Prediction model of chloride diffusion coefficients for concrete based on RBF neural network

Corrosion of reinforced concrete is a chronic infrastructure problem, particularly in areas with deicing salt and marine exposure. And the diffusion behavior of the chloride ions in concrete is a more complex and complicated transport process than what can be described by Fick´s law of diffusion. To maintain structural integrity, a prediction model of radial basis function (RBF) network is presented to predict the chloride diffusion coefficient of concrete in this paper. Three influence factors, water-cement ratio, cement content and cement admixture ratio are chosen as input vectors, and the chloride diffusion coefficient is defined as output vector. The prediction results based on the RBF model are compared with that from three other RBF network models, BP network model, and the experimental data. It can be seen that, the prediction accuracy of the present model is higher than that of other models. The key point for developing a neural network model is to choose the input vectors. Both the relative amount and absolute amount of influence factors should be taken into account. Moreover, the independence of input vector are also ought to be considered to avoid data binding. It is concluded that RBF neural network is absolutely a new method to evaluated chloride diffusion coefficient in concrete, and has promising applications in durability problems of concrete structures.