GPU-accelerated geometric algorithms for computational modeling and simulation

Thumbnail Image
Date
2021-12
Authors
Shah, Harshil
Major Professor
Advisor
Krishnamurthy, Adarsh
Ganapathysubramanian, Baskar
Hsu, Ming-Chen
Oliver, James
Sarkar, Soumik
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Altmetrics
Authors
Research Projects
Organizational Units
Journal Issue
Is Version Of
Versions
Series
Department
Mechanical Engineering
Abstract
The boundaries of engineering design are always pushed with a better understanding of problems, novel methods to tackle them, and the capabilities to solve complex problems in an efficient yet robust way. The advancement in modeling of the vision of an engineer in a virtual setup and perform testing with all the in-service scenarios have played a massive part in speeding up the process of getting the design from board to user. Nevertheless, we find new ways to create a more realistic model and test scenarios. As we move closer to replicating the actual model and analysis, we run into new types of challenges. In this research work, we have addressed some of the challenges faced in 3 domains: modeling, analysis, and visualization. The trifecta of capability is always the benchmark for comparison between any commercial tools. The first stage of any virtual setup is to generate a model as close to reality as possible. This modeling step is critical, especially for thin wall structures and multi-laminate composites. We have developed a method to generate volumetric NURBS using surface offset to generate closer-to-reality thin-walled and multi-layer structures. We have added the capability to generate different thickness volumes in multi-layer and variable thickness along the span of the surface element. We have validated these models with analytical models for the same geometry. With improvement in environment mapping, navigation in space without any collision has become a challenge regarding computation speed and analysis consistency. We present here a GPU accelerated collision analysis method for a pre-computed path in a point cloud environment. We use model voxelization to perform point-voxel classification to identify voxels part of the model that may be colliding with the environment. We also add the capability to perform clearance analysis by adding voxel layers using the Minkowski sum operation. We have developed two types of classification that can be used based on the complexity of the model. The output is visualized in Unreal Engine that can handle a large point cloud environment. Finally, it is difficult to visualize any isogeometric analysis (IGA) results due to difficulty rendering NURBS surfaces. We have developed a GPU accelerated volume visualization method using a modified ray intersection test. Our direct voxelization approach eliminated the need to store intersection data with surface elements. The variable density model allows us to visualize any analysis results with normalization. We tested our method on two different types of models and analysis to showcase its versatility. We visualized the results in an interactive animation that allows us to understand the results in-depth. Our tests show that we can keep the frame rate of the animation over 30fps for all the tested voxelization resolutions.
Comments
Description
Keywords
Citation
Source
Copyright