A novel tire-pavement related parameter for improved rolling resistance predictions

Abstract

Accurate prediction of rolling resistance (RR) is essential for improving vehicle fuel efficiency and supporting policymakers in making sustainable environmental decisions. This study introduces a novel framework that integrates both data-driven and physics-based approaches to enhance RR prediction by incorporating tire-penetration level indicator, the Delta (δ) parameter. The research investigates the relationships between RR, the δ parameter, and texture properties to refine predictive modelling. A portable device was built to measure the in-field δ parameter using tire-pavement interaction. Machine learning (ML) techniques, including multiple linear regression (MLR), random forest regressor (RFR), artificial neural networks (ANN) and finite element method-based (FEM) tire-pavement interaction models were employed to develop and validate the framework. Findings from the FEM tire-pavement interaction model confirmed the reliability of the δ parameter. Exploratory data analysis (EDA) highlighted the strong correlation between texture metrices such as MPD, ETD, and RMS, reinforcing the δ parameter's role in tire-pavement interactions. Comparative analysis of different pavement surfaces revealed that worn surfaces contribute to higher δ parameter values and increased RR. The improvement resulting from the inclusion of the δ parameter is particularly evident in the ANN and RF models, confirming nonlinear interaction effects between tire penetration and surface texture. It was also observed that the obtained RR data follow a non-normal distribution, which most of the previous studies did not consider. A deeper statistical insight showed that the δ parameter has a significant impact on RRC prediction. The primary contribution of this study lies in demonstrating the feasibility of integrating a physics-based tire-pavement interaction parameter into ML models for rolling resistance prediction, thereby bridging mechanistic modelling and machine learning within pavement engineering.