Abstract
This paper provides a clear perspective on existing several different approaches to robust design optimization of structures. We primarily consider three approaches: the worst-case optimization, the discrepancy (i.e., the maximum gap between the objective values in a nominal case and a possibly occurring case) minimization, and the variance minimization. Some other formulations can also be linked with one of these three approaches. To investigate how the solutions derived by these three approaches differ from each other, we present two numerical examples. This direct comparison clarifies different features of these approaches.
Notes
It should be clear again that, in general, \(\hat {x}_{\text {b}}^{\prime }\) is different both from \(x^{*}_{\text {b}}\) and \(\hat {x}_{\text {b}}\).
Semidefinite programming (SDP) is the optimization of a linear objective function under a linear matrix inequality. Namely, it has the form
$$ \begin{array}{@{}rcl@{}} &&{\kern-.7pc} \text{{Minimize}} {\quad} \sum\limits_{i=1}^{m} b_{i} y_{i} \\ && {\kern-.7pc}\text{{subject~to}}\quad \sum\limits_{i=1}^{m} x_{i} A_{i} + C \succeq 0 , \end{array} $$where \(A_{1},\dots ,A_{m}\), \(C \in \mathbb {R}^{n \times n}\) are constant symmetric matrices and \(b_{1},\dots ,b_{m} \in \mathbb {R}\) are constants.
The stiffness matrix of a truss satisfies K(x + y) = K(x) + K(y) for any x, y ∈ X. Therefore, for any λ ∈ [0, 1], we have
$$ \begin{array}{@{}rcl@{}} &&{\kern-.6pc}{\pi}(\lambda {\boldsymbol{x}} + (1-\lambda){\boldsymbol{y}};{\boldsymbol{q}} ) \\ &&{\kern-.7pc}= \sup_{{\boldsymbol{u}}} \left\{ 2 {\boldsymbol{q}}^{\top} {\boldsymbol{u}} - \left[ \lambda {\boldsymbol{u}}^{\top} K({\boldsymbol{x}}) {\boldsymbol{u}} + (1-\lambda) {\boldsymbol{u}}^{\top} K({\boldsymbol{y}}) {\boldsymbol{u}} \right] \right\} \\ &&{\kern-.7pc}\le \lambda \sup_{{\boldsymbol{u}}} \{ 2 {\boldsymbol{q}}^{\top} {\boldsymbol{u}} - {\boldsymbol{u}}^{\top} K({\boldsymbol{x}}) {\boldsymbol{u}} \} + (1-\lambda) \sup_{{\boldsymbol{u}}} \{ 2 {\boldsymbol{q}}^{\top} {\boldsymbol{u}} - {\boldsymbol{u}}^{\top} K({\boldsymbol{y}}) {\boldsymbol{u}} \} \\ &&{\kern-.7pc}= \lambda {\pi}({\boldsymbol{x}} ;{\boldsymbol{q}} ) + (1-\lambda) {\pi}({\boldsymbol{y}};{\boldsymbol{q}} ) , \end{array} $$which shows the convexity of π(⋅; q). See, e.g., Kanno (2011, Proposition 3.1.2) for more accounts.
Using a vector-valued function as the objective function in (5a) means that this problem is a multi-objective optimization problem (particularly, problem (5a) is a bi-objective optimization problem). That is, we attempt to minimize both \({\pi }(\textit {\textbf {x}}; \tilde {\textit {\textbf {q}}})\) and πworst(x; α). We call x∗∈ X a Pareto optimal solution of (5a) if there exists no \({\boldsymbol {x}}^{\prime } \in \mathbb {R}^{n}\)\(({\boldsymbol {x}}^{\prime } \not = {\boldsymbol {x}}^{*})\) satisfying \({\pi }({\boldsymbol {x}}^{\prime }; \tilde {{\boldsymbol {q}}}) \le {\pi }({\boldsymbol {x}}^{*}; \tilde {{\boldsymbol {q}}})\) and \({\pi }^{{\text {worst}}}({\boldsymbol {x}}^{\prime }; \alpha ) \le {\pi }^{{\text {worst}}}({\boldsymbol {x}}^{*}; \alpha )\). See, e.g., Boyd and Vandenberghe (2004, Section 4.7) and Chankong et al. (1985) for fundamentals of multi-objective optimization.
