Research Papers: Design Automation

Efficient Filtering in Topology Optimization via B-Splines1

[+] Author and Article Information
Mingming Wang

Mechanical, Materials and Aerospace
Engineering Department,
Illinois Institute of Technology,
Chicago, IL 60616
e-mail: mwang11@hawk.iit.edu

Xiaoping Qian

Department of Mechanical Engineering,
University of Wisconsin,
Madison, WI 53706
e-mail: qian@engr.wisc.edu

An earlier version of this paper appeared in 2014 ASME International Design Engineering Technical Conferences.

Matlab implementation of the 2D version is available at http://cdm.me.wisc.edu/code/btop85.htm.

2Corresponding author.

Contributed by the Design Automation Committee of ASME for publication in the JOURNAL OF MECHANICAL DESIGN. Manuscript received April 9, 2014; final manuscript received December 3, 2014; published online January 9, 2015. Assoc. Editor: Shinji Nishiwaki.

J. Mech. Des 137(3), 031402 (Mar 01, 2015) (10 pages) Paper No: MD-14-1228; doi: 10.1115/1.4029373 History: Received April 09, 2014; Revised December 03, 2014; Online January 09, 2015

This paper presents a B-spline based approach for topology optimization of three-dimensional (3D) problems where the density representation is based on B-splines. Compared with the usual density filter in topology optimization, the new B-spline based density representation approach is advantageous in both memory usage and central processing unit (CPU) time. This is achieved through the use of tensor-product form of B-splines. As such, the storage of the filtered density variables is linear with respect to the effective filter size instead of the cubic order as in the usual density filter. Numerical examples of 3D topology optimization of minimal compliance and heat conduction problems are demonstrated. We further reveal that our B-spline based density representation resolves the bottleneck challenge in multiple density per element optimization scheme where the storage of filtering weights had been prohibitively expensive.

Copyright © 2015 by ASME
Your Session has timed out. Please sign back in to continue.



Grahic Jump Location
Fig. 1

Checkerboard appear in (a) can be suppressed either with fewer knot intervals (b) or high B-spline degrees (c). (a) p = 2, D28 × 56 × 112, (b) p = 2, D7 × 14 × 28, and (c) p = 8, D28 × 56 × 112. The cross sections of three respective designs are shown in the second row: (d) cross section of (a), (e) cross section of (b), and (f) cross section of (c).

Grahic Jump Location
Fig. 2

Obtaining bivariate basis function N(x, y) through tensor product Bp(x)Bq(y). (a) Univariate basis functions B(x) and B(y) and (b) bivariate basis functions N(x, y).

Grahic Jump Location
Fig. 3

Representing density distribution in 3D by B-splines. The intensity in B-spline coefficients corresponds to material density values.

Grahic Jump Location
Fig. 4

Optimized density representation from degree p = 3, q = 2 B-splines of 30 × 10 knot intervals with analysis done by 30 × 10 quadrilateral linear elements [2]. (a) Distribution of 33 × 12 B-spline coefficients ρi (small circle) and (b) 3D view of density distribution ρ˜(x) and the B-spline coefficients ρi.

Grahic Jump Location
Fig. 5

The CPU time (a) and memory usage (b) of density filter with different filter radius R for four tasks in each optimization iteration

Grahic Jump Location
Fig. 6

Problem specification of (a) cantilever beam and (b) heat conduction

Grahic Jump Location
Fig. 7

Topology optimization results using different degrees of B-splines at 14 × 18 × 56 knot intervals. (a) p2, I326, c6470, (b) p6, I583, c5884, and (c) p10, I942, c6325.

Grahic Jump Location
Fig. 8

3D results optimized with degree 2 B-splines of different B-spline knot intervals. (a) Δ3, I460, c5230 (b) Δ6, I326, c6470, and (c) Δ12, I378, c7238.

Grahic Jump Location
Fig. 9

3D results optimized with the density filter with analysis mesh 28 × 36 × 112. (a) R1.5, I452, c5324, (b) R3.5, I367, c6389, and (c) R5.0, I675, c7357.

Grahic Jump Location
Fig. 10

The plot of CPU time (a) and memory usage (b) of B-spline representation for cantilever beam problem with different number of knot intervals

Grahic Jump Location
Fig. 11

Topology optimization results of heat conduction problem with degree 2 B-splines of different knot interval numbers. The analysis mesh is 100 × 100 × 100. (a) p2, D100 × 100 × 100, memory: 84 MB, (b) p2, D80 × 80 × 80, memory: 84 MB, (c) p2, D60 × 60 × 60, memory: 84 MB, and (d) p2, D40 × 40 × 40, memory: 84 MB.

Grahic Jump Location
Fig. 12

The plot of CPU time (a) and memory usage (b) of B-spline representation with different number of knot intervals for heat conduction problems

Grahic Jump Location
Fig. 13

Illustration of multiple density Q4/M25 [12]. (a) MTOP scheme.

Grahic Jump Location
Fig. 14

Density filter with analysis mesh: 14 × 18 × 36. R∕h = 2. (a) M = 1, c = 5656, (b) M = 23, c = 5777, (c) M = 33, c = 6479, and (d) M = 43, c = 7543.

Grahic Jump Location
Fig. 15

Using quadratic B-splines with analysis mesh 42 × 54 × 108. (a) M = 1, c = 3697, (b) M = 23, c = 3637, (c) M = 33, c = 3884, and (d) M = 43, c = 4126.




Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In