基于并行连续碰撞检测的虚拟装配技术

发布时间：2018-06-03 03:04

本文选题：多核GPU + 并行运算　；参考：《杭州电子科技大学》2015年硕士论文

【摘要】：作为衡量一个虚拟装配系统优越与否的重要标准，，碰撞检测问题受到人们的广泛关注。当前，实时性和精确性是判断碰撞检测算法的关键标准，因此如何利用日渐成熟的计算机硬件的并行架构，来实现复杂场景下的精确的实时碰撞检测成为人们当前的研究重点。在复杂场景中，连续碰撞检测（Continuous Collision Detection，CCD）依然很难做到实时效果。随着CPU并行架构的实现，本文提出了一种CPU多核加速的碰撞检测算法来提高虚拟装配系统性能。通过利用硬件的多核性能，提出了一个空间分解理论，然后利用动态任务分配策略把碰撞检测任务均衡的分配到所有内核上，使每个内核上的的任务负载均衡。最后通过数据分析，该算法可以有效提高碰撞检测效率。相较于具有较强逻辑计算能力CPU来说，具有强大浮点计算能力的GPU在有庞大运算量的装配系统中更有优势。由于人们在进行数学建模时不可避免出现误差，从而导致本不应该有碰撞接触的零部件出现重叠或者嵌套现象。针对以上问题，我们利用现代GPU强大的高密度计算能力及其多核的性能，提出了一种实时的GPU加速的并行碰撞检测算法。主要工作内容如下：（1）提出了一个新的基于GPU加速的并行碰撞处理框架，利用连续碰撞检测保证物体在移动过程中没有碰撞遗漏现象。（2）为了解决CPU与GPU之间由于带宽限制所造成的数据吞吐量较小而引起的通信瓶颈问题，改进了基于碰撞流优化的CCD方法，以改善碰撞检测的整体性能，实现系统的实时效果。（3）将离散碰撞检测（Discrete Collision Detection，DCD）与连续碰撞检测结合。把三角行碰撞检测（Triangle Intersection Detection，TID）结合到当前的基于CCD的多核碰撞检测算法中，利用混合的碰撞检测来避免由于建模误差的缘故导致在初始化时彼此间有嵌套、重叠现象而无法移动部件的问题。
[Abstract]:As an important criterion to measure the superiority of a virtual assembly system, collision detection has been paid more and more attention. At present, real-time and accuracy are the key criteria for judging collision detection algorithms. Therefore, how to use the increasingly mature parallel architecture of computer hardware to achieve accurate real-time collision detection in complex scenes has become the focus of current research. In complex scenarios, continuous Collision detection is still difficult to achieve real-time effects. With the implementation of CPU parallel architecture, a CPU multi-core accelerated collision detection algorithm is proposed to improve the performance of virtual assembly system. By using the multi-core performance of hardware, a spatial decomposition theory is proposed, and then the collision detection task is evenly distributed to all kernels using dynamic task allocation strategy, which makes the task load balance on each kernel. Finally, through data analysis, the algorithm can effectively improve the efficiency of collision detection. Compared with CPU with strong logical computing power, GPU with strong floating-point computing capability has more advantages in assembly systems with large computational complexity. Due to the inevitable errors in mathematical modeling, there is overlap or nesting of parts which should not have collision contact. To solve the above problems, we propose a real-time parallel collision detection algorithm based on GPU acceleration using the powerful high density computing power of modern GPU and its multi-core performance. The main tasks are as follows: A new parallel collision processing framework based on GPU acceleration is proposed. Continuous collision detection is used to ensure that there is no collision omission in the moving process. In order to solve the communication bottleneck problem between CPU and GPU caused by the low data throughput caused by bandwidth limitation, the CCD method based on collision flow optimization is improved to improve the overall performance of collision detection and realize the real-time effect of the system. The discrete Collision Detection (DCD) is combined with continuous collision detection. In this paper, Triangle Intersection Detection (TID) is combined with the current multi-core collision detection algorithm based on CCD, and the hybrid collision detection is used to avoid nesting each other during initialization due to modeling errors. The problem of overlapping and immovable parts.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TG95;TP391.9

【参考文献】