基于PDE灵敏度滤波器的算法研究

孟换利; 张岐良; 王杰

doi:10.21656/1000-0887.430064

基于PDE灵敏度滤波器的算法研究

doi: 10.21656/1000-0887.430064

孟换利^1,,
张岐良^1, ,,
王杰²

1.
中山大学系统科学与工程学院，广州 510006
2.
天津航天长征火箭制造有限公司，天津 300462

基金项目: 中央高校基本科研业务费（76210-31610007）

详细信息

作者简介:
孟换利（1995—），女，硕士生（E-mail：menghli@mail2.sysu.edu.cn）

张岐良（1985—），男，助理教授，博士，硕士生导师（通讯作者. E-mail：zhangqliang@mail.sysu.edu.cn）

中图分类号: O241.82
计量
- 文章访问数: 624
- HTML全文浏览量: 590
- PDF下载量: 137
- 被引次数: 0
出版历程
- 收稿日期: 2022-03-01
- 修回日期: 2022-12-15
- 网络出版日期: 2023-01-06
- 刊出日期: 2023-01-01

Algorithm Research Based on the PDE Sensitivity Filter

1.
School of Systems Science and Engineering, SUN Yat-Sen University, Guangzhou 510006, P.R.China
2.
Tianjin Aerospace Long March Rocket Manufacturing Co. Ltd. , Tianjin 300462, P.R.China

摘要

摘要:
采用PDE灵敏度滤波器可以消除连续体结构拓扑优化结果存在的棋盘格现象、数值不稳定等问题，且PDE灵敏度滤波器的实质是具有Neumann边界条件的Helmholtz偏微分方程。针对大规模PDE灵敏度滤波器的求解问题，有限元分析得到其代数方程，分别采用共轭梯度算法、多重网格算法和多重网格预处理共轭梯度算法对代数方程进行求解，并且研究精度、过滤半径以及网格数量对拓扑优化效率的影响。结果表明：与共轭梯度算法和多重网格算法相比，多重网格预处理共轭梯度算法迭代次数最少，运行时间最短，极大地提高了拓扑优化效率。
- PDE滤波器 /
- Helmholtz方程 /
- 多重网格算法 /
- 共轭梯度算法 /
- 拓扑优化
Abstract:
The PDE sensitivity filter can eliminate the checkerboard patterns and numerical instability existing in the topology optimization results of continuum structures, and the essence of the PDE sensitivity filter is the Helmholtz partial differential equation with Neumann boundary conditions. To solve the large-scale PDE sensitivity filter problem, the conjugate gradient algorithm, the multigrid algorithm and the multigrid preconditioned conjugate gradient algorithm were used to solve the algebraic equations obtained by finite element analysis, and the effects of accuracy, filter radius and grid numbers on the efficiency of topology optimization were studied. The results show that, compared with the conjugate gradient algorithm and the multi-grid algorithm, the multi-grid preconditioned conjugate gradient algorithm has the least number of iterations and the shortest running time, which greatly improves the efficiency of topology optimization.
- PDE filter /
- Helmholtz equation /
- multigrid algorithm /
- conjugate gradient algorithm /
- topology optimization

HTML全文

图 1 基于SIMP插值模型连续体结构拓扑优化实现流程图

Figure 1. The flow chart of continuum structure topology optimization based on the SIMP interpolation model

下载: 全尺寸图片幻灯片

图 2 悬臂梁

Figure 2. The cantilever beam

下载: 全尺寸图片幻灯片

图 3 Michell结构

Figure 3. The Michell structure

下载: 全尺寸图片幻灯片

图 4 不同精度下CG、MG和MGCG算法的悬臂梁的优化结构：(a) P=1E−2 (CG)；(b) P=1E−2 (MG)；(c) P=1E−2 (MGCG)；(d) P=1E−3 (CG)；(e) P=1E−3 (MG)；(f) P=1E−3 (MGCG)；(g) P=1E−4 (CG)；(h) P=1E−4 (MG)；(i) P=1E−4 (MGCG)；(j) P=1E−5 (CG)；(k) P=1E−5 (MG)；(l) P=1E−5 (MGCG)

