Embedded Files
Generative AI Models for Inverse Design
I developed a generative AI framework using a variational autoencoder (VAE) to generate pressure distributions around airfoils. The VAE model was used as a surrogate to overcome the limitations of traditional inverse design optimization, which often requires repeated iterations for different target distributions. By training the VAE on existing aerodynamic performance data, I was able to generate realistic target pressure distributions. This generative approach allows for more flexible and efficient exploration of potential airfoil designs, providing a faster and more versatile way to optimize aerodynamic performance without the need for complex iterative processes.
In addition to generating pressure distributions, the VAE framework enabled the direct prediction of quantities of interest (QoIs) and airfoil shapes from the generated distributions. This process bypasses the need for predefined performance distributions and multiple constraints often associated with traditional methods. The use of a VAE in this context demonstrated the potential of generative AI to automate the inverse design process while maintaining accuracy, efficiency, and flexibility in generating airfoil shapes that meet specific performance requirements.
Click for more details
Generative AI Models for Flow Fields Reconstruction
The next work extends the use of VAEs for the generation of transonic flow fields, highlighting their potential for reduced-order modeling (ROM). In this work, the VAE was trained to encode high-dimensional flow field data into a lower-dimensional latent space, capturing the essential features of transonic flows around airfoils. By leveraging this latent space, I could efficiently generate new flow field configurations, providing a valuable tool for aerodynamic analysis. This generative AI approach enabled quick predictions of transonic flow behavior, reducing the computational cost typically associated with high-fidelity simulations.
The usefulness of the VAE in this application lies in its ability to identify key physical parameters, such as the Mach number and angle of attack, which were automatically encoded in the latent variables. By exploring the latent space, the VAE could generate flow fields under different conditions, facilitating a more comprehensive understanding of transonic flow dynamics. This demonstrated how generative AI can play a crucial role in both data augmentation and the rapid exploration of complex flow phenomena, aiding in the overall design process.
Page updated
Google Sites
Report abuse