基于BP算法的二维阵面推进网格划分方法的优化与应用  

Optimization and application of two -dimensional advancing grid generationmethod based on BP algorithm

作　　者：刘翰林 崔会敏 张珍 韩智铭 刘庆宽[1,2,4] LIU Han-lin;CUI Hui-min;ZHANG Zhen;HAN Zhi-ming;LIU Qing-kuan(State Key Laboratory of Mechanical Behavior and System Safety of Traffic Engineering Structures,Shijiazhuang Tiedao University,Shijiazhuang 050043,China;Innovation Center for Wind Engineering and Wind Energy Technology of Hebei Province,Shijiazhuang 050043,China;Department of Mathematics and Physics,Shijiazhuang Tiedao University,Shijiazhuang 050043,China;School of Civil Engineering,Shijiazhuang Tiedao University,Shijiazhuang 050043,China)

机构地区：[1]石家庄铁道大学省部共建交通工程结构力学行为与系统安全国家重点实验室,石家庄050043 [2]河北省风工程和风能利用工程技术创新中心,石家庄050043 [3]石家庄铁道大学数理系,石家庄050043 [4]石家庄铁道大学土木工程学院,石家庄050043

出　　处：《计算力学学报》2024年第4期626-633,共8页Chinese Journal of Computational Mechanics

基　　金：河北省自然科学基金面上项目(E2022210069);河北省自然科学基金创新研究群体项目(E2022210078);河北省教育厅青年拔尖人才项目(BJ2019004);国家自然科学基金青年项目(11802186,12202291);科技冬奥专项(21475402D);河北省高端人才项目(冀办[2019]63号)资助项目.

摘　　要：在计算流体力学中,网格质量和生成效率对整个算例数值结果精度及运算效率的影响至关重要.本文在BP算法非结构网格阵面推进法的基础上,为达到优化效果,利用阵元法向建立节点筛选算法对生成的网格分类筛选,并构建神经网络结构以及重新设置损失函数和激活函数.最后,利用优化后的方法生成了若干二维非结构三角形网格(以圆柱传热问题、NACA-0012翼型流动问题和一组单节理裂纹扩展问题为例),采用特定方法测试了网格质量和生成耗时,并将其与传统阵面推进法和优化前的结合神经网络阵面推进法进行对比.结果表明,在保证网格质量不下降的前提下,尤其在针对复杂外形网格划分中,优化后的方法较传统阵面推进法缩短了约43%生成时间,较优化前的结合神经网络阵面推进法缩短约25%生成时间,可见网格生成效率明显提高.In computational fluid dynamics,mesh quality and generation efficiency are very important to the accuracy and efficiency of numerical analysis.In this paper,a node-screening algorithm is established by using the normal of advancing front to classify and screen the generated grids on the basis of the BP algorithm of unstructured grid front advancing method.We also construct a neural network structure and reset loss function and activation function in order to achieve the optimization effect.Finally,several two-dimensional unstructured triangular meshes are generated by using the optimized method (taking cylinder heat transfer problems,NACA-0012 airfoil flow problems and a group of single joint crack diffusion problems as examples).Mesh quality and generation time are tested using specific methods.The results are compared with the traditional front advancing method and the combined neural network front advancing method before optimization.The results show that,on the premise that the grid quality is not degraded,especially for the grid division of complex shapes,the optimized method reduces the generation time by about 43%,compared with the traditional front advancing method,and about 25%,compared with the combined neural network front advancing method before optimization,which shows that the grid generation efficiency is significantly improved.

关 键 词：计算流体力学 机器学习 BP算法 阵面推进法 网格生成 

分 类 号：O35[理学—流体力学]