Prediction of Tension Softening Curve in Concrete Using Artificial Neural Networks

Abstract

Knowledge of the tension softening process of concrete is essential to understand fracture mechanism, further to analyze fracture behaviour, and further to evaluate properties of concrete. For the last eight years, many different tests on uniaxial tension with elimination of secondary flexure were performed in Tohoku Institute of Technology. The paper is dedicated to predict tension softening curve of concrete by using artificial neural networks (ANNs) based on experimental data of five different mixtures of concrete (including High Performance Concrete). It is an advantage to predict a proper tension softening curve without performing uniaxial tension tests. Several artificial neural networks with different architectures (with various hidden neurons and layers) were studied using software - Statistica Neural Network. In order to evaluate the prediction accuracy, tension softening curve and other fracture parameters were predicted for each mix from the other four mixes and compared with the omitted data of the relevant mix. High accuracy was obtained in the all predicted tension softening curves and the fracture parameters were also well predicted.