Predicting Marshal Stability And Marshal Flow Usnig Artificial Intelligence: A Comparison Of Gene Expression Programming and Artificial Neural Network

Abstract

The type and proportion of materials being used in the construction of a highway facility, as well as many other criteria, influence its comfort, ride quality, and service life. While constructing the bituminous layers of a highway facility, the type and composition of mixes must be carefully considered. These layers must be built with certain care since they are directly impacted by the applied load and environmental conditions. Properties of these layers are affected by number of factors which are described in the form of Marshal Quotient of hot mix asphalt. Calculating the marshal quotient leads the project to an uneconomical as calculating this parameter is based on trails and error and requires skilled labor and extensive time for calculation. A computer-based model has been developed using a data set composed of 110 lab experiments collected from a construction firm working on Jehagira to Risalpur road (Khyber Pakhtunkhwa, Pakistan) that can predict the values of marshal stability and marshal flow. The collected data was first screened and all the inappropriate data points were removed. Prior to modeling, insignificant variables were removed to generate a better model. Models were developed using GEP and ANN for both Marshal Stability and Marshal Flow using seven input variables. The performance of the models developed has been validated using coefficient of determination, RMSE, MAE and Adjusted R2. Results shows that the GEP model performs better than ANN and has more better predicting power than ANN. Performance of the developed model was also validated using unseen data collected from N95 Swat (Khyber Pakhtunkhwa, Pakistan) where the models performed significant and were able to predict the output quiet accurately. A sensitivity analysis has also been performed to access the relative contribution of every variable in predicting the outputs. It was also concluded that the marshal stability increases with the increase in air voids and reduction in bitumen content while marshal flow increases upon increase in bitumen content and decrease in air voids.