Neural network-based prediction of sideway force coefficient for asphalt pavement using high-resolution 3D texture data

The pavement friction has been recognised as a critical contributor to traffic safety and an important pavement functional characteristic. The sideway force coefficient routine investigation machine (SCRIM) has been used in many studies for pavement friction evaluation. However, this device is costly to purchase and run which limits its accessibility to many transportation agencies. This study explores an artificial neural network model to predict the sideway force coefficient (SFC) from SCRIM using pavement micro- and macro-texture information. On the selected field site, pavement texture was evaluated by a digital sand patch tester and a portable high-resolution 3D laser scanner, while pavement friction was measured using a SCRIM. The obtained high-resolution 3D texture data was decomposed into micro- and macro-textures via discrete Fourier transform and Butterworth filters. Then, height, feature, and hybrid texture parameters were calculated to characterise pavement 3D texture at the micro- and macro-levels. Next, the obtained pavement texture parameters were used to predict SFC through linear and neural network models. The neural network model, including pavement micro- and macro-texture parameters, shows better performance than other models and is adequate to predict SFC. Besides, pavement micro-texture shows more contribution to SFC than macro-texture.