Prediction and optimization of tire-pavement noise for ultra-thin friction course based on surface texture

Abstract

Tire–road noise is a major component of traffic noise and becomes increasingly dominant under medium- and high-speed driving conditions. Pavement surface texture plays a critical role in tire–road noise generation; however, existing prediction approaches remain limited in capturing complex nonlinear texture–noise relationships and supporting inverse texture optimization. To address these limitations, this study focused on ultra-thin friction course (UTFC) pavements and established a UTFC-focused dataset, supplemented by conventional asphalt pavement sections as references, linking multiscale surface texture characteristics with tire–road noise. Three-dimensional laser scanning was used to acquire pavement point-cloud data, from which representative texture parameters were extracted. Four ensemble-learning models, including Random Forest (RF), Gradient Boosting Decision Tree (GBDT), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), were developed and compared for tire–road noise prediction. Optuna was subsequently employed for hyperparameter optimization, and a genetic algorithm (GA) was integrated with the best-performing model to establish a data-driven inverse optimization framework for low-noise texture design. The results showed that RF achieved the best predictive performance among the four models, with a test-set R² of 0.9432. After Optuna optimization, the RF model performance further improved, with R² increasing to 0.9648 and MSE decreasing from 0.3091 to 0.1918. Based on the optimized model, the GA–Optuna–RF framework identified a target texture parameter combination corresponding to a model-predicted tire–road noise level of 90.3 dB(A), 4.2 dB(A) lower than the dataset average. These results demonstrate the potential of integrating multiscale texture characterization, nonlinear machine-learning prediction, and inverse optimization to support low-noise UTFC texture design. The proposed framework provides a data-driven basis for identifying favorable texture characteristics, while further experimental validation is required before direct engineering application.