high performance computing on graphics processing units: hgpu.org

The efficient planning of automated machining processes is unthinkable without the use of offline CAM systems. Though machining programs can be written and input manually, right at the machine controller, if the workpiece geometry is complex, or if the machined features are numerous, the help of CAM software is essential for generating the program both accurately and quickly. This is particularly true in the case of multi-axis machining, when complex, often sculpture-like shapes are machined in a single set-up. The determination of the optimal toolpath in accordance with the CAD part designs requires sophisticated computerized algorithms. On the other hand, it is indispensable to allow the user to verify and – if it’s needed – manually modify the computer-generated operations in the course of the interactive planning process. The most evident mode of the verification is the ocular inspection of the planned machining operations on the screen of the CAM software. The visual representation of the machining processes has several levels, from the simple linear drawing of the toolpath until the detailed displaying of the altering workpiece geometry during the material removal process. While the first helps shifting out the rough errors of the automated toolpath generation, the latter can bring the unwanted geometrical errors of the machined part to light. The first part of the dissertation deals with this higher level of machining process representation. The simulation of the material removal process presents a special challenge for the CAM software developers. To get the final shape of the workpiece the simulation of the whole material removal process has to be fulfilled. Considering that the resolution of the displayed objects must satisfy the requirements of the visual verification, a huge amount of data, which carries enough geometrical information for the detailed description, has to be processed within a short time. In the course of time several simulation methods have come to life to perform this task, with various levels of performance. One of the oldest, and still the most effective solution up to now exploited that the operations of the material removal can be performed in a highly parallelized thus effective manner by the use of graphics processing units (GPU). Unfortunately this method proved to be viable only in the simple geometrical circumstances of 3-axis machining. The complex geometry of a workpiece that is shaped during multi-axis machining cannot be represented with data structures used by the traditional GPUs, and the sequential execution by the CPU cannot even approach the performance of the GPU based solution.