Neural network modelling of chloride binding

Abstract

The capacity of cement to bind chloride is considered to be one of the controlling parameters in the process of chloride-induced corrosion of steel in concrete. In this work, a literature review has been undertaken to identify factors affecting chloride binding. Data from 21 previously published works have been collated and used to develop a neural network model to predict the free chloride concentration as a function of 18 input variables. These predictions are relatively free of the indiscriminate variations present in individual measurements. The influence of factors not yet fully evaluated is identified. In addition, the relative importance of a wide range of factors is quantified. The neural network model provides a powerful tool for generating binding isotherms for use in models ofchloride ingress into concrete.