Modeling of Complex Large-Scale Flow Phenomena
Public Deposited- Last Modified
- March 19, 2019
- Creator
-
Golas, Abhinav
- Affiliation: College of Arts and Sciences, Department of Computer Science
- Abstract
- Flows at large scales are capable of unmatched complexity. At large spatial scales, they can exhibit phenomena like waves, tornadoes, and a screaming concert audience; at high densities, they can create shockwaves, and can cause stampedes. Though strides have been made in simulating flows like fluids and crowds, extending these algorithms with scale poses challenges in ensuring accuracy while maintaining computational efficiency. In this dissertation, I present novel techniques to simulate large-scale flows using coupled Eulerian-Lagrangian models that employ a combination of discretized grids and dynamic particle-based representations. I demonstrate how such models can efficiently simulate flows at large-scales, while maintaining fine-scale features. In fluid simulation, a long-standing problem has been the simulation of large-scale scenes without compromising fine-scale features. Though approximate multi-scale models exist, accurate simulation of large-scale fluid flow has remained constrained by memory and computational limits of current generation PCs. I propose a hybrid domain-decomposition model that, by coupling Lagrangian vortex-based methods with Eulerian velocity-based methods, reduces memory requirements and improves performance on parallel architectures. The resulting technique can efficiently simulate scenes significantly larger than those possible with either model alone. The motion of crowds is another class of flows that exhibits novel complexities with increasing scale. Navigation of crowds in virtual worlds is traditionally guided by a static global planner, combined with dynamic local collision avoidance. However, such models cannot capture long-range crowd interactions commonly observed in pedestrians. This discrepancy can cause sharp changes in agent trajectories, and sub-optimal navigation. I present a technique to add long-range vision to virtual crowds by performing collision avoidance at multiple spatial and temporal scales for both Eulerian and Lagrangian crowd navigation models, and a novel technique to blend both approaches in order to obtain collision-free velocities efficiently. The resulting simulated crowds show better correspondence with real-world pedestrians in both qualitative and quantitative metrics, while adding a minimal computational overhead. Another aspect of real-world crowds missing from virtual agents is their behavior at high densities. Crowds at such scales can often exhibit chaotic behavior commonly known as emph{crowd turbulence}; this phenomenon has the potential to cause mishaps leading to loss of life. I propose modeling inter-personal stress in dense crowds using an Eulerian model, coupled with a physically-based Lagrangian agent-based model to simulate crowd turbulence. I demonstrate how such a hybrid model can create virtual crowds whose trajectories show visual and quantifiable similarities to turbulent crowds in the real world. The techniques proposed in this thesis demonstrate that hybrid Eulerian-Lagrangian modeling presents a versatile approach for modeling large-scale flows, such as fluids and crowds, efficiently on current generation PCs.
- Date of publication
- May 2015
- Keyword
- Subject
- DOI
- Identifier
- Resource type
- Rights statement
- In Copyright
- Advisor
- Manocha, Dinesh
- Lin, Ming
- Sewall, Jason
- Lastra, Anselmo
- Adalsteinsson, David
- Degree
- Doctor of Philosophy
- Degree granting institution
- University of North Carolina at Chapel Hill Graduate School
- Graduation year
- 2015
- Language
- Publisher
- Place of publication
- Chapel Hill, NC
- Access right
- There are no restrictions to this item.
- Date uploaded
- June 23, 2015
Relations
- Parents:
This work has no parents.
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Golas_unc_0153D_15324.pdf | 2019-04-12 | Public | Download |