A Fast Method to Detect Collision for Five-axis Machining with GPU

Abstract

The applications of five-axis machining have been used widely recently. Owing to the expensive cost of machine tools, how to detect the collision in real time has become a critical issue. Indeed, in order to ensure that G-codes will not result in the collision, the developers may use some tools before the stage to process the five-axis machine tool on off line. Moreover, to reduce long execution time on off line, we propose a parallel method to remedy it in this paper. The objective of the proposed approach aims at improving the performance to detect collision in parallel by utilizing the functions of a GPU (Graphics Processing Unit).We address the issue above by inducing six separating axis in plan and 11 separating axis in non-plan for two triangle meshes. Then we propose a parallel approach by implementing a CUDA ( Compute Unified Device Architecture ) program based on a GPU. Finally, with our domain knowledge and experiences, we attempt to optimize the proposed work with loop unrolling and prefetching techniques to improve performance.. The result shows that our work is very efficiently by using the two techniques.