Presumption of shear strength of steel fiber reinforced concrete beam using artificial neural network model

The shear force characteristics of steel fiber reinforced concrete (SFRC) are investigated with varying shapes and mixture ratios. However, because experimental characterization of SFRC is experimentally demanding in terms of time and equipment, the characterized SFRC data are used with limitation. Therefore, for various applications, an easier approach is required to predict the shear force characteristics of unsaturated soils. In consideration of such a situation, a method to ascertain the shear force characteristics of SFRC is suggested and applied to this study as a neural network theory. The backpropagation algorithm is applied as a learning algorithm for a neural network, and learning is performed in order to converge within an error range of 0.001. In addition, a nonlinear function is used as an objective function and the problem of overfitting is resolved with a more generalized method by adopting the Bayesian regularization technique as a generalization process. To identify the reliability of this artificial neural network model, we compare values from the shear strength test of SFRC beams with the values from the model. They show correspondence between them. Therefore, it is concluded that, if many test variables and data are used as input for learning in the neural network model developed in this study, it is possible to attain a much more reliable prediction. © 2006 Wiley Periodicals, Inc. J Appl Polym Sci 103: 2351–2358, 2007