Predicting the Fracture Characteristics of Concrete Using Ensemble and Meta-heuristic Algorithms

Abstract

Fracture properties are crucial for the design and application of concrete materials. The fracture energy which represents the fracture performance of concrete always suffers the influence of various factors, making it difficult to accurately predict the fracture energy of concrete with traditional methods. Thus, the artificial intelligence (AI) approaches are employed to establish the predictive models for the concrete fracture energy. Additionally, the firefly meta-heuristic algorithm (FA) is used to search for the optimum model hyper-parameters. The FA algorithm is proved to be efficient in tuning the hyper-parameters of the three models, and optimum values are obtained within ten iterations. Compared with the single classification and regression tree algorithm (CART), the ensemble algorithm appears to be more accurate and generalizable. Moreover, through the importance evaluation of different features in the optimum predictive model, the aggregate characteristics (aggregate size distribution and maximal aggregate size) and specimen size have been proven to be dominant factors for the predictive models, which should be carefully considered in predictive work regarding concrete fracture properties.