We adopt the constraint method (a.k.a. ε-constraint method) for scalarization of multi-objective optimization. Application of standard constraint method converts problem (5a) into problem (6a). See, e.g., Haimes et al. (1971), Haimes and Hall (1974), Cohon and Marks (1974), Hwang et al. (1980), Chankong et al. (1985) for details of the constraint method.
We make use of the equality
$$ \begin{array}{@{}rcl@{}} &&{\kern-.7pc}\sup\{ {\boldsymbol{x}}^{\top} {\boldsymbol{y}} \mid \| {\boldsymbol{x}} \|_{\infty} \le 1 \} = \sup\{ {\boldsymbol{x}}^{\top} {\boldsymbol{y}} \mid \| {\boldsymbol{x}} \|_{\infty} = 1 \} \\ &&~~~~~{\kern-.7pc}= \sum\limits_{i=1}^{n} \sup\{ x_{i} y_{i} \mid |x_{i}| = 1 \} = \| {\boldsymbol{y}} \|_{1} . \end{array} $$Here, ∥⋅∥1 is called the dual norm of \(\| \cdot \|_{\infty }\).
Indeed, for any x, y ∈ X and for any λ ∈ [0, 1], we have
$$ \begin{array}{@{}rcl@{}} &&{\kern-.7pc}{\pi}^{{\text{worst}}}(\lambda {\boldsymbol{x}} + (1-\lambda) {\boldsymbol{y}} ;\alpha) \\ &&~~~~~~{\kern-.7pc}= \sup_{{\boldsymbol{q}} \in Q(\alpha)} \{ {\pi}(\lambda {\boldsymbol{x}} + (1-\lambda) {\boldsymbol{y}} ;{\boldsymbol{q}}) \} \\ &&~~~~~~{\kern-.7pc}\le \sup_{{\boldsymbol{q}} \in Q(\alpha)} \{ \lambda {\pi}({\boldsymbol{x}} ;{\boldsymbol{q}}) + (1-\lambda) {\pi}({\boldsymbol{y}} ;{\boldsymbol{q}}) \} \\ &&~~~~~~{\kern-.7pc}\le \lambda \sup_{{\boldsymbol{q}} \in Q(\alpha)} \{ {\pi}({\boldsymbol{x}} ;{\boldsymbol{q}}) \} + (1-\lambda) \sup_{{\boldsymbol{q}} \in Q(\alpha)} \{ {\pi}({\boldsymbol{y}} ;{\boldsymbol{q}}) \} \\ &&~~~~~~{\kern-.7pc}\le \lambda {\pi}^{{\text{worst}}}({\boldsymbol{x}} ;\alpha) + (1-\lambda) {\pi}^{{\text{worst}}}({\boldsymbol{y}} ;\alpha) , \end{array} $$where the first inequality follows the convexity of π(⋅; q). See, e.g., Hiriart-Urruty and Lemaréchal (1993, Proposition IV.2.1.2) for more accounts.
See also Calafiore and Dabbene (2008, Proposition 2).
We have \(\sup \{ {\boldsymbol {x}}^{\top } {\boldsymbol {y}} \mid \| {\boldsymbol {x}} \|_{1} \le 1 \} = \| {\boldsymbol {y}} \|_{\infty }\), because dual of the dual norm is the norm itself; see, e.g., Hiriart-Urruty and Lemaréchal (1993, Proposition V.3.2.1).
