Structural Engineering Graduate Course

Civil Engineering Department STR651

Cairo University Faculty of Engineering

STR651
High-Rise Building

Topic # 1 (Part 2)
Structural Optimization

Prepared by: Dr. Hazem Elanwar

1
Outline

 Motivational Example

 Introduction to topology optimization

 Introduction to optimization

 Topology optimization

2
Example

10t
 For the shown truss:
 Find the top lateral deflection
 Assume all member to be IPE100
−10 2 𝑡 5m
 E=2100 t/cm2 +10𝑡

 Solution
• IPE-100  Area=10.3 cm2 10t +10𝑡
• dtop= No  N1 dL
 E A
−20 2 𝑡
−10𝑡 5m
+30𝑡

5m
3
Example

1t
 For the shown truss:
 Find the top lateral deflection
 Assume all member to be IPE100
−1 2 𝑡 5m
 E=2100 t/cm2 +1𝑡

 Solution
• IPE-100  Area=10.3 cm2 +1𝑡
• dtop= No  N1 dL
 E A
• dtop=1/EI*(10x1x500+10 2 x1 2 x500 2 −1 2 𝑡
+10x1x500+30x2x500+20 2x1 2x500 2 −1 𝑡 5m
+ 10x1x500)=4.04cm +2𝑡

5m
4
Example

10t
 For the shown truss:
 Find the top lateral deflection
 Assume all member to be IPE100
 E=2100 t/cm2 5m

 Solution
• IPE-100  Area=10.3 cm2 10t
• dtop= No  N1 dL
 E A
• dtop=1/EA*(…)=4.5cm
5m

• Then the first system is more efficient than

the second system
• How many systems should be studied?

5m
5
Example

10t
 For the shown truss:
 Find the top lateral deflection
 Assume all member to be IPE100
 E=2100 t/cm2 5m

 Solution The shorter the load path:

• IPE-1001-
 The
Area=10.3 cm2 10t
more optimum design
2- The
• dtop= No

N1 dL
smaller the deflection
E A
• dtop=1/EA*(…)=4.5cm
5m

• Then the first system is more efficient than

the second system
• How many systems should be studied?

5m
6
Example

 How many topologies need to be checked?

7
Intro. to Topology Optimization

8
Introduction

 For reference check the following seminar:

 William F. Baker: "On the Harmony of Theory and Practice in the Design
of Tall Buildings"

9
Introduction: William F. Baker Seminar Lecture

 Optimizing the structural components:

 Reduces the project cost
 Protects the environment
• 37% of embodied energy

10
Introduction: William F. Baker Seminar Lecture

 The lateral system optimization is based on Maxwell theorem on

load paths.

11
Introduction: William F. Baker Seminar Lecture

 The lateral system optimization is based on Maxwell’s theorem on

load paths.
 It states that: the sum of compression and tension load paths is always
constant

12
Introduction: William F. Baker Seminar Lecture

 Maxwell’s Theorem on Load Path: Example

 By selecting any topology of the truss, the difference between the
tension and compression path shall be “P.B”.

13
Introduction: William F. Baker Seminar Lecture

 Maxwell’s Theorem on Load Path: Example

 It is logic to assume a truss geometry that follows the moment diagram
of a cantilever beam.

14
Introduction: William F. Baker Seminar Lecture

 Maxwell’s Theorem on Load Path: Example

 However, Warren shape truss is more economic and yields less
deflection!

15
Introduction: William F. Baker Seminar Lecture

 Choice of topology significantly impacts the cost of the structure:

 Truss A: requires 27% more volume of material to satisfy safety limits.
 Truss A: requires 60% more volume of material to give the same
deflection as Truss B

16
Introduction: William F. Baker Seminar Lecture

 Maxwell’s Theorem:

 Since the sum of the tension path

and the compression path is
constant, then:
 The longer the tension path the
longer the compression path and
vice versa
 The efficiency is paid twice in the
tension and compression path
 By knowing the value of P.r, you
need only to calculate the tension or
comp. path to know the other
 If you have only tension or
compression path then this is the
optimum structure
17
Introduction: William F. Baker Seminar Lecture

 Maxwell’s Theorem:

 Since the difference between the

tension path and the compression
path is constant, then:
 The longer the tension path the
longer theTherefore,
compressiona path
benchandmark (lower
vice versa bound solution) is essential to
guide
 In efficiency the twice
is paid designer’s
in the decisions
tension and compression path
 By knowing the value of P.r, you
need only to calculate the tension or
comp. path to know the other
 If you have only tension or
compression path then this is the
optimum structure
18
Introduction: William F. Baker Seminar Lecture

 In order to reach the bench mark design:

 Use analytical approaches: i.e. Michell’s optimal trusses.
 Use optimization techniques: i.e. topology optimization

19
Introduction

 Topology optimization might provide a solution for selecting the

most optimum geometry, shown a comparison between Michell’s
solution and topology optimization software (TopOpt).

20
Introduction

 Topology optimization might provide a solution for selecting the

most optimum geometry, shown a comparison between Michell’s
solution and topology optimization software (TopOpt).

