Prediction of chloride diffusivity in concrete using artificial neural network: Modelling and performance evaluation

Chloride ingression is the main reason for causing durability degradation of reinforced concrete (RC) structures. In this study, the distinguishing features of artificial neural network (ANN) technique are utilized to develop a rational and effective predictive model for chloride diffusion coefficient of concrete. An extensive and reliable database comprising of 653 distinctive diffusion coefficient results, from literature, was utilized for establishing the network model. The developed ANN models used 13 most influential parameters, varying from concrete constituents, mechanical property and experimental process, as input to incorporate complex underlying physical phenomena for prediction of diffusion coefficient. The significance of normalization of the variables is highlighted through a comparative study. The performance of the developed model is assessed by conducting several in-depth statistical error analysis as per the recommendations in literature. The results of the study reveals that the models are robust and possess a strong prediction potential. The findings revealed that ANN can be an effective tool to identify the discrepancies in the experimental findings, and would be particularly useful for evaluating the chloride resistance of RC structures serving in complex or harsh environment. (C) 2020 Elsevier Ltd. All rights reserved.

Automated summary

This study developed a predictive model for chloride diffusion coefficient of concrete using artificial neural network (ANN) technique, which showed strong prediction potential. The performance of the model was assessed through normalization of variables and in-depth statistical error analysis, indicating its robustness and reliability.