formula-based method suffers from severe drawbacks such as the difficulty of parameter estimation [11] and unsatisfactory prediction accuracy [12], [13]. The reason is that the dependence between chloride diffusion in concrete and its conditioning factors is inherently complex and time-dependent [11], [14]. These facts demand more advanced tools for modeling the phenomenon of chloride ion diffusion in concrete.

Hodhod and Ahmed [15] attempted to model the chloride diffusivity process in high performance concrete with the application of an Artificial Neural Network (ANN). Eskandari, Nik and Eidi [16] constructed an ANN-based inference model for estimating compressive strength of mortar in marine environment. Intelligent models for predicting chloride content in concrete based on an ANN and regression tree have been examined by Asghshahr, Rahai and Ashrafi [17]. Liao, Chen, Wu, Chen and Yeh [12] studied the chloride diffusion in cement mortar by means of the Least Squares Support Vector Regression. Recent applications indicate that advanced machine learning methods provide much better tools for characterizing the chloride diffusion process [18].

Nevertheless, since the chloride ion diffusion in cement mortar is indeed a complex phenomenon, other advanced machine learning approaches should be investigated for tackling with the problem of interest. Moreover, most studies employed learning algorithms that cannot yield explicit model structures. The current study attempts to fill this gap in the literature by examining the possibility of constructing chloride diffusion modeling equations from experimental data. These modeling equations can provide convenient tools for researchers and engineers to express the relationship between the chloride ion diffusion in concrete and its conditioning variables.

The Multi-Gene Genetic Programming (MGGP) and the Multivariate Adaptive Regression Splines (MARS) are selected in this research due to their successful applications in other fields of study [19], [20]. Furthermore, to train and validate the prediction models, a data set including 132 cement mortar specimens in simulated marine environment has been collected. The rest of the paper is organized as follows: the second section describes the research method; the experimental setting and results are reported in the next section; the final part provides some conclusions on this study.

Section snippets

The experimental data set

To establish a data set for constructing and verifying the machine learning solutions, a total number of 132 mortar specimens with different features has been prepared. It is noted that the fresh mixing water is in compliance with the specifications of ASTM C494 – Standard specification for chemical admixtures for concrete [21]. The mortar specimens are made with Portland type I cement, which is complied with the specifications of ASTM C150 – Standard specification for Portland cement [22]. The

Experimental setting

The purpose of this section is to construct the modeling equation from experimental data. The two machine learning approaches (MGGP and MARS) are employed to establish the mapping function that determine the relationship between input variables (the mortar age, the depth of measured point, the diffusion dimension, and the presence of reinforcement) and the targeted output (the chloride ion concentration). It is noted that before the training and prediction phases, the variables within the data

Conclusion

This study investigates the possibility of employing machine learning algorithms, including MGGP and MARS, for constructing prediction equations for the modeling of the diffusion of chloride ion in cement mortar. To train and verify these machine learning approaches, a data set containing 132 records of mortar specimens has been collected. The mortar age, the depth of measured point, the diffusion dimension, and the presence of reinforcement have been employed to characterize the chloride ion

References (42)

  • S. Chithra et al.

    A comparative study on the compressive strength prediction models for high performance concrete containing nano silica and copper slag using regression analysis and Artificial Neural Networks

    Constr. Build. Mater.

    (2016)
  • Y. Wang et al.

    Experiments of chloride ingression in loaded concrete members under the marine environment

    J. Mater. Civ. Eng.

    (2014)
  • Z. Ma et al.

    Effects of hydrostatic pressure on chloride ion penetration into concrete

    Mag. Concrete Res.

    (2016)
  • E. Ryan et al.

    Comparison of two methods to assess the resistance of concrete to chloride ion penetration

    J. Mater. Civ. Eng.

    (2014)
  • S. Hong et al.

    Experimental monitoring of chloride-induced reinforcement corrosion and chloride contamination in concrete with ground-penetrating radar

    Struct. Infrastruct. Eng.

    (2015)
  • S. Caré et al.

    Application of a n-phase model to the diffusion coefficient of chloride in mortar

    Transport Porous Med.

    (2004)
  • H. Zhang et al.

    Chloride penetration in concrete under marine atmospheric environment – analysis of the influencing factors

    Struct. Infrastruct. Eng.

    (2016)
  • C. Sosdean et al.

    Experimental and numerical determination of the chloride penetration in cracked mortar specimens

    Eur. J. Environ. Civ. Eng.

    (2016)
  • H.-L. Wang et al.

    Time-dependent and stress-dependent chloride diffusivity of concrete subjected to sustained compressive loading

    J. Mater. Civ. Eng.

    (2016)
  • K.-W. Liao et al.

    Investigation of chloride diffusion in cement mortar via statistical learning theory

    Mag. Concrete Res.

    (2016)
  • J. Liu, F. Xing, B.Q. Dong, H.Y. Ma, D. Pan, New equation for description of chloride ions diffusion in concrete under...
  • Cited by (73)

    View all citing articles on Scopus
    View full text