Assignment 0: Introduction Name: Sushrut Pavanaskar (20560943) Department and Major: Mechanical Engineering (Manufacturing) Research Area: CAD/CAM, Advanced algorithms for CAD using GPUs (GPGPU) Applications: Many of advanced 3D modeling software use the power of a graphics processor (GPU) present on a video card. Solidworks is one such software where GPU based algorithms are written for more than one function. To name a few such algorithms performs complex curve calculations, collision detection and perform advanced rendering. To make GPUs work for such applications, one needs to program the graphics processor at a sufficiently low level. nVidia provides its own language CUDA for this purpose and there are some other languages like CG too. All such programs have to be parallel since a G PU is inherently parallel in architecture. Thus my aim behind taking CS267 is to learn parallel programming fundamentals, implementation details for effectively programming the GPUs. I am also registered for ME290R which is GPGPU programming. Most of the GPU programming done today, for CAD /CAM is implemented on normal graphics cards and so I am not really acquainted with supercomputers. These applications perform significantly well though professional benchmarking or quantitative results are available only for specific applications (e.g. NURBS interpolator performs 5 times faster as compared to on a CPU etc.) Most of them use CG/CUDA. Details of a particular Application: MACHINABILITY ANALYSIS OF FREE FORM SURFACES ON THE GPU Aim: This application is developed to aid in machinability analysis of freeform surfaces. Details of the process: Freeform surfaces are generally found in automotive design and need to be machined using ball milling or similar 5 axis milling processes. Such processes are difficult to control as the surface changes its curvature at every point. Design of freeform surfaces is a very important step as it takes maximum time in a product’s life cycle. Costly design mistakes from lack of knowledge often result in increased cost of manufacture or in the worst case, non‐manufacturable parts. Design for Manufacturability (DFM) thus a commonly employed methodology developed to simultaneously consider design goals and manufacturing constraints/limitations in order to eliminate or minimize manufacturing problems during the design process. Other than hard constraining the designer to use standard parts, some efforts to implement DFM employ feature recognition techniques to evaluate manufacturability of the parts, which imposes no constraints on the designers’ freedom. Most of these methods pertain to prismatic parts and use continuous space solid modelers such as ACIS and Parasolid for geometric analysis. These methods are slow due to the complexity of geometric computations and highly susceptible to numeric round‐off errors. Moreover, the inbuilt algorithms in these solid modelers run serially on a CPU and are not real‐time especially in analyzing complex surfaces. Also such algorithms, while planning a complex tool‐path employ methods like Z‐buffer for a sampled set of points on the surface. Although these applications use the graphics pipeline and the associated hardware in generating tool‐paths, they did not use programmability of the graphics hardwa re, and therefore most of the computations are performed on the CPU. Therefore the speed of such applications is limited. This application uses programmable GPUs to perform the same calculations but in parallel fashion. Novelty of approach lies in understanding the unique parallel computing model of the GPUs and formulating the machinabi lity analysis problem in terms of a specialized 3‐D shading problem. The graphics pipeline is a good match for the stream computing model, on which most of the GPUs are based. It is structured in stages of computation connected by data flow between stages. Data flow is highly localized with data produced by one stage immediately consumed by the subsequent stage. On the GPU, data streams are represented as textures and kernels are represented by shaders. Machinability problem: A surface is deemed machinable if every point on the surface is ‘accessible’ to the ‘smallest’ available cutting tool. This means that the cutting tool does not intersect any part of the surface when placed at each and every point on the surface. In this method, analysis is conducted in the image space model r esulting from the projected image of any surface. First, the entire surface is rendered into a position and normal map, and then a shader (GPU program) is used to compute accessibility from the data. Figure shows results run on a sample model and colored part represents NON‐MACHINABL E surface due to inaccessibility. On the left is the region inaccessible to a smaller tool while on the right colored region represents inaccessible region of a bigger diameter tool. Naturally more area is accessible to a smaller tool. Implementation Details: The program is written using CG / Microsoft Visual Studio using DirectX and the GPU used is NVIDIA GeForce 7950. (Source: “Real time machinability analysis of free form surfaces on the GPU.” Mikola Lysenko, Keyvan Rahmani and Roshan D’souza, Proceedings of the ASME 2007 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference, IDETC/CIE 2007, September 4‐7, 2007, Las Vegas, Nevada, USA)
View Full Document
Unlocking...