Size-dependent aerodynamic interference between adjacent bluff bodies: A two-stage generative adversarial network-super-resolution approach

Abstract

Accurate prediction of wind-induced interference effects in building groups is essential for wind-resistant design, yet traditional approaches based on computational fluid dynamics (CFD) or wind tunnel tests are prohibitively expensive for large-scale parametric investigations. This study presents a deep learning-aided framework for rapid prediction and analysis of high-resolution wind pressure fields, enabling systematic investigation of aerodynamic interference effects between adjacent non-identical square-sectioned bluff bodies. Given building positions and section sizes as inputs, the framework first predicts a pressure-field image at 128 × 64 resolution using a conditional generative adversarial network. A geometry-conditioned super-resolution network then refines this prediction and reconstructs the pressure-field representation at 512 × 256 resolution using the corresponding high-resolution CFD contour as the supervised target. The model was trained on a dataset comprising 12 758 cases covering size ratios from 0.5 to 2.0 and diverse relative building configurations. Parametric analyses reveal that first, upstream interfering buildings can influence downstream structures at separations of up to 20 times the characteristic building dimension, with surface pressure reductions exceeding 70%. Second, side-by-side arrangements produce the strongest interference effects throughout the flow field, whereas tandem configurations with the interfering building located three times the characteristic building dimension downstream exhibit the weakest effects. Third, the size-dependent variation of interfering building is the dominant parameter governing spatiotemporal evolution of vortex shedding, thereby controlling both the intensity and spatial extent of aerodynamic interference. The proposed framework provides an efficient geometry-conditioned surrogate for high-resolution pressure-field image reconstruction and supports the investigation of size-dependent interference trends within building groups.