DC decomposition of a DC function is not unique. For example, for any convex function \(f : \mathbb {R}^{n} \to \mathbb {R}\), we see that πdisc(⋅; α) can be decomposed as
$$ \begin{array}{@{}rcl@{}} {\pi}^{{\text{disc}}}({\boldsymbol{x}}; \alpha) = ({\pi}^{{\text{worst}}}({\boldsymbol{x}}; \alpha) + f({\boldsymbol{x}})) - ({\pi}({\boldsymbol{x}}; \tilde{{\boldsymbol{q}}}) + f({\boldsymbol{x}})) , \end{array} $$which is also a DC decomposition of πdisc(⋅; α).
Indeed, with the first-order approximation of \({\pi }(\tilde {{\boldsymbol {q}}}+{\boldsymbol {\zeta }})\), (18) follows
$$ \begin{array}{@{}rcl@{}} {\mathrm{E}}[{\pi}(\tilde{{\boldsymbol{q}}}+{\boldsymbol{\zeta}})] \simeq {\mathrm{E}}[ {\pi}(\tilde{{\boldsymbol{q}}}) + \nabla{\pi}(\tilde{{\boldsymbol{q}}})^{\top} {\boldsymbol{\zeta}} ] = {\pi}(\tilde{{\boldsymbol{q}}}) + \nabla{\pi}(\tilde{{\boldsymbol{q}}})^{\top} {\mathrm{E}}[{\boldsymbol{\zeta}}] \end{array} $$and (20) follows
$$ \begin{array}{@{}rcl@{}} &&{\kern-.7pc}{\text{Var}}[{\pi}(\tilde{{\boldsymbol{q}}}+{\boldsymbol{\zeta}})] \\ &&{\kern-.7pc}~~~~~~~= {\mathrm{E}} \left[ ({\pi}(\tilde{{\boldsymbol{q}}}+{\boldsymbol{\zeta}}) - {\mathrm{E}}[{\pi}(\tilde{{\boldsymbol{q}}}+{\boldsymbol{\zeta}})])^{2} \right] \\ &&{\kern-.7pc}~~~~~~~\simeq {\mathrm{E}} \left[ ({\pi}(\tilde{{\boldsymbol{q}}}) + \nabla{\pi}(\tilde{{\boldsymbol{q}}})^{\top} {\boldsymbol{\zeta}} - {\pi}(\tilde{{\boldsymbol{q}}}))^{2} \right] \\ &&{\kern-.7pc}~~~~~~~= {\mathrm{E}}\left[ \nabla{\pi}(\tilde{{\boldsymbol{q}}})^{\top} ({\boldsymbol{\zeta}} {\boldsymbol{\zeta}}^{\top}) \nabla{\pi}(\tilde{{\boldsymbol{q}}}) \right] \\ &&{\kern-.7pc}~~~~~~~= \nabla{\pi}(\tilde{{\boldsymbol{q}}})^{\top} {\mathrm{E}}\left[ {\boldsymbol{\zeta}} {\boldsymbol{\zeta}}^{\top} \right] \nabla{\pi}(\tilde{{\boldsymbol{q}}}) . \end{array} $$Here, the explicit dependency of π on x has been suppressed for notational simplicity.
It is clear that \(\max \limits \{ {\pi }(\textit {\textbf {x}};\textit {\textbf {q}}) \mid \textit {\textbf {q}} \in Q(\alpha ) \}\), as well as \(\min \limits \{ {\pi }({\boldsymbol {x}};{\boldsymbol {q}}) \mid {\boldsymbol {q}} \in Q(\alpha ) \}\), is attained at a point on the boundary of Q(α).
For α = 10kN, there exists almost no difference between the solution with minimal nominal-case compliance and the solution with minimal worst-case compliance.
In the following we assume without loss of generality that the original design optimization problem is formulated as a minimization problem, as the case in the previous sections.
