Cracking performance evaluation and modelling of RAP mixtures containing different recycled materials using deep neural network model
Meisam Khorshidi, Mahmoud Ameri, Ahmad Goli
Abstract
This study evaluates the cracking resistance of recycled asphalt pavement (RAP) mixtures including waste engine oil (WEO), crumb rubber (CR), and steel slag aggregates using the Illinois flexibility index test (I-FIT). Performance indices, derived from both this study and another, were predicted by comparing deep neural network (DNN), linear, and polynomial regression models via a k-fold cross-validation process. I-FIT test results demonstrated that WEO, steel slag aggregates, and specific CR proportions enhance cracking resistance while RAP utilisation decreased it. In terms of modelling, it was found that the most appropriate prediction model for the dataset structure of this study is the deep neural network model. The DNN model sensitivity analysis identified WEO as key for high and intermediate temperature (I-FIT) performance. Meanwhile, CR significantly impacted intermediate temperatures (IDEAL-CT), while RAP influenced moisture susceptibility. This model proves reliable and efficient, suggesting its potential for predicting the performance of recycled mixtures.