Generative Inverse Aerodynamic Design of a Single-Stage Turbine Using Conditional Denoising Diffusion Probabilistic Model
Journal of Turbomachinery · 01.04.2026 · 148 (4)
DOI: 10.1115/1.4069801 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Turbomachinery, online erschienen am 23.10.2025. Verfasst von Yueteng Wu und 5 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 43 Literaturangaben.
Abstract (Originalsprache)
The aerodynamic design of turbomachinery is crucial for the overall performance of aero engines and has long been a paramount field of research. Traditional design methods are categorized into forward and inverse designs. Forward design relies on predefined geometric parameters and iteratively adjusts them to achieve the target performance, while the inverse design starts with the desired performance and directly derives the blade geometry. However, these existing methods face challenges such as a heavy reliance on expert knowledge, high computational complexity, and limited robustness. This article proposes a generative inverse aerodynamic design framework based on a conditional denoising diffusion probabilistic model to mitigate the drawbacks of traditional methods. The framework takes turbomachinery aerodynamic performance and other key indicators as input and directly generates the corresponding design parameters. It consists of two main components: forward process modeling and inverse process modeling. In the forward process, a transformer performance surrogate model is utilized to construct a large-scale, low-cost database based on limited simulation data and to evaluate the performance of different geometries. The inverse process employs a prior knowledge-guided conditional diffusion model to achieve generative inverse design of turbomachinery according to the specified performance targets. The proposed framework is applied to the design of a single-stage turbine. Results show that a single RTX 4090 GPU can produce multiple feasible geometries within 2 s. Further computational fluid dynamics (CFD) validations confirm that the generated designs achieve a mean absolute relative error of less than 2% compared to target performance. This significantly improves design efficiency and reduces dependence on expert knowledge of designers, offering a promising direction for intelligent aerodynamic design of turbomachinery.
Zitieren
APA
Wu, Y., Zhang, M., Xu, X., Yin, Y., Ba, D., & Du, J. (2026). Generative Inverse Aerodynamic Design of a Single-Stage Turbine Using Conditional Denoising Diffusion Probabilistic Model. Journal of Turbomachinery, 148(4). https://doi.org/10.1115/1.4069801
Vancouver
Wu Y, Zhang M, Xu X, Yin Y, Ba D, Du J. Generative Inverse Aerodynamic Design of a Single-Stage Turbine Using Conditional Denoising Diffusion Probabilistic Model. Journal of Turbomachinery. 2026;148(4). doi:10.1115/1.4069801
BibTeX
@article{Wu2026,
author = {Wu, Yueteng and Zhang, Min and Xu, Xiaobin and Yin, Yue and Ba, Dun and Du, Juan},
title = {Generative Inverse Aerodynamic Design of a Single-Stage Turbine Using Conditional Denoising Diffusion Probabilistic Model},
journal = {Journal of Turbomachinery},
year = {2026},
volume = {148},
number = {4},
doi = {10.1115/1.4069801}
}
EUVIMED