Figure 4. The optimized cantilever beams by CG, MG and MGCG algorithms with different degrees of precision: (a) P=1E−2 (CG); (b) P=1E−2 (MG); (c) P=1E−2 (MGCG); (d) P=1E−3 (CG); (e) P=1E−3 (MG); (f) P=1E−3 (MGCG); (g) P=1E−4 (CG); (h) P=1E−4 (MG); (i) P=1E−4 (MGCG); (j) P=1E−5 (CG); (k) P=1E−5 (MG); (l) P=1E−5 (MGCG)

下载: 全尺寸图片幻灯片

图 5 不同精度下CG、MG和MGCG算法的Michell结构的优化结构：(a) P=1E−2 (CG)；(b) P=1E−2 (MG)；(c) P=1E−2 (MGCG)；(d) P=1E−3 (CG)；(e) P=1E−3 (MG)；(f) P=1E−3 (MGCG)；(g) P=1E−4 (CG)；(h) P=1E−4 (MG)；(i) P=1E−4 (MGCG)；(j) P=1E−5 (CG)；(k) P=1E−5 (MG)； (l) P=1E−5 (MGCG)

Figure 5. The optimized Michell structures by CG, MG and MGCG algorithms with different degrees of precision: (a) P=1E−2 (CG); (b) P=1E−2 (MG); (c) P=1E−2 (MGCG); (d) P=1E−3 (CG); (e) P=1E−3 (MG); (f) P=1E−3 (MGCG); (g) P=1E−4 (CG); (h) P=1E−4 (MG); (i) P=1E−4 (MGCG); (j) P=1E−5 (CG); (k) P=1E−5 (MG); (l) P=1E−5 (MGCG)

下载: 全尺寸图片幻灯片

图 6 悬臂梁结构优化时间分布图

Figure 6. The optimized time distribution of the cantilever beam

下载: 全尺寸图片幻灯片

图 7 Michell结构优化时间分布图

Figure 7. The optimized time distribution of the Michell structure

下载: 全尺寸图片幻灯片

图 8 不同过滤半径下CG、MG和MGCG算法的悬臂梁的优化结构：(a) R=3 (CG)；(b) R=3 (MG)；(c) R=3 (MGCG)；(d) R=6 (CG)；(e) R=6 (MG)；(f) R=6 (MGCG)；(g) R=9 (CG)；(h) R=9 (MG)；(i) R=9 (MGCG)；(j) R=12 (CG)；(k) R=12 (MG)；(l) R=12 (MGCG)

Figure 8. The optimized cantilever beams by CG, MG and MGCG algorithms with different filter radius: (a) R=3 (CG); (b) R=3 (MG); (c) R=3 (MGCG); (d) R=6 (CG); (e) R=6 (MG); (f) R=6 (MGCG); (g) R=9 (CG); (h) R=9 (MG); (i) R=9 (MGCG); (j) R=12 (CG); (k) R=12 (MG); (l) R=12 (MGCG)

下载: 全尺寸图片幻灯片

图 9 不同过滤半径下CG、MG和MGCG算法的Michell结构的优化结构：(a) R=3 (CG)；(b) R=3 (MG)；(c) R=3 (MGCG)；(d) R=6 (CG)；(e) R=6 (MG)；(f) R=6 (MGCG)；(g) R=9 (CG)；(h) R=9 (MG)；(i) R=9 (MGCG)；(j) R=12 (CG)；(k) R=12 (MG)；(l) R=12 (MGCG)

Figure 9. The optimized Michell structures by CG, MG and MGCG algorithms with different filter radius: (a) R=3 (CG); (b) R=3 (MG); (c) R=3 (MGCG); (d) R=6 (CG); (e) R=6 (MG); (f) R=6 (MGCG); (g) R=9 (CG); (h) R=9 (MG); (i) R=9 (MGCG); (j) R=12 (CG); (k) R=12 (MG); (l) R=12 (MGCG)

下载: 全尺寸图片幻灯片

图 10 悬臂梁结构优化时间分布图(不同过滤半径)

Figure 10. The optimized time distribution of the cantilever beam (different filter radius)

