LAUSR

Abstract The fracture properties (e.g. tensile strength, fracture toughness) are regarded as the material constants, which could be used to predict the fracture behavior of concrete. However, the scattered distribution and size effect of test results make the prediction in conventional methods difficult. In this study, the artificial intelligence (AI) approaches and the Boundary Effect Model (BEM) closed-form solution were tried to analyze such complex relations between the fracture test results and specimen geometries in the 3 PB tests. Firstly, a cluster of data were collected and divided into the training set and testing set. Then, based on the training data, the ensemble algorithm (Random Forests) and the Particle Swarm Optimization (PSO) were combined to establish the hybrid AI predictive model, and the fracture properties and predictive domain were determined with the BEM closed-form solution and normal distribution analysis. After that, the testing data were used to evaluate the behavior of these two predictive methods. The performance of the AI predictive model was quantified with R2 = 0.947, and the unknown data in testing set all fell into the predicted domain which was determined by using the BEM predictive model. The merits of the two predictive methods in predicting the fracture performance of concrete specimens were compared and expected to integrate together in the future work.