Fatigue life prediction of asphalt concrete using neural networks

The service life of asphalt concrete (AC) is strongly influenced by its resistance to fatigue. Traditional methods for fatigue life assessment are time consuming and require a significant amount of resources, and previous applications of artificial neural networks (ANNs) have not provided detailed descriptions of their architectures, nor have they used large datasets. Our study uses ANNs to predict the fatigue life of ACs, focusing on the level of overfitting, binder content, and spacing, and working with a dataset of 245 test fixtures. To enhance the prediction capability of ANN, we optimized the hyperparameters and used the log root mean square error as the loss function. Our results show improved fatigue life prediction accuracy, with special importance given to the binder content. This approach advances the capabilities and transparency of ANN application in AC fatigue life prediction.