下载: 全尺寸图片幻灯片

图 11 Michell结构优化时间分布图(不同过滤半径)

Figure 11. The optimized time distribution of the Michell structure (different filter radius)

下载: 全尺寸图片幻灯片

图 12 不同网格数量下CG、MG和MGCG算法的悬臂梁的优化结构：(a) 网格为240 × 80 (CG)；(b) 网格为240 × 80 (MG)；(c) 网格为240 × 80 (MGCG)；(d) 网格为480 × 160 (CG)；(e) 网格为480 × 160 (MG)；(f) 网格为480 × 160 (MGCG)；(g) 网格为960 × 320 (CG)；(h) 网格为960 × 320 (MG)；(i) 网格为960 × 320 (MGCG)

Figure 12. The optimized cantilever beams by CG, MG and MGCG algorithms with different grids: (a) grid 240 × 80 (CG); (b) grid 240 × 80 (MG); (c) grid 240 × 80 (MGCG); (d) grid 480 × 160 (CG); (e) grid 480 × 160 (MG); (f) grid 480 × 160 (MGCG); (g) grid 960 × 320 (CG); (h) grid 960 × 320 (MG); (i) grid 960 × 320 (MGCG)

下载: 全尺寸图片幻灯片

图 13 不同网格数量下CG、MG和MGCG算法的Michell结构的优化结构：(a) 网格为240 × 120 (CG)；(b) 网格为240 × 120 (MG)；(c) 网格为240 × 120 (MGCG)；(d) 网格为480 × 240 (CG)；(e) 网格为480 × 240 (MG)；(f) 网格为480 × 240 (MGCG)；(g) 网格为960 × 480 (CG)；(h) 网格为960 × 480 (MG)；(i) 网格为960 × 480 (MGCG)

Figure 13. The optimized Michell structures by CG, MG and MGCG algorithms with different grids: (a) grid 240 × 120 (CG); (b) grid 240 × 120 (MG); (c) grid 240 × 120 (MGCG); (d) grid 480 × 240 (CG); (e) grid 480 × 240 (MG); (f) grid 480 × 240 (MGCG); (g) grid 960 × 480 (CG); (h) grid 960 × 480 (MG); (i) grid 960 × 480 (MGCG)

下载: 全尺寸图片幻灯片

图 14 悬臂梁结构优化时间分布图(不同网格数量)

Figure 14. The optimized time distribution of the cantilever beam (different grids)

下载: 全尺寸图片幻灯片

图 15 Michell结构优化时间分布图(不同网格数量)

Figure 15. The optimized time distribution of the Michell structure (different grids)

下载: 全尺寸图片幻灯片

算法1
1 input:
2　level k;
3　initial guess $ {\boldsymbol{u}}_k^m $;
4　right hand side vector $ {{\boldsymbol{f}}_k} $;
5　coefficient matrix under level $ k $ grid $ {{\boldsymbol{A}}_k} $;
6 output:
7　updated solution $ {\boldsymbol{u}}_k^{m + 1} $
8 pre-smooth: $ {\boldsymbol{u}}_k^{m + 1} = {{\boldsymbol{S}}^{{\mu _1}}}({{\boldsymbol{A}}_k},{{\boldsymbol{f}}_k},{\boldsymbol{u}}_k^m) $
9 coarse grid correction:
10　compute residual: $ {\boldsymbol{r}}_k^m = {{\boldsymbol{f}}_k} - {{\boldsymbol{A}}_k}{\boldsymbol{u}}_k^m $
11　restrict the residual to coarse-grid: $ {\boldsymbol{r}}_{k + 1}^m = {\boldsymbol{R}}_k^{k + 1}{\boldsymbol{r}}_k^m $
12　if k + 1=L then
13　　solve approximate error solution $ \tilde {\boldsymbol{e}}_{k + 1}^m = {\boldsymbol{A}}_{k + 1}^{ - 1}{\boldsymbol{r}}_{k + 1}^m $
14　else
15　　$ \tilde {\boldsymbol{e}}_{k + 1}^m{\text{ = }}V \cdot {\text{cycle(}}{{\boldsymbol{A}}_{k + 1}},{\boldsymbol{r}}_{k + 1}^m,0,k + 1) $
16　end
17 interpolate:
18　$ \tilde {\boldsymbol{e}}_k^m = P_{k + 1}^k\tilde {\boldsymbol{e}}_{k + 1}^m $
19　$ {\boldsymbol{u}}_k^{m + 1} = {\boldsymbol{u}}_k^{m + 1} + \tilde {\boldsymbol{e}}_k^m $
20 post-smooth:
21　$ {\boldsymbol{u}}_k^{m + 1} = {{\boldsymbol{S}}^{{\mu _2}}}({{\boldsymbol{A}}_k},{{\boldsymbol{f}}_k},{\boldsymbol{u}}_k^{m + 1}) $。

