Friction Coefficient Prediction of Asphalt Pavement Based on Point Cloud Data in Different Water-film Thickness

The skid resistance performance of asphalt pavements is closely related to road traffic safety. To study the skid resistance performance of asphalt pavements and their influencing factors, this paper proposes a deep learning-based model for predicting the friction coefficient of asphalt pavements. High-precision three-dimensional laser equipment Gocator 3110 was used to scan the three-dimensional texture of asphalt mixture specimens, and the British Pendulum Tester (BPT) was utilized to measure the surface friction coefficient of the specimens. Quality improvement measures were applied to address issues such as missing and distorted three-dimensional texture point cloud data, and correlation analysis was employed to select macro-texture features that significantly impact the skid resistance performance of asphalt pavements. Ultimately, a self-attention mechanism was introduced to uncover potential correlations among samples, resulting in the development of a friction coefficient prediction model for asphalt pavements. Compared to two models, XGBoost (eXtreme Gradient Boosting) and LightGBM (Light Gradient Boosting Machine), the proposed model demonstrates superior performance in prediction accuracy, achieving an R² value of up to 98.4%. This value indicates that the model can effectively predict the friction coefficient of asphalt pavements under different water-film thickness conditions. This model provides an effective scientific tool for assessing the skid resistance performance of asphalt pavements, thereby offering strong data support for the field of road traffic safety.