References
Alefeld G, Mayer G (2000) Interval analysis: theory and applications. J Comput Appl Math 121:421–464
Asadpoure A, Tootkaboni M, Guest JK (2011) Robust topology optimization of structures with uncertainties in stiffness—application to truss structures. Comput Struct 89:1131–1141
Ben-Haim Y (1994) Fatigue lifetime with load uncertainty represented by convex model. J Eng Mech (ASCE) 120:445–462
Ben-Haim Y (1995) A non-probabilistic measure of reliability of linear systems based on expansion of convex models. Struct Saf 17:91–109
Ben-Haim Y, Elishakoff I (1990) Convex models of uncertainty in applied mechanics. Elsevier, New York
Ben-Tal A, El Ghaoui L, Nemirovski A (2009) Robust optimization. Princeton University Press, Princeton
Beyer H-G, Sendhoff B (2007) Robust optimization—a comprehensive survey. Comput Methods Appl Mech Eng 196:3190–3218
Boyd S, Vandenberghe L (2004) Convex optimization. Cambridge University Press, Cambridge
Calafiore GC, Dabbene F (2008) Optimization under uncertainty with applications to design of truss structures. Struct Multidiscip Optim 35:189–200
Chankong V, Haimes YY, Thadathil J, Zionts S (1985) Multiple criteria optimization; a state of the art review. In: Haimes YY, Chankong V (eds) Decision making with multiple objectives. Springer, Berlin, pp 36–90
Chen W, Allen JK, Tsui K-L, Mistree F (1996) A procedure for robust design: minimizing variations caused by noise factors and control factors. J Mech Des 118:478–485
Chen W, Fu W, Biggers SB, Latour RA (2000) An affordable approach for robust design of thick laminated composite structure. Optim Eng 1:305–322
Chen S, Lian H, Yang X (2002) Interval static displacement analysis for structures with interval parameters. Int J Numer Methods Eng 53:393–407
Cherkaev E, Cherkaev A (2003) Principal compliance and robust optimal design. J Elast 72:71–98
Cherkaev E, Cherkaev A (2008) Minimax optimization problem of structural design. Comput Struct 86:1426–1435
Choi JH, Lee WH, Park JJ, Youn BD (2008) A study on robust design optimization of layered plate bonding process considering uncertainties. Struct Multidiscip Optim 35:531–540
Cohon JL, Marks DH (1974) A review and evaluation of multiobjective programing techniques. Water Resour Res 11:208–220
Collobert R, Sinz F, Weston J, Bottou L (2006) Large scale transductive SVMs. J Mach Learn Res 7:1687–1712
Cornuéjols G, Pena J, Tütüncü R (2018) Optimization methods in finance, 2nd edn. Cambridge University Press, Cambridge
da Silva GA, Cardoso EL, Beck AT (2019) Non-probabilistic robust continuum topology optimization with stress constraints. Struct Multidiscip Optim 59:1181–1197
de Gournay F, Allaire G, Jouve F (2008) Shape and topology optimization of the robust compliance via the level set method. ESAIM: Control Optim Calc Var 14:43–70
Ganzerli S, Pantelides CP (1999) Load and resistance convex models for optimum design. Struct Optim 17:259–268
Grant M, Boyd S (2008) Graph implementations for nonsmooth convex programs. In: Blondel V, Boyd S, Kimura H (eds) Recent advances in learning and control (a tribute to M. Vidyasagar). Springer, pp 95–110
Grant M, Boyd S (2019) CVX: Matlab software for disciplined convex programming, Ver. 2.1 http://cvxr.com/cvx/ (Accessed: July 2019)
Haimes YY, Hall WA (1974) Multiobjectives in water resource systems analysis: the surrogate worth trade off method. Water Resour Res 10:615–624
Haimes YY, Lasdon LS, Wismer DA (1971) On a bicriterion formulation of the problems of integrated system identification and system optimization. IEEE Trans Syst Man Cybern 1:296–297
Han JS, Kwak BM (2004) Robust optimization using a gradient index: MEMS applications. Struct Multidiscip Optim 27:469–478