下载: 导出CSV

算法2
input:
symmetric positive definite coefficient matrix $ {\boldsymbol{A}} $;
right hand side vector $ {\boldsymbol{b}} $;
initiate a guess for $ {{\boldsymbol{x}}_0} \in {\mathbb{R}^n} $
output:
solution $ {\boldsymbol{x}} $;
1 compute the residual on the finest grid $ {{\boldsymbol{r}}_0} \mathop {\rm{ = }}\limits^\Delta {\boldsymbol{b}} - {\boldsymbol{A}}{{\boldsymbol{x}}_0} $;
2 for $ i = 1,2,3, \cdots $ do
3 solve $ {\boldsymbol{A}}{{\boldsymbol{z}}_0} = {{\boldsymbol{r}}_0} $ by the multigrid V-cycle algorithm
4　initialize $ {{\boldsymbol{p}}_0} = {{\boldsymbol{z}}_0} $
5　compute a step length the $ {\alpha _i} = ({\boldsymbol{r}}_{i - 1}^{\text{T}}{{\boldsymbol{z}}_{i - 1}})/({\boldsymbol{p}}_{i - 1}^{\text{T}}{\boldsymbol{A}}{{\boldsymbol{p}}_{i - 1}}) $
(where $ {{\boldsymbol{z}}_{i - 1}} $ can be solved by the multigrid V-cycle algorithm)
6　update the approximate solution $ {{\boldsymbol{x}}_i} = {{\boldsymbol{x}}_{i - 1}} + {\alpha _i}{{\boldsymbol{p}}_{i - 1}} $
7　update the residual $ {{\boldsymbol{r}}_i} = {{\boldsymbol{r}}_{i - 1}} - {\alpha _i}{\boldsymbol{A}}{{\boldsymbol{p}}_{i - 1}} $
8　solve $ {\boldsymbol{A}}{{\boldsymbol{z}}_i} = {{\boldsymbol{r}}_i} $ by the multigrid V-cycle algorithm
9　compute a gradient correction factor $ {\beta _i} = ({\boldsymbol{r}}_i^{\rm{T}}{{\boldsymbol{z}}_i})/({\boldsymbol{r}}_{i - 1}^{\rm{T}}{{\boldsymbol{z}}_{i - 1}}) $
10　set the new search direction $ {{\boldsymbol{p}}_i} = {{\boldsymbol{z}}_i} + {\beta _i}{{\boldsymbol{p}}_{i - 1}} $
11 end
12 until convergence。

下载: 导出CSV

表 1 两种结构的拓扑优化迭代次数和柔度值

Table 1. Total numbers of iterations and compliance for topology optimization of 2 structures

structure	precision P	CG algorithm			MG algorithm			MGCG algorithm
structure	precision P	compliance C	time T/s	iterations I	compliance C	time T/s	iterations I	compliance C	time T/s	iterations I
cantilever beam	1E−2	186.885 2	87	170	187.306 9	96	167	184.738 3	64	114
	1E−3	186.883 4	82	155	186.939 3	100	160	184.414 7	54	100
	1E−4	186.884 2	84	155	186.939 2	105	160	184.014 0	75	133
	1E−5	186.883 9	84	155	186.939 1	111	160	185.215 5	64	113
Michell structure	1E−2	17.125 2	331	387	16.909 5	70	78	16.923 8	68	78
	1E−3	17.123 4	231	271	17.119 9	260	277	16.923 9	67	78
	1E−4	17.123 1	230	271	17.119 9	269	277	16.952 6	80	93
	1E−5	17.123 1	231	271	17.074 7	267	267	16.943 6	62	71

