PI: Levent Kara
Co-PI(s): Jonathan Cagan
University: Carnegie Mellon University
Industry partner: Ansys Inc.
Generative design and topology optimization (TO) algorithms treat the structural performance as minimizing a combination of compliance and volume fraction, generating lightweight structures. However, the optimized shape of TO is rarely directly manufacturable. While previous research has attempted to model manufacturability constraints, it has not included various objectives and constraints imposed by different manufacturing technologies. The long-term vision of this project is to develop the means for an automatic, purely requirements-driven part design that jointly considers engineering (performance), and manufacturing concerns. The objective of this particular project is to improve the generation of 3D mechanical parts that are derived by TO and are manufactured by a prescribed manufacturing process. Because of the instant inference of deep learning models, we expect to obtain ready-to-manufacture structure significantly faster compared to manual modification. As this work explores the potential of encoding multiple machining constraints in one neural network, it’s also expected to save the effort of designing analytical formulations for every single machining method.