Explainable machine learning-enhanced aerodynamic characteristic analysis of bluff bodies under interference effects

Aerodynamic interference has a critical impact on wind-resistant design. This study presents a computational fluid dynamics (CFD)-based explainable machine learning framework to predict and interpret the aerodynamic characteristics of bluff bodies under interference effects. A large-scale two-dimensional CFD dataset comprising 3360 three-bluff-body interference case configurations was established, covering 140 layouts and 24 wind angles. Detailed aerodynamic analysis shows that the observed behavior does not arise from a single interference mechanism like shielding effects, channeling effects, or wake interference, but rather from their coupled action. Their interactions reshape pressure distributions and global aerodynamic forces across different interference layouts. Based on this dataset, eight machine learning (ML) models were benchmarked for predicting aerodynamic force coefficients, among which the random forest (RF) model achieved the highest accuracy, yielding an R² of 0.962. To enhance interpretability, global, conditional, and local SHapley Additive exPlanations (SHAP) analyses were conducted to quantify the contributions of wind angle of attack and interference location features to the aerodynamic loads, thereby linking dominant features to the underlying flow mechanisms. In addition, the RF model was further applied to predict surface pressure distributions with an R 2 of up to 0.974, and SHAP analysis was performed to quantify the influence of interference location parameters on representative pressure points. Furthermore, uncertainty quantification was conducted to evaluate prediction reliability, providing confidence estimates alongside aerodynamic predictions for individual interference configurations. The proposed framework enables rapid and interpretable assessment for large numbers of interference scenarios, facilitating aerodynamic evaluation during early-stage design.