High-resolution flow field data are essential for accurately evaluating the aerodynamic performance of aircraft. However, acquiring such data via large-scale numerical simulations or wind tunnel experiments is highly resource-intensive. Flow field super-resolution techniques aim to reconstruct high-resolution information from low-resolution data, significantly improving data acquisition efficiency. With the rapid advancement of artificial intelligence, especially deep learning, neural network-based super-resolution methods have been widely adopted for flow field reconstruction. Nonetheless, these models typically rely on fixed end-to-end mappings and lack design space exploration capabilities. As a result, they may recover global flow structures but often fail to accurately capture local details, particularly in complex regions such as shock waves. Recently, generative diffusion models have shown promise in learning the error patterns between low-resolution and high-resolution data via noise modeling, enhancing the model's ability to recover local features.
Recently, a research team of Professor Rongqian Chen from Xiamen University, China, introduced a novel FlowViT-Diff framework that integrates a Vision Transformer (ViT) with an enhanced denoising diffusion probabilistic model for super-resolution reconstruction of high-resolution flow fields from low-resolution inputs. In this framework, the ViT's results serve as a prior to guide the diffusion model during reconstruction, enabling high-fidelity recovery of both global structures and local details. This approach effectively addresses the limitations of conventional methods in capturing localized flow features.
The team published their work in the Chinese Journal of Aeronautics on June 12, 2025.
In this study, the team first developed a ViT-based prediction model that leverages attention mechanisms to rapidly generate high-quality initial flow fields based on airfoil geometry and flow conditions. Subsequently, the noise injection and denoising processes of the diffusion model were optimized to integrate the ViT predictions, thereby enhancing the final high-resolution flow field generation. A series of experiments on supercritical airfoils under various downsampling rates were conducted to evaluate the framework's performance. Comparative analyses with classical super-resolution methods confirmed the superior accuracy and robustness of FlowViT-Diff in flow field reconstruction.
"This method fully leverages the complementary strengths of ViT and diffusion models," said Jinhua Lou. "The ViT provides a fast approximation of high-quality flow fields, though it may contain local detail errors. Our improved diffusion model uses the ViT output as the mean to guide the denoising process on low-resolution flow inputs, effectively correcting the ViT prediction and enhancing both global consistency and local detail accuracy."
Jinhua Lou further noted, "The proposed method supports two error correction strategies: directly reconstructing high-resolution flow fields from low-resolution inputs, or learning the residual error between them. Additionally, by incorporating transfer learning, the model can be rapidly generalized across multiple downsampling low-resolution flow fields, greatly enhancing its applicability."
"Compared to traditional methods, it delivers more consistent global and local features. With transfer learning, the training speed increased 3.6 times, and the model achieved a reconstruction accuracy of 99.7% under ultra-low downsampling." said Jinhua Lou.
"At present, FlowViT-Diff has demonstrated excellent performance in steady flow field reconstruction. However, its effectiveness in unsteady flow fields remains to be explored. The future work will focus on extending the framework to unsteady flow conditions, reconstructing 3D flow fields around full aircraft configurations, and improving reconstruction efficiency." said Professor Rongqian Chen.
Other contributors include Zelun Lin, Jiaqi Liu, Yue Bao, Hao Wu, Professor Yancheng You from the School of Aerospace Engineering at Xiamen University in Xiamen, China.
Original Source
Jinhua LOU, Rongqian CHEN, Zelun LIN, Jiaqi LIU, Yue BAO, Hao WU, Yancheng YOU. A general framework for airfoil flow field reconstruction based on transformer-guided diffusion models [J]. Chinese Journal of Aeronautics, 2025, https://doi.org/10.1016/j.cja.2025.103624.
About Chinese Journal of Aeronautics
Chinese Journal of Aeronautics (CJA) is an open access, peer-reviewed international journal covering all aspects of aerospace engineering, monthly published by Elsevier. The Journal reports the scientific and technological achievements and frontiers in aeronautic engineering and astronautic engineering, in both theory and practice. CJA is indexed in SCI (IF = 5.7, Q1), EI, IAA, AJ, CSA, Scopus.
