BIM and data-driven multi-objective optimization of asphalt pavement structure combinations

To address low modeling efficiency and multiple design factors affecting pavement performance, an integrated data-driven method combining building information modeling (BIM), finite element method (FEM), and deep learning (DL) for optimizing asphalt pavement design is proposed. The rapid BIM-FEM interaction enables quick modeling and calculations of rutting and fatigue life, creating a DL database. A convolutional neural network (CNN), temporal convolutional network (TCN), and attention mechanisms (CNN-TCN-Attention) models that captures complex nonlinear relationships are proposed for accurate pavement performance prediction. Subsequently, an improved multi-objective evolutionary algorithm based on decomposition (MOEA/D) dynamically adjusts neighborhood sizes are developed to optimize design features. Case study indicates that BIM-FEM framework improves modeling efficiency by 68.66 %, while CNN-TCN-Attention model achieved precise predictions for pavement performance. After optimization, rutting decreased by 10.03 mm and fatigue life increased by 0.67 billion cycles. This method holds potential for road structure health monitoring and digital twin applications.