Neural Network Analysis of Chloride Diffusion in Concrete

Abstract

Chloride diffusion is one of the major causes of deterioration of concrete structures. A large amount of research has been conducted to study the chloride diffusion of concrete, both theoretically and experimentally. Because chloride diffusion experiments are time consuming, it is desired to develop a model to predict the chloride profiles in both steady and unsteady states. This paper studies the feasibility of using a neural network as an adaptive synthesizer as well as a predictor to meet such a requirement. The cascade-correlation algorithm is adopted in the present investigation to predict the chloride profiles diffused through concrete specimens. It is found that the predictions given by the cascade-correlation algorithm are in good agreement with the test results in both steady and unsteady states. The cascade-correlation algorithm can also synthesize an appropriate architecture and train the network simultaneously. It has the potential of becoming an effective tool in the prediction of durability problems. The investigation results demonstrate that the addition of fly ash and microsilica improves the resistance of mixtures to chloride diffusion. However, the addition of calcium nitrite solution degrades the improvement caused by the incorporation of these mineral admixtures, so calcium nitrite should be used in practical engineering with caution.