Estimating Fracture Eenergy of Concrete Using Artificial Neural Networks

Concrete is one of the most common industrial and constructional materials due to its economical aspects. In recent years, fracture parameters of cement-based materials, such as concrete have been studied using the development of mathematical models derived from various experimental data and methods. The role of these parameters in design of structures is an important issue. In this paper, a fracture model based on artificial neural network, as an alternative way and soft computing to reduce time and calculation content, has been applied to estimate the fracture parameters of concrete (specific fracture energy i.e. the area under the complete stress- strain curve) by loading under a three-point bending (3PB) specimen. Fracture parameters of concrete, model’s output parameter, is predicted by input parameters: test age, aggregate type, maximum aggregate size (dmax), compressive-tensile strengths ratio (f´c/f´t), water-cement ratio (w/c), young’s modulus (E). Because of the fewness of data, thirty nine data were used as training data and eight data were used as validation or testing data. By using the neural network and its proper training to create logical relationship between input and output variables, an optimal model for each series of data could be made, which can be applied to evaluate the network reliability as an effective tool to estimate the fracture energy of concrete. The results showed that the mathematical models can be used to estimate the fracture parameters of cement-based materials and concrete. Also it was shown that the feed-forward multi-layer perception back-propagation artificial neural network with Lovenberg-Marquardt training algorithm and with 2 layers which has 6 and 1 neurons in the first and second layer respectively with R=0.90 and RMSE=0.04 is most appropriate neural network for the estimation of fracture energy of concrete.