TY - JOUR

T1 - An operator splitting algorithm for Tikhonov-regularized topology optimization

AU - Talischi, Cameron

AU - Paulino, Glaucio H.

N1 - Funding Information:
The authors acknowledge the support by the Department of Energy Computational Science Graduate Fellowship Program of the Office of Science and National Nuclear Security Administration in the Department of Energy under contract DE-FG02-97ER25308, and the National Science Foundation (NSF) through grant #1234243 (Civil, Mechanical and Manufacturing Innovation Division). The information presented in this paper is the sole opinion of the authors and does not necessarily reflect the views of the sponsoring agencies.

PY - 2013/1/1

Y1 - 2013/1/1

N2 - In this work, we investigate a Tikhonov-type regularization scheme to address the ill-posedness of the classical compliance minimization problem. We observe that a semi-implicit discretization of the gradient descent flow for minimization of the regularized objective function leads to a convolution of the original gradient descent step with the Green's function associated with the modified Helmholtz equation. The appearance of " filtering" in this update scheme is different from the current density and sensitivity filtering techniques in the literature. The next iterate is defined as the projection of this provisional density onto the space of admissible density functions. For a particular choice of projection mapping, we show that the algorithm is identical to the well-known forward-backward splitting algorithm, an insight that can be further explored for topology optimization. Also of interest is that with an appropriate choice of the projection parameter, nearly all intermediate densities are eliminated in the optimal solution using the common density material models. We show examples of near binary solutions even for large values of the regularization parameter.

AB - In this work, we investigate a Tikhonov-type regularization scheme to address the ill-posedness of the classical compliance minimization problem. We observe that a semi-implicit discretization of the gradient descent flow for minimization of the regularized objective function leads to a convolution of the original gradient descent step with the Green's function associated with the modified Helmholtz equation. The appearance of " filtering" in this update scheme is different from the current density and sensitivity filtering techniques in the literature. The next iterate is defined as the projection of this provisional density onto the space of admissible density functions. For a particular choice of projection mapping, we show that the algorithm is identical to the well-known forward-backward splitting algorithm, an insight that can be further explored for topology optimization. Also of interest is that with an appropriate choice of the projection parameter, nearly all intermediate densities are eliminated in the optimal solution using the common density material models. We show examples of near binary solutions even for large values of the regularization parameter.

KW - Topology optimization

UR - http://www.scopus.com/inward/record.url?scp=84870545146&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84870545146&partnerID=8YFLogxK

U2 - 10.1016/j.cma.2012.05.024

DO - 10.1016/j.cma.2012.05.024

M3 - Article

AN - SCOPUS:84870545146

VL - 253

SP - 599

EP - 608

JO - Computer Methods in Applied Mechanics and Engineering

JF - Computer Methods in Applied Mechanics and Engineering

SN - 0374-2830

ER -