下载: 导出CSV

表 2 两种结构的拓扑优化迭代次数和柔度值(不同过滤半径)

Table 2. Total numbers of iterations and compliance for topology optimization of 2 structures (different filter radius)

structure	filter radius R	CG algorithm			MG algorithm			MGCG algorithm
structure	filter radius R	compliance C	time T/s	iterations I	compliance C	time T/s	iterations I	compliance C	time T/s	iterations I
cantilever beam	3	180.759 1	505	917	182.218 3	147	225	182.352 0	58	103
	6	186.884 2	82	155	186.939 2	101	160	184.014 0	73	133
	9	192.522 9	98	183	193.053 7	64	100	193.096 7	53	97
	12	197.418 7	73	136	197.669 5	67	105	197.641 9	62	115
Michell structure	3	16.738 5	840	926	16.864 5	165	168	16.894 1	80	90
	6	17.123 1	239	271	17.119 9	280	277	16.952 6	84	93
	9	17.382 9	117	133	17.322 6	115	117	17.168 6	59	67
	12	17.696 4	99	114	17.696 5	107	114	17.686 4	100	114

下载: 导出CSV

表 3 两种结构的拓扑优化迭代次数和柔度值(不同网格数量)

Table 3. Total numbers of iterations and compliance for topology optimization of 2 structures (different grids)

structure	grid size	CG algorithm			MG algorithm			MGCG algorithm
structure	grid size	compliance C	time T/s	iterations I	compliance C	time T/s	iterations I	compliance C	time T/s	iterations I
cantilever beam	240 × 80	186.484 0	16	133	186.315 3	19	133	183.148 4	12	96
	480 × 160	186.884 2	82	155	186.939 2	101	160	184.014 0	73	133
	960 × 320	188.099 5	430	167	188.513 5	495	173	185.695 5	419	155
Michell structure	240 × 120	15.995 4	42	233	15.867 0	20	91	15.816 0	15	83
	480 × 240	17.123 1	235	271	17.119 9	274	277	16.952 6	80	93
	960 × 480	18.293 9	1319	292	18.288 6	148 1	306	18.095 9	361	91

下载: 导出CSV

参考文献(20)

