On loosely-coupled parallel implementation of algorithms for computer-aided manufacturing

Many manufacturing processes, such as numerically controlled (NC) machining and robotic spraying, are conducted under direct numerical control of computers. The process of creating such numerical control (NC) programs remains quite complex and error-prone for many sophisticated tasks. The final result of tool motions under control of such programs in machining complex-shaped parts is often uncertain. Proofing runs on actual NC machines are time consuming and expensive. The high cost of conducting test runs of such programs has created a strong demand for simulators and verifiers for such programs. Current NC simulation software on sequential machines greatly improves the verification procedure. However, such NC simulators are still inaccurate and most can handle only 3-axis milling with the simplest types of cutters. Given the former dearth of sophisticated software for generating optimal 5-axis toolpaths for more complex tools, this inadequacy of simulators or verifiers was not very noticeable. However, the current trend toward higher-capability NC programming systems raises the strong need for more sophisticated and accurate verification. This trend highlights the tradeoff between verification time and accuracy of verification, strongly raising the desire for parallelization of NC verification or simulation processes.; This research presents an algorithm which can address both concerns--high efficiency and accuracy, using parallel (or distributed) processing. It not only guarantees that tessellated surfaces are within a user-specified tolerance, but also pre-estimates work load directly from the original geometry parameters, such as surface order, control points and knots. The proposed algorithm pre-evaluates the sculptured surfaces to be machined for parallel processing, estimates the load balance for the processors, discretizes the nominal sculptured surfaces based on the surface curvature and user-defined error tolerance, distributes the computational job onto the given number of processors or workstations, and uses a parallel processing approach to directly compute the possible interference between the surfaces being machined and the envelope of the moving tool, without solving the swept volume problem as a solid modeling operation. The geometric model uses the ruled surface defined by the axis of the cutting tool to define the center of the tool envelope. The surface pre-evaluation guarantees that near-optimal computational performance will be achieved on a given number of processors. We believe the proposed algorithm could be easily implemented on a distributed system.; Some results from a sequential 5-axis NC toolpath verification system implemented by the authors are presented first, and this is used as the basis for the parallelization work described here. Some published parallel approaches for NC machining are reviewed, then the new scheme for parallelization is presented. Finally, the performance of this scheme on parallel vs. single CPU machines is reported and future work is discussed.