滑动轴承混合润滑多线程并行计算数值方法

Mixed lubrication study of multi-thread parallel computing algorithm for journal bearings

针对混合润滑数值分析将动压效应、弹性变形和界面接触特性耦合而非常耗时的问题,基于共享内存并行系统的多线程程序设计语言OpenMP,提出一种多线程混合润滑并行计算数值方法——红黑线交叉并行计算法.该并行计算模型是将雷诺方程求解域分成两个相互独立的子求解域,依次对两个子求解域进行并行数值求解,可以有效克服CPU线程间数据争用问题,加快求解速度.着重研究了并行计算核数、网格数量和工作站配置对并行计算性能的影响,分析结果表明:并行计算模型能够有效提高滑动轴承混合润滑计算速度,并行计算速度的提升幅度与并行计算核数成非线性关系,随着CPU核数的增加计算速度的增加幅度逐渐减小;此外,与内存和缓存相比,CPU的主频对并行计算速度有非常大的影响.

The complex and time-consuming mixed-（ elastic hydrodynamic lubrication） problems are introduced by the contact between the journal and bearing require high-speed computation. So a faster mixed-EHL parallel computing algorithm based on the OpenMP was proposed, which exhibits higher efficiency than the normal SOR（successive over relaxation） method. In this method, the Reynolds computing domain was separated into two indecently sub-domains, which can avoid the data race is- sues among the CPU（central processing unit） computing threads. A group of simulations were pres- ented to evaluate the effects of core numbers, meshes size and workstation configurations on the paral- lel computing performance. The results show that parallel computing method can significantly improve the mixed-EHL computing speed, however the increment of speed has a non-linear relationship with cores numbers. Besides, the clock rate has significantly effects on the computing parallel performance comparing to CPU cache and RAM（random-access memory）.