Hashimoto D, Kanno Y (2015) A semidefinite programming approach to robust truss topology optimization under uncertainty in locations of nodes. Struct Multidiscip Optim 51:439–461
Hiriart-Urruty J-B, Lemaréchal C (1993) Convex analysis and minimization algorithms, vol I. Springer, Berlin
Holmberg E, Thore C-J, Klarbring A (2015) Worst-case topology optimization of self-weight loaded structures using semi-definite programming. Struct Multidiscip Optim 52:915–928
Huan Z, Zhenghong G, Fang X, Yidian Z (2019) Review of robust aerodynamic design optimization for air vehicles. Arch Comput Methods Eng 26:685–732
Hwang CL, Paidy SR, Yoon K, Masud ASM (1980) Mathematical programming with multiple objectives: a tutorial. Comput Oper Res 7:5–31
Ito M, Kogiso N, Hasegawa T (2018) A consideration on robust design optimization problem through formulation of multiobjective optimization. J Adv Mech Des Syst Manuf 12:18–00076
Kanno Y (2011) Nonsmooth mechanics and convex optimization. CRC Press, Boca Raton
Kanno Y (2015) A note on formulations of robust compliance optimization under uncertain loads. J Struct Construct Eng (Trans AIJ) 80:601–607
Kanno Y (2018) Robust truss topology optimization via semidefinite programming with complementarity constraints: a difference-of-convex programming approach. Comput Optim Appl 71:403–433
Kanno Y (2019) A data-driven approach to non-parametric reliability-based design optimization of structures with uncertain load. Struct Multidiscip Optim 60:83–97
Keshavarzzadeh V, Fernandez F, Tortorelli DA (2017) Topology optimization under uncertainty via non-intrusive polynomial chaos expansion. Comput Methods Appl Mech Eng 318:120–147
Kim N-K, Kim D-H, Kim D-W, Kim H-G, Lowther DA, Sykulski JK (2010) Robust optimization utilizing the second-order design sensitivity information. IEEE Trans Magn 46:3117–3120
Kitayama S, Yamazaki K (2014) Sequential approximate robust design optimization using radial basis function network. Int J Mech Mater Des 10:313–328
Kogiso N, Ahn W-J, Nishiwaki S, Izui K, Yoshimura M (2008) Robust topology optimization for compliant mechanisms considering uncertainty of applied loads. J Adv Mech Des Syst Manuf 2:96–107
Kriegesmann B, Lüdeker JK (2019) Robust compliance topology optimization using the first-order second-moment method. Struct Multidiscip Optim 60:269–286
Lee K-H, Park G-J (2006) A global robust optimization using kriging based approximation model. JSME Int J Ser C Mech Syst Mach Element Manuf 49:779–788
Le Thi HA, Pham Dinh T (2005) The DC (difference of convex functions) programming and DCA revisited with DC models of real world nonconvex optimization problems. Ann Oper Res 133:23–46
Le Thi HA, Pham Dinh T (2018) DC Programming and DCA: thirty years of developments. Math Program 169:5–68
Lipp T, Boyd S (2016) Variations and extension of the convex–concave procedure. Optim Eng 17:263–287
Luo Y, Kang Z, Luo Z, Li A (2009) Continuum topology optimization with non-probabilistic reliability constraints based on multi-ellipsoid convex model. Struct Multidiscip Optim 39:297–310
Markowitz H (1952) Portfolio selection. J Financ 7:77–91
Nakazawa Y, Kogiso N, Yamada T, Nishiwaki S (2016) Robust topology optimization of thin plate structure under concentrated load with uncertain load position. J Adv Mech Des Syst Manuf 10:16–00232
Neumaier A, Pownuk A (2007) Linear systems with large uncertainties, with applications to truss structures. Reliab Comput 13:149–172
Pantelides CP, Ganzerli S (1998) Design of trusses under uncertain loads using convex models. J Struct Eng (ASCE) 124:318–329
Park G-J, Lee T-H, Lee KH, Hwang K-H (2006) Robust design: an overview. AIAA J 44:181–191
Pólik I (2005) Addendum to the SeDuMi user guide: Version 1.1. Technical Report, Advanced Optimization Laboratory. McMaster University, Hamilton. http://sedumi.ie.lehigh.edu/sedumi/
