Crack width in concrete using artificial neural networks

Most of the rules for predicting the crack width of reinforced concrete structures, in existing building codes, are based on the statistical results obtained for normal strength concrete (NSC) members with normal concrete cover. Therefore, these rules need to be adjusted for high strength concrete (HSC) members with thick concrete cover. This paper presents a method for the use of neural networks for the proper estimation of crack width in thick concrete elements at the serviceability stress limit state stated by ACI 318-08. Two kinds of neural networks were used: the radial basis and the feed forward back propagation neural networks. It has been showed that both types of neural networks yield better results than results obtained using existing building codes’ rules. The radial basis neural network needs smaller design and training time and provides better results than the classical feed forward back propagation neural network. sults of the present study show that predictions of the average crack width for both thick and thin concrete members using neural networks are more accurate than those results obtained using the rules in existing building codes. There is good agreement between the neural networks results and experimental results.