Enhancing CAD-integrated automatic feature recognition of weld joints with GPU-accelerated multi-directional slicing

Abstract

In this paper, a GPU implementation for a previously proposed weld feature recognition in CAD is presented. The intention was to support downstream manufacturing process planning for large welded structures and automate feature recognition processes, essentially saving what could accumulate to thousands of man-hours spent. However, when running on larger and more complex structures, computation can still reach hours. The GPU implementation presented in this paper aims to further accelerate this process to make it more user-friendly while simultaneously providing an outline for the underlying implementation processes into CAD environments. To do this, Siemens NX is used as the CAD framework with the NXOpen API allowing for functions and tools to be accessed in a programmatic manner. The GPU utilization can then be imported through the Python Numba library along with its CUDA programming architecture. The proposed implementation is shown to perform more than a thousand times faster than its CPU serial execution counterpart for the most complex models. While for the most simple model, the CPU algorithm slightly outperforms the GPU implementation due to the just-in-time compilation nature of Numba.