PE-WFormer: Physical enhanced sparse data-driven wind field reconstruction for bluff body

Wind field distribution around bluff bodies is a critical factor in structural wind resistance design and aerodynamic performance evaluation. Traditional computational fluid dynamics (CFD) methods face challenges such as high computational cost, long test cycles, and difficulty in capturing full-field dynamic characteristics. Physics-Informed Neural Networks (PINNs) have emerged as a promising tool for getting fluid dynamics via sparse data, but conventional MLP-based PINNs neglect temporal dependencies in unsteady bluff body flow, leading to inaccurate full-field reconstruction from sparse data. In this paper, we propose an improved data-driven framework tailored for bluff body flow reconstruction via sparse data, leveraging Transformer’s strong capability in capturing spatiotemporal dependencies to address the limitations of traditional PINNs. We validate the proposed framework using CFD data of a typical bluff body. Experimental results demonstrate that the proposed framework outperforms conventional PINNs, in terms of reconstruction accuracy. This work provides a high-efficiency, high-accuracy data-driven method for wind field reconstruction of bluff body, offering practical support for structural wind resistance design and aerodynamic optimization.