Using artificial neural network to predict the fracture properties of the interfacial transition zone of concrete at the meso-scale

Abstract Concrete is a multi-phase heterogeneous material in which the interfacial transition zone (ITZ) between aggregates and mortar significantly affects the cracking behaviour of concrete, especially under tensile load. In this paper, the artificial neural network (ANN) method is applied in predicting the fracture properties of ITZ in concrete. To form the data pool for the training of the ANN, a large number of two-dimensional (2D) meso-scale fracture simulation of concrete under direct tensile load is conducted. Cohesive crack elements are used in simulating arbitrary cracking in the ITZ and mortar. After the verification of the trained ANN model, the tensile strength and fracture energy of the concrete ITZ are predicted by using the RILEM direct tensile test results, i.e., stress–displacement curve, as the input for the ANN. It has been found that, the trained ANN performs well in predicting the ITZ properties and the computed stress–displacement curve together with the optimized ITZ fracture parameters has a good agreement with that from the RILEM test results. The randomness of aggregates has little effect on the predicted ITZ tensile strength while it becomes slightly bigger on the predicted ITZ fracture energy. The tensile strength ratio of ITZ to mortar are calculated 0.38–0.47 and the fracture energy ratio of ITZ to mortar are 0.18–0.58. These inversely predicted fracture properties of concrete ITZ can be useful complementation to the existing dataset and 2D fracture simulation of concrete structures.