Date of Award
Doctor of Philosophy (PhD)
Dr. Laine Mears, Committee Chair
Dr. Thomas R. Kurfess
Dr. Tommy Tucker
Dr. Stan Birchfield
This work has created a completely new geometry representation for the CAD/CAM area that was initially designed for highly parallel scalable environment. A methodology was also created for designing highly parallel and scalable algorithms that can use the developed geometry representation. The approach used in this work is to move parallel algorithm design complexity from an algorithm level to a data representation level. As a result the developed methodology allows an easy algorithm design without worrying too much about the underlying hardware. However, the developed algorithms are still highly parallel because the underlying geometry model is highly parallel. For validation purposes, the developed methodology and geometry representation were used for designing CNC machine simulation and tool path planning algorithms. Then these algorithms were implemented and tested on a multi-GPU system. Performance evaluation of developed algorithms has shown great parallelizability and scalability; and that main algorithm properties are required for modern highly parallel environment. It was also proved that GPUs are capable of performing work an order of magnitude faster than traditional central processors. The last part of the work demonstrates how high performance that comes with highly parallel hardware can be used for development of a next level of automated CNC tool path planning systems. As a proof of concept, a fully automated tool path planning system capable of generating valid G-code programs for 5-axis CNC milling machines was developed. For validation purposes, the developed system was used for generating tool paths for some parts and results were used for machining simulation and experimental machining. Experimental results have proved from one side that the developed system works. And from another side, that highly parallel hardware brings computational resources for algorithms that were not even considered before due to computational requirements, but can provide the next level of automation for modern manufacturing systems.
Konobrytskyi, Dmytro, "Automated CNC Tool Path Planning and Machining Simulation on Highly Parallel Computing Architectures" (2013). All Dissertations. 1779.