Prediction of a Healing Index of Asphalt Concrete Mixture with Artificial Neural Network

Roads are one of the most critical assets of any country, and a large part of the country's budget is spent annually on repair and maintenance operations to eliminate cracks. One of the factors that can effectively increase the useful life of asphalt pavement is the self-healing potential of asphalt mixtures. This study considers the factors affecting the self-healing index (such as type of additive, percentage of additive, gradation of aggregate, type of bitumen, crack repair cycle, type of heating, and heating time), and using a neural network model to predict this index. For this purpose, multilayer perceptron neural network (MLP), m multilayer perceptron neural network with particle swarm optimized algorithm (PSO), radial basis function neural network (RBF), and statistical analysis with SPSS software were used, and the results of these methods were compared. The results showed that the multilayer perceptron neural network (MLP) with a correlation coefficient of 0.96 has a better performance in predicting the self-healing index than other methods and to evaluate the generalization power of the neural network using data that were not used during modelling, multilayer perceptron neural network (MLP) and radial basis function neural network (RBF) work best. Also, three samples were made in the laboratory by replacing 60% of steel slag with coarse aggregates (4.75-9.5 mm), and their self-healing results were evaluated with neural networks.