Engineering design involves manipulating product geometry to meet user demands, satisfy manufacturing constraints and maximize physical performance. In practice, integrating multiple components into cohesive systems presents challenges in achieving optimal performance while maintaining structural integrity. In the long run, we aim to develop a new framework that leverages deep geometric learning to automate the parameterization and co-optimization of multi-part systems while ensuring adherence to geometric constraints. This will increase design efficiency and effectiveness in industries such as aerospace and automotive engineering through fully AI-assisted Computer-Aided Design and Engineering (CAD/CAE).
Objective
In this project, we will extend an existing single-part modeling technique [1] to handle complex assemblies. The primary challenges include:
- Joint Design Space Exploration: Developing methods to explore a unified design space that encapsulates all the components and move beyond isolated per-part optimizations. Implementing optimization strategies that consider the interdependencies among multiple components to achieve holistic system performance.
- Geometric Constraints Enforcement: Ensuring that the optimized components fit together seamlessly, and adhere to predefined geometric constraints to maintain structural integrity and manufactural feasibility.
Key steps to achieving this goal include designing a unified parameterization scheme, implementing an interactive co-optimization algorithm, and incorporating hard constraints into deep learning models.
Prerequisites
- Proficiency with PyTorch and Python.
- Solid background in mathematics or physics.
- Familiarity with computer assisted engineering would be a plus.
References
[1] Wei, Z., Yang, A., Li, J., Bauerheim, M., Liem, R. P., & Fua, P. (2024). DeepGeo: Deep geometric mapping for automated and effective parameterization in aerodynamic shape optimization. In AIAA AVIATION FORUM AND ASCEND 2024.