Scribd
Upload
23 views

Optimization in Matlab Part 1

The document discusses structural optimization problems and methods. It defines structural optimization, objective functions, design variables, constraints, and different types of structural optimization problems. It then describes various solution strategies and methods for solving constrained structural optimization problems.

Uploaded by

Nhon Nguyen
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
23 views

Optimization in Matlab Part 1

The document discusses structural optimization problems and methods. It defines structural optimization, objective functions, design variables, constraints, and different types of structural optimization problems. It then describes various solution strategies and methods for solving constrained structural optimization problems.

Uploaded by

Nhon Nguyen
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 43

Department of Mechanics, Faculty of Mathematics &

Computer Science, University of Science, Vietnam


National University HCMC

University of Science, October-2012


11 Introduction and concepts

2 Formulation of structural optimization problems

3 Examples of structural optimization problems

14 Solution strategies for no-constrained optimization

5 Solution strategies for constrained optimization

6 Methods for structural optimization problems

17 Reliability-based structural optimization

8 References
Structure Optimization Structural optimization

“any assemblage of Making things the best The subject of


materials which is making an
intended to sustain assemblage of
loads.” materials sustain
loads in the best way.
But “best” in what sense?
Objective A function used to classify designs: measures weight, displacement in
function
(f) a given direction, effective stress or even cost of production

Design
A function or vector that describes the design, and which can be changed
variable
during optimization. It may represent geometry or choice of material (the (x)
area of a bar, or the thickness of a sheet).

State
variable For a given structure, y is a function or vector that represents the
(y) response of the structure (displacement, stress, strain or force…)
Single-Objective Structural Optimization
minimize f(x, y) with respect to x and y
Behavioral constrains on y
(SO)
Subject to Design constraints on x
Equilibrium constraint

Multi-Objective Structural Optimization

minimize f1(x, y), f2(x, y), f3(x, y), ....., fn(x, y)


Behavioral constrains on y
(SO)
Subject to Design constraints on x
Equilibrium constraint
Pareto optimality: a design is Pareto optimal if there does not exist any other design that
satisfies all of the objectives better.

n
Common way to solve  i=1
wi fi ( x, y)
(1) Behavioral constraints: constraints on the state variable y.
Usually they are written g(y) ≤ 0, where g is a function which
represents, e.g., a displacement in a certain direction.

(2) Design constraints: constraints involving the design


variable x. Obviously, these two types of constraints can be
Three
combined. types of
constraints
(3) Equilibrium constraint: K ( x)u  F( x)

 min f ( x, u( x))
Nested formulation (SO)nf  x
s.t. g ( x, u( x))  0
First-order optimality condition
Second-order optimality condition
Constrained optimality
Constrained optimality
Constrained optimality
Constrained optimality
Three types of structural optimization problems

(1) Sizing optimization: This is when x is some type of structural thickness, i.e., cross -
sectional areas of truss members, or the thickness distribution of a sheet.
Three types of structural optimization problems

(2) Shape optimization: x represents the form or contour of some part of the boundary of the
structural domain. The state y is described by a set of partial differential equations.
Connectivity of the structure is not changed.
Three types of structural optimization problems

(3) Topology optimization: In this way the connectivity of nodes is variable. Design variables
x (often present geometric feature) can take the value zero, i.e., bars are removed from the
truss.
(1) Weight Minimization of a Two-Bar Truss Subject to Stress Constraints
(2) Weight Minimization of a Two-Bar Truss Subject to Stress and Instability Constraints
(3) Weight Minimization of a Two-Bar Truss subject to Stress and Displacement Constraints
(4) Weight Minimization of a Three-Bar Truss Subject to Stress Constraints
(5) Fiber orientation optimization in laminate composite

+ The problem with many design variables


 Objective function to be minimized is the strain energy of homogeneous symmetric
laminates including 2*n plies.
 Fibers’s orientation is the design variables.
 Thickness of each plies is constant and equal to t = 1 cm.
 Laminate is in a state of plane stress. The design optimization problem is presented in such
that:

