Prediction of frictional characteristics of bituminous mixes using group method of data handling and multigene symbolic genetic programming

Abstract

A safe road transport requires adequate friction between vehicle tires and pavement surface for safe travels. Adequate friction is essential for a vehicle to safely maneuver. Inadequate friction is directly correlated to the accident hazard, particularly in wet weather conditions. Quality of pavement materials has a prodigious effect on skid resistance and high-quality materials allow adequate pavement frictional resistance for an extended period. Science evaluation of frictional characteristics depends on physical, chemical, and mineralogical properties of the aggregate, type of mix, binder/bitumen content, water film thickness, etc., and it necessities a costly and time-consuming test protocol. In the present research, a model is developed for the evaluation of skid resistance in terms of the British pendulum number (BPN), using experimental observations, with the aid of machine learning tools. In the present work, group method of data handling (GMDH) and multigene symbolic genetic programming (MSGP) have been used to model the BPN. Developed model is capable to simplify extremely nonlinear deviations in data as well as forecast the frictional performance from experimental data. It is also found that the performance of the MSGP (R2 = 0.99) is more encouraging and better than that of the GMDH model (R2 = 0.98) for the prediction of BPN. The analytical expression obtained through MSGP in the present study has been also subjected to sensitivity analysis to assess the effect of individual parameters in prediction of BPN.