Speaker
Description
In shape optimization, especially for designs like airfoils and car bodies, computational fluid dynamics (CFD) simulations are essential for assessing aerodynamic performance. To speed up the optimization process, surrogate models—such as machine learning techniques can be used. These models approximate CFD outcomes at a fraction of the cost, enabling faster design exploration and optimization while maintaining sufficient accuracy. This makes it possible to perform efficient design iterations, making the process more feasible for real-world engineering.
In our work, we develop a differentiable model that uses a parameterized mesh as input and outputs key performance metrics like drag (CD) and lift (CL) coefficients. This approach allows for automatic shape optimization using gradient-based methods.
We also present preliminary results comparing different methods, including geodesic convolutional networks and traditional convolutional techniques, where shapes are represented as images for optimization.