1
min  T A
i 2
0   i  180
(6) Topology optimization in mechanics

N
min c(x)  U KU   ( xe ) p uTe k 0ue
T
x
e 1

 S ( x) / S 0

KU  F
0  x  x  1
 min
(7) Limit analysis of plates

t/2 ε  Bu

min L   0 
t / 2
 T ( z ) dz W (f , u)  1
u  0

(8) Weight Minimization of stiffer in stiffened plate problems

min f  bh
 w1  wload

MU  KU  0
b  0; h  0

Two strategies

Line search Trust regions


4.1 First strategy: Line search min f  xk   pk 
 0

1. Steepest descent (need first derivatives)


2. Newton method (need first and second derivatives)
3. Quasi-Newton (need 1st derivative + approximated 2nd derivative)
4. BFGS method (need 1st derivative + approximated 2nd derivative)
5. Conjugate Gradient (need 1st derivatives)
6. Truncated Newton method (need first and second derivatives)
7. Golden section search (no derivatives)
8. Successive Parabolic Interpolation (SPI) (no derivatives)
9. Safeguarded method (combine Golden section and SPI)
10. Nonlinear least square (need first and second derivatives)
11. Gauss-Newton method (need first derivatives)
12. Levenberg-Marquardt method (need first derivatives)
4.2 Second strategy: trust regions
The information gathered about f is used to construct a model function mk whose behavior
near the current point xk is similar to that of the actual objective function f.

min mk  xk  p  where xk  p lies inside the trust region


p

Model mk is usually defined to be a quadratic function of the form


1 T
mk  xk  p   f k  p f k  p Bk p
T

2
where fk ,∇ fk, and Bk are a scalar, vector, and matrix, respectively. The matrix Bk is either the
Hessian ∇2fk or some approximation to it.
In trust region, we first choose a maximum distance—the trust-region radius k—and then
seek a direction that attain the best improvement possible subject to this distance constraint.
If this step proves to be unsatisfactory, we reduce the distance measure k and try again.
5.1 Equality-constrained optimization
Convert to un-constrained optimization by Lagrange multiply method and then apply
Newton method (or other methods)
5.1 Equality-constrained optimization
5.2 Inequality-constrained optimization

a. If inequality constrains and objective function are linear


=> linear program => Simplex method
5.2 Inequality-constrained optimization
b. If inequality constrains are nonlinear => Barrier method
=> Interior-point method or Primal-dual interior-point method
For the structural problems, it is in general impossible to obtain the explicit form of the
objective function and constraints of the design variables. It is hence necessary to
generate of explicit sub-problems that are approximations of the original problem and
solve these sub-problems instead.
In addition, most problems in structural optimization are non-convex. It is hence
necessary to choose approximations that are convex.
1. Sequential Linear Programming (SLP)
2. Sequential Quadratic Programming (SQP)
3. Convex Linearization (CONLIN)
4. Method of Moving Asymptotes (MMA)
5. Genetic Algorithm (GA)
6. Pattern search
7. Optimality Criteria Method (OC)
8. Discrete level-set topology optimization
N
min c(x)  U KU   ( xe ) p uTe k 0ue
T
x
e 1
 S ( x) / S 0
KU  F

subject to 
0  x min  x  1
   U   t

Two methods:
Method 1:
a. First, conduct the reliability analysis of random variables
b. Perform the structural optimization => optimal design variables
c. Using FORM2 to verify the reliability index of optimal solution
=> Fast, can use various methods
Method 2:
Combine both Reliability analysis (FORM2) and structural optimization simultaneously =>
long, require suitable methods
2. Jorge Nocedal,Stephan J. Wright,
(2000). Numerical Optimization.
Springer.

1. Christensen, P. W., Klarbring, A.,


(2009). An Introduction to Structural
Optimization. Springer.
3. Spillers, W. R., MacBain, K. M.,
(2009). Structural Optimization.
Springer.

4. Bendsoe, M. P., Sigmund, O,


(2003). Topology Optimization.
Springer.
6. Pablo, Pedregal, (2004).
Introduction to optimization,
Springer.

5. Venkataraman, P, (2002 - 2009).


Applied optimization with
Matlab programing, Wiley.

You might also like