Implementing and Optimizing an Artificial Neural Network to Predict Fatigue Resistance of Asphalt Mixtures

As more agencies move toward implementing the balanced mix design (BMD) of asphalt mixtures, laboratory performances, including resistance to cracking, will be required to meet established criteria during mix design. There may be cases when several candidate mixes with different components need to be evaluated for fatigue cracking resistance prior to final mix design selection, which can be time-consuming and sometimes non-practical. To reduce the number of laboratory tests, a quick prediction method to identify the mixtures with better cracking resistance without need to complete the test is required. Artificial neural network has shown potential to predict certain properties of asphalt mixtures in recent research. This study investigated implementing one type of artificial neural network—convolutional neural network (CNN), to predict the cracking resistance of asphalt mixtures, represented by the overlay test based on the material properties and test setup parameters. The architecture of the CNN was established and optimized, with predicting capability tested and evaluated. The predicting quality of the CNN was slightly better than the linear regression model, demonstrating potential to preliminarily predict the cracking resistance so the laboratory testing effort can be reduced.