Airfoil design and surrogate modeling for performance prediction based on deep learning method

Du, Qiuwan; Xie, Yonghui; Yang, Like; Li, Liangliang; Zhang, Di; Xie, Yonghui

doi:10.1063/5.0075784

Cited by 50 publications

(13 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Different from the previous work which only used the SDF matrix (Bhatnagar et al 2019) or the latent codes (Du et al 2022) to construct the input for predicting flow fields, we utilize the SDF matrix coupled with latent codes generated by Bézier GAN to construct the sparse input and enhance the airfoil geometry information. To explore the effect of information enhancement on the prediction performance of the SCNN model, we trained another SCNN model but only 10(a) shows the convergence history during the training of the two models.…”

Section: Influence Of the Input Parameters On The Scnnmentioning

confidence: 99%

Computationally effective estimation of supersonic flow field around airfoils using sparse convolutional neural network

Peng

Zhi-ming³

et al. 2023

Fluid Dyn. Res.

View full text Add to dashboard Cite

This work proposes an innovative approach for supersonic flow field modeling around airfoils based on sparse convolutional neural networks (SCNN) and Bézier generative adversarial network (GAN), where 1) the SCNN model is built to end-to-end predict supersonic compressible physical flow fields around airfoils from spatially-sparse geometries and 2) the trained Bézier-GAN is utilized to generate plenty of smooth airfoils as well as the latent codes representing airfoils. The spatially-sparse positions of airfoil geometry are represented using Signed Distance Function (SDF). Particularly, the latent codes are merged with the SDF matrix and the Mach number to form the input of the SCNN model, effectively making the SCNN model possess more robust geometric adaptability to different flow conditions. The most significant contribution compared to the regular convolutional neural network is that SCNN introduces sparse convolutional operations to process spatially-sparse input matrix, specifically, which only focuses on the local area with flow information when performing convolution, eventually saving memory usage and improving the network’s attention on the flow area. Further, the testing results show that the SCNN model can more accurately predict supersonic flow fields with a mean absolute error lower than 5% and save 40% of GPU (Graphics Processing Unit) memory. These results indicate that the proposed SCNN model can capture the shock wave features of supersonic flow fields and improve learning efficiency and computing efficiency.

show abstract

Section: Influence Of the Input Parameters On The Scnnmentioning

confidence: 99%

Computationally effective estimation of supersonic flow field around airfoils using sparse convolutional neural network

Peng

Zhi-ming³

et al. 2023

Fluid Dyn. Res.

View full text Add to dashboard Cite

show abstract

“…It is not only with regard to the flowfield that pressure is important, but also its distribution on aerodynamic surfaces. For instance, pressure distribution data is crucial for aerodynamic design optimization [19] and many surrogate models have recently been developed to reduce the computational time involved in predicting aerodynamic forces during the design process [20,21,22,23]. In the context of unsteady aerodynamics, pressure distribution data are used to provide insights on the turbulent structures responsible for the far-field noise generated by unsteady airfoils [4,24,5] and also to estimate the instant at which the flow separates near the leading edge [25,26,27].…”

Section: Introductionmentioning

confidence: 99%

Flow imaging as an alternative to pressure transducers through vision transformers and convolutional neural networks

Miotto¹,

Wolf²

2023

Preprint

View full text Add to dashboard Cite

In this work, we propose a framework whereby flow imaging data is leveraged to extract relevant information from flowfield visualizations. To this end, a vision transformer (ViT) model is developed to predict the unsteady pressure distribution over an airfoil under dynamic stall from images of the flowfield. The network is capable of identifying relevant flow features present in the images and associate them to the airfoil response. Results demonstrate that the model is effective in interpolating and extrapolating between flow regimes and for different airfoil motions, meaning that ViT-based models may offer a promising alternative for sensors in experimental campaigns and for building robust surrogate models of complex unsteady flows. In addition, we uniquely treat the image semantic segmentation as an image-to-image translation task that infers semantic labels of structures from the input images in a supervised way. Given an input image of the velocity field, the resulting convolutional neural network (CNN) generates synthetic images of any corresponding fluid property of interest. In particular, we convert the velocity field data into pressure in order to subsequently estimate the pressure distribution over the airfoil in a robust manner.

show abstract

“…The findings demonstrate that machine learning is more computationally efficient than conventional genetic algorithms for optimizing lift-todrag characteristics. Du et al [23] established a deep-learning-based convolutional neural network framework (DPCNN) for airfoil design and performance optimization. They optimized aerodynamic performance parameters using the gradient descent method, achieving airfoil database parameterization and performance prediction with superior robustness and convergence.…”

Section: Introductionmentioning

confidence: 99%

Machine Learning Assisted Prediction of Airfoil Lift-to-Drag Characteristics for Mars Helicopter

et al. 2023

View full text Add to dashboard Cite

The aerodynamic properties of rotor systems operating within low Reynolds number flow field conditions are profoundly influenced by their geometric and flight parameters. Precise estimation of optimal airfoil parameters at different angles of attack is indispensable for enhancing these aerodynamic properties. This study presents a technique for optimizing the airfoil parameters of a Mars helicopter by employing machine learning methods in conjunction with computational fluid dynamics (CFD) simulations, thereby circumventing the need for expensive experiments and simulations. The effectiveness of diverse machine learning algorithms for prediction is evaluated, and the resultant models are utilized for airfoil optimization. Ultimately, the aerodynamic properties of the optimized airfoil are experimentally validated. The experimental findings exhibit agreement with the simulated predictions, indicating the successful optimization of the aerodynamic properties. This research offers valuable insights into the influence of airfoil parameters on the aerodynamic properties of the Mars helicopter, along with guidance for airfoil optimization.

show abstract

Airfoil design and surrogate modeling for performance prediction based on deep learning method

Cited by 50 publications

References 42 publications

Computationally effective estimation of supersonic flow field around airfoils using sparse convolutional neural network

Computationally effective estimation of supersonic flow field around airfoils using sparse convolutional neural network

Flow imaging as an alternative to pressure transducers through vision transformers and convolutional neural networks

Machine Learning Assisted Prediction of Airfoil Lift-to-Drag Characteristics for Mars Helicopter

Contact Info

Product

Resources

About