Rao SS, Cao L (2002) Optimum design of mechanical systems involving interval parameters. J Mech Des (ASME) 124:465–472
Shimoyama K, Lim JN, Jeong S, Obayashi S, Koishi M (2009) Practical implementation of robust design assisted by response surface approximation and visual data-mining. J Mech Des 131:061007
Shin YS, Grandhi RV (2001) A global structural optimization technique using an interval method. Struct Multidiscip Optim 22:351–36
Sobieszczanski-Sobieski J, Morris A, van Tooren MJL (2015) Multidisciplinary design optimization supported by knowledge based engineering. Wiley, Chichester
Sriperumbudur BK, Lanckriet GRG (2009) On the convergence of the concave-convex procedure. Adv Neural Inf Process Syst 22:1759–1767
Sturm JF (1999) Using SeDuMi 1.02, a MATLAB toolbox for optimization over symmetric cones. Optim Methods Softw 11–12:625–653
Su J, Renaud JE (1997) Automatic differentiation in robust optimization. AIAA J 35:1072–1079
Sun G, Song X, Baek S, Li Q (2014) Robust optimization of foam-filled thin-walled structure based on sequential Kriging metamodel. Struct Multidiscip Optim 49:897–913
Takezawa A, Nii S, Kitamura M, Kogiso N (2011) Topology optimization for worst load conditions based on the eigenvalue analysis of an aggregated linear system. Comput Methods Appl Mech Eng 200:2268–2281
Tootkaboni M, Asadpoure A, Guest JK (2012) Topology optimization of continuum structures under uncertainty—a polynomial chaos approach. Comput Methods Appl Mech Eng 201–204:263– 275
Toyoda M, Kogiso N (2015) Robust multiobjective optimization method using satisficing trade-off method. J Mech Sci Technol 29:1361–1367
Valdebenito MA, Schuëller GI (2010) A survey on approaches for reliability-based optimization. Struct Multidiscip Optim 42:645–663
Yao W, Chen X, Luo W, van Tooren M, Guo J (2011) Review of uncertainty-based multidisciplinary design optimization methods for aerospace vehicles. Prog Aerosp Sci 47:450–479
Yonekura K, Kanno Y (2010) Global optimization of robust truss topology via mixed integer semidefinite programming. Optim Eng 11:355–379
Yuille AL, Rangarajan A (2003) The concave-convex procedure. Neural Comput 15:915–936
Zang C, Friswell MI, Mottershead JE (2005) A review of robust optimal design and its application in dynamics. Comput Struct 83:315–326
Zhang X, He J, Takezawa A, Kang Z (2018a) Robust topology optimization of phononic crystals with random field uncertainty. Int J Numer Methods Eng 115:1154–1173
Zhang W, Kang Z (2017) Robust shape and topology optimization considering geometric uncertainties with stochastic level set perturbation. Int J Numer Methods Eng 110:31–56
Zhang Y, Li X, Guo S (2018b) Portfolio selection problems with Markowitz’s mean–variance framework: a review of literature. Fuzzy Optim Decis Making 17:125–158
Zhao Z, Han X, Jiang C, Zhou X (2010) A nonlinear interval-based optimization method with local-densifying approximation technique. Struct Multidiscip Optim 42:559–573
Funding
The work described in this paper is partially supported by Research Grant from the Maeda Engineering Foundation and JSPS KAKENHI 17K06633.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The author declares no conflict of interest.
Additional information
Responsible Editor: Xiaoping Du
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Replication of results
Source codes for solving the optimization problems presented in Section 5 are available online at https://github.com/ykanno22/relative_robust/.
Rights and permissions
About this article
Cite this article
Kanno, Y. On three concepts in robust design optimization: absolute robustness, relative robustness, and less variance. Struct Multidisc Optim 62, 979–1000 (2020). https://doi.org/10.1007/s00158-020-02503-9
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00158-020-02503-9