[1]	DEATON J D, GRANDHI R V. A survey of structural and multidisciplinary continuum topology optimization: post 2000[J]. Structural and Multidisciplinary Optimization, 2014, 49: 1-38. doi: 10.1007/s00158-013-0956-z
[2]	ZHU J H, ZHOU H, WANG C, et al. A review of topology optimization for additive manufacturing: status and challenges[J]. Chinese Journal of Aeronautics, 2021, 34(1): 91-110. doi: 10.1016/j.cja.2020.09.020
[3]	卫志军, 申利敏, 关晖, 等. 拓扑优化技术在抑制流体晃荡中的数值模拟研究[J]. 应用数学和力学, 2021, 42(1): 49-57 WEI Zhijun, SHEN Limin, GUAN Hui, et al. Numerical simulation of topology optimization technique fortank sloshing suppression[J]. Applied Mathematics and Mechanics, 2021, 42(1): 49-57.(in Chinese)
[4]	SIGMUND O. Design of material structures using topology optimization[D]. PhD Thesis. Copenhagen : Technical University of Denmark, 1994.
[5]	BRUNS T E, TORTORELLI D A. Topology optimization of non-linear elastic structures and compliant mechanisms[J]. Computer Methods in Applied Mechanics & Engineering, 2001, 190(26/27): 3443-3459.
[6]	吴一帆, 郑百林, 何旅洋, 等. 结构拓扑优化变密度法的灰度单元等效转换方法[J]. 计算机辅助设计与图形学学报, 2017, 29(4): 759-767 WU Yifan, ZHENG Bailin, HE Lüyang, et al. Equivalent conversion method of gray-scale elements for SIMP in structures[J]. Journal of Computer-Aided Design & Computer Graphics, 2017, 29(4): 759-767.(in Chinese)
[7]	XING J, QIE L. A simple way to achieve black-and-white designs in topology optimization[J]. Journal of Physics Conference Series, 2021, 1798(1): 012043. doi: 10.1088/1742-6596/1798/1/012043
[8]	廉睿超, 敬石开, 何志军, 等. 拓扑优化变密度法的灰度单元分层双重惩罚方法[J]. 计算机辅助设计与图形学学报, 2020, 32(8): 1349-2583 LIAN Ruichao, JING Shikai, HE Zhijun, et al. A hierarchical double penalty method of gray-scale elements for SIMP in Topology optimization[J]. Journal of Computer-Aided Design & Computer Graphics, 2020, 32(8): 1349-2583.(in Chinese)
[9]	雷阳, 封建湖. 基于参数化水平集法的材料非线性子结构拓扑优化[J]. 应用数学和力学, 2021, 42(11): 1150-1160 LEI Yang, FENG Jianhu. Topology optimization of nonlinear material structures based on parameterized level set and substructure methods[J]. Applied Mathematics and Mechanics, 2021, 42(11): 1150-1160.(in Chinese)
[10]	LIU B, WU Y, GUO J, et al. Finite difference Jacobian based Newton-Krylov coupling method for solving multi-physics nonlinear system of nuclear reactor[J]. Annals of Nuclear Energy, 2020, 148: 107670. doi: 10.1016/j.anucene.2020.107670
[11]	肖文可, 陈星玎. 求解PageRank问题的重启GMRES修正的多分裂迭代法[J]. 应用数学和力学, 2022, 43(3): 330-340 XIAO Wenke, CHEN Xingding. A modified multi-splitting iterative method with the restarted GMRES to solve the PageRank problem[J]. Applied Mathematics and Mechanics, 2022, 43(3): 330-340.(in Chinese)
[12]	LIU T. A nonlinear multigrid method for inverse problem in the multiphase porous media flow[J]. Applied Mathematics and Computation, 2018, 320(C): 271-281.
[13]	LIU T. A multigrid-homotopy method for nonlinear inverse problems[J]. Computers & Mathematics With Applications, 2020, 79(6): 1706-1717.
[14]	DENG L J, HUANG T Z, ZHAO X L, et al. Signal restoration combining Tikhonov regularization and multilevel method with thresholding strategy[J]. Journal of the Optical Society of America A: Optics, Image Science, and Vision, 2013, 30(5): 948-955. doi: 10.1364/JOSAA.30.000948
[15]	WAN F, YIN Y, ZHANG Q, et al. Analysis of parallel multigrid methods in real-time fluid simulation[J]. International Journal of Modeling, Simulation, and Scientific Computing, 2017, 8(4): 121-128.
[16]	AMIR O, AAGE N, LAZAROV B S. On multigrid-CG for efficient topology optimization[J]. Structural and Multidisciplinary Optimization, 2014, 49(5): 815-829. doi: 10.1007/s00158-013-1015-5
[17]	LIAO Z, ZHANG Y, WANG Y, et al. A triple acceleration method for topology optimization[J]. Structural and Multidisciplinary Optimization, 2019, 60(2): 727-744. doi: 10.1007/s00158-019-02234-6
[18]	LAZAROV B S, SIGMUND O. Filters in topology optimization based on Helmholtz-type differential equations[J]. International Journal for Numerical Methods in Engineering, 2011, 86(6): 765-781. doi: 10.1002/nme.3072
[19]	LAZAROV B S, SIGMUND O. Sensitivity filters in topology optimisation as a solution to Helmholtz type differential equation[C]// Proceedings of the Eighth World Congress on Structural and Multidisciplinary Optimization. Lisbon, Portugal, 2009.
[20]	ANDREASSEN E, CLAUSEN A, SCHEVENELS M, et al. Efficient topology optimization in MATLAB using 88 lines of code[J]. Structural and Multidisciplinary Optimization, 2011, 43(1): 1-16. doi: 10.1007/s00158-010-0594-7