Employing a hybrid GA-ANN method for simulating fracture toughness of RCC mixture containing waste materials

The present study is conducted to investigate the efficiency of evolutionary algorithms such as genetic algorithm (GA)-evolved neural network in estimating fracture properties of roller compacted concrete pavement (RCCP) mixtures with different compositions. The effect of waste materials, including reclaimed asphalt pavement (RAP) and crumb rubber on fracture toughness in both pure mode I and pure mod II, were investigated using a real coded GA and an evolution with a back-propagation algorithm. Moreover, the geometry effect of SCB and 4 PB specimens on predicting fracture toughness was evaluated using GA. To evaluate the GA-based neural network's performance, the NSE criterion was applied for fitness function, a different approach for fitting in this area. Many researchers have studied the fracture behavior of RCC in diverse modes. Still, their studies have been restricted to the materials' fracture behavior, which has rarely come to a model. As pure mode II fracture toughness experiment is a complicated procedure with high uncertainties, the introduced model provides a powerful tool for predicting mode II fracture toughness via mix composition and pure mode I fracture toughness of RCC based on GA. The proposed model outperforms the traditional artificial neural network (ANN) models. The geometry effect survey on predicting fracture toughness shows that the 4 PB fracture model has been better fitted to observed data. Also, the MSE results show that prismatic specimens present relatively reliable results than SCB specimens.