Evaluate asphalt pavement frictional characteristics based on IGWO-NGBoost using 3D macro-texture data

Abstract

Friction, or skid resistance, is an important feature of road pavements. It is influenced by a combination of factors. Accurately predicting the friction coefficient and thus effectively assessing the skid resistance performance is a complex nonlinear problem. For multi-feature and non-linear data, the accuracy of traditional methods is limited. In addition, the prediction accuracy of some methods depends on the original sample size, so it is difficult to achieve accurate prediction under the condition of small sample size. To solve the above problem, the friction coefficient prediction model based on the improved gray wolf optimization (IGWO) and natural gradient boosting (NGBoost) is proposed. First, to ensure the diversity of samples, asphalt mixture specimens with different gradation types are made. Then, friction and three-dimensional (3D) macro-texture data are collected from the specimen surfaces. Next, twenty-seven 3D macro-texture features are extracted from macro-texture data to describe macro-texture details. A correlation coefficient evaluation method is used to eliminate redundant features, and a feature importance analysis model based on gradient boosting model is constructed to obtain the key factors affecting the skid resistance. Finally, the friction coefficient prediction model based on IGWO-NGBoost is constructed. The IGWO algorithm is used to adjust the hyperparameters of NGBoost to optimize the model structure. The results show that IGWO-NGBoost can effectively fit the friction coefficient with a goodness of fit R2 of 97.31% compared with state-of-the-art methods. The model can effectively analyze the change mechanism of pavement skid resistance under the combined influence of multiple factors.