S2E2
Episode 2 (October 11, 2020)
Charles Yang
UC Berkeley
Jin Yang
University of Wisconsin - Madison
Yifan Wang
California Institute of Technology
Applying Deep Learning to Composite Design
Abstract:
The rise of additive manufacturing has opened up an enormous, previously unexplored design space. Finding new designs efficiently with desireable mechanical properties is a difficult computational and experimental problem. I will present my work on using deep learning to serve as an emulator for finite element simulations in order to more efficiently explore the design space for binary composites. I will also explore some interpretability methods and explain how we incorporated deep learning into our research problem.
Adaptive Augmented Lagrangian Digital Image/Volume Correlation
Abstract:
Digital image/volume correlation (DIC/DVC) is a powerful, non-invasive experimental method for extracting 2D and 3D volumetric full-field deformation information. The basic idea of this method is to compare images of an object painted with a speckle pattern before and after deformation, and thereby to compute displacements and strains. Most current DIC/DVC algorithms can be categorized into either local or finite element based global methods. However, there are some drawbacks with either of these methods. In the local method, since all of the local subset deformations are estimated independently, the computed displacement field may not be compatible, and the deformation gradients can be noisy, especially for small subsets. Although the global method can incorporate kinematic compatibility, it is generally much more computational expensive than its local counterpart. Here we present a new hybrid algorithm, the augmented Lagrangian digital image/volume correlation (AL-DIC/DVC), that combines the advantages of both the local (fast computation times) and global (compatible displacement field) methods. I will show that the AL-DIC/DVC has higher accuracy and behaves more robustly compared to both current local and global DIC/DVC methods. Finally, I will demonstrate that this new AL-DIC/DVC technique can be implemented with adaptive meshing capability, which can further save computation time one order of magnitude.
Adaptive and Active Lattices for Dynamic Applications
Abstract:
Architected lattices are materials that derive their properties from the selection of both their constitutive materials and the geometry of their micro-and meso-structure. Most existing architected lattices are intrinsically passive, with properties fixed once fabricated. This limits their applications in areas where material adaptivity and tunability are required. In this talk, I will present the development of architected lattices whose mechanical properties can be actively controlled and can adapt to different dynamic conditions. I will first demonstrate an architected lattice filled with granular particles whose damping properties can be controlled to achieve optimal energy absorption, over a range of impact energies. Then, I will discuss actively modulated phononic lattices that allow non-reciprocal wave propagation and unidirectional vibration mitigation. These works open routes towards creating the next generation of structured materials which adapt to varying environmental conditions.