Prediction of Interface Shear Stiffness Modulus of Asphalt Pavement using Bagging Ensemble-based Hybrid Machine Learning Model

Interface shear stiffness modulus (K) is one of the important bonding properties of layers. It is also used to evaluate interface shear strength between asphalt layers of asphalt pavement. Direct determination of K parameter in field or laboratory requires time, cost and special equipment. In this article, K has been estimated based on three interlayer shear strength affecting factors namely maximum size of the asphalt concrete aggregate (D-max), normal pressure and temperature using Machine Learning (ML) methods such as Multilayer Perception Neural Network, Bagging Random Forest (Bagging-RF), and Bagging Reduced Error Pruning Tree (Bagging-REPT). The ML models for the prediction of shear strength were built based on the laboratory shear tests results of 180 double-layer asphalt samples. The data was divided randomly into a ratio of 70/30 to train and test model, respectively. Standard statistical measures were used to evaluate and validate the models' performance. All the developed models performed well in correctly predicting K value of AC, but performance of the Bagging-RF model is the best as it is giving Correlation Coefficient (R) value 0.88 between estimated value and determined value. The proposed ML predictive models will reduce the field and laboratory experimental efforts and increase the efficiency in estimating the K parameter for the safe designing, construction and maintenance of asphalt concrete pavements.

Automated summary

In this article, the shear stiffness modulus (K) of asphalt layers was estimated based on three factors affecting interlayer shear strength. Machine Learning (ML) methods were used to build the predictive models, which performed well in correctly predicting the K value. Among the models, Bagging-RF showed the best performance with a correlation coefficient of 0.88 between estimated and determined values. These ML models will reduce the experimental efforts and increase the efficiency in estimating the K parameter for asphalt pavement design, construction, and maintenance.