21
Introduction

 Topology optimization might provide a solution for selecting the

most optimum system

22
Introduction to Optimization

23
Optimization

 Optimization can be classified into 3 main categories

 Size
 Shape
 Topology

Topology Optimization: Theory, Methods, and Applications, M.P. Bendsoe, O.Sigmund, Springer, 2004

24
Optimization

 Optimization can be either:

 Gradient Based
 Non-Gradient (Random) Based

 Gradient based optimization

 (+) Utilize the gradient and Hessian
matrix
 (+) Computationally efficient
 (+) Feasible for large scale problems
 (-) The function must be continuous and
differentiable
 (-) The output depends on the initial
guess
 (-) Can be trapped in Local minimum
solution
 Examples: Interior point method, Golden 25
search method, etc.
Optimization

 Optimization can be either:

 Gradient Based
 Non-Gradient (Random) Based

 Non- Gradient based optimization

 (-) Requires long computational time
 (-) Infeasible for large scale problems
 (+) The function doesn’t need to be
continuous nor differentiable
 (+) The output doesn’t depends on the
initial guess
 (+) More efficient when dealing with
Local minimum
 (+) Can handle problems even if the
function is not accessible
 Examples: Genetic Algorithm, Ant 26
Colony, etc.
Optimization

 Optimization problem is divided into:

 Objective (cost) function
 Design variables
 Constraints

 Objective function
 The value that need to be maximized or minimized
 i.e. minimize the volume of member A
 Design variables
 The parameter that can be changed to satisfy the objective function
 i.e. the distance “X”
 Constraints
 The set of limitations that must be satisfied
 i.e. The upper and lower bounds for “X”
 i.e. Linear and non-linear constraints on stress limits, buckling, etc. 27
Optimization: Example

 Solve the following problem:

Prof. Glaucio Paulino Lecture Notes

28
Optimization: Example

 Solve the following problem:

1. Graphical

Prof. Glaucio Paulino Lecture Notes

29
Optimization: Example

 Solve the following problem:

1. Graphical

Prof. Glaucio Paulino Lecture Notes

30
Optimization: Example

 Solve the following problem:

1. Graphical

Prof. Glaucio Paulino Lecture Notes

31
Optimization: Example

 Solve the following problem:

1. Graphical

Prof. Glaucio Paulino Lecture Notes

32
Optimization: Example

 Solve the following problem:

1. Graphical

Prof. Glaucio Paulino Lecture Notes

33
Optimization: Example

 Solve the following problem:

1. Graphical
2. Linear programing (i.e. MATLAB)

Prof. Glaucio Paulino Lecture Notes

34
Optimization: Example

 Solve the following problem:

1. Graphical
2. Linear programing (i.e. MATLAB)
3. Genetic Algorithm (i.e. MATLAB)
Prof. Glaucio Paulino Lecture Notes

35
Optimization: Example

 Solve the following problem:

1. Graphical
2. Linear programing (i.e. MATLAB)
3. Genetic Algorithm (i.e. MATLAB)
Prof. Glaucio Paulino Lecture Notes
4. Interior Point Method (i.e. fmincon)

36
Optimization: Example

 Tools such as GA and IPM are very useful in case of having a

group of complicated linear and non-linear constraints

Prof. Glaucio Paulino Lecture Notes

37
Optimization: Example#2

 Find the dimensions of the box with largest volume if the total
surface area is 64 cm2:
 Obj Fnc: max(X*Y*Z)
 Variables: X, Y, & Z
 Constraints: 2XY+2YZ+2XZ=64 XY+YZ+XZ=32
 Solve using Lagrange Multiplier
 L=XYZ+λ(XY+YZ+XZ-32)
 L,X=dL/dX=YZ+ λ(Y+Z)=0  (*X) -XYZ=λX(Y+Z) …(1)
 L,Y=dL/dY=XZ+ λ(X+Z)=0  (*Y) -XYZ=λY(X+Z) …(2)
 L,Z=dL/dZ=XY+ λ(X+Y)=0  (*Z) -XYZ=λZ(X+Y) …(3)
 L,λ=dL/dλ=XY+YZ+XZ-32=0 …(4)

38
Optimization: Example#2

 Find the dimensions of the box with largest volume if the total
surface area is 64 cm2:
 From eq(1) & eq(2)  λX(Y+Z)=λY(X+Z)
 λ(XZ-YZ)=0
• Either λ=0 (refused, it means the constraint is not applied)
• Or (XZ-YZ)=0  X=Y
 Repeat the same procedure eq(2) & eq(3) to get X=Y=Z
 From eq(4) X2+X2+X2=32
 X=Y=Z=3.266

39
Topology Optimization

40
Introduction

 Topology optimization:
 Given:
• Feasible domain (area or volume)
• Boundary conditions (B.C.)
• Load conditions (L.C.)
• Required openings or holes
 Variable: The density of each
element in the domain should take
a value either 0 or 1.
 Objective function: There are Sigmund et.al, 2011 “ Efficient topology
optimization in MATLAB using 88 lines of code.
several objective functions that
can be utilized. For example min.
compliance problem.
• Min.: L(u)
 Constraints
• aE(u,v)=L(v)

For Future Reference: Topology Optimization, Theory, Methods and Applications (Bendsoe and Sigmund) 41
Objective Function

 A simple objective function is the minimum compliance (maximum

global stiffness)
 Assumptions:
 Eijkl(x) is variable over the domain

 Virtual work at equilibrium (u) and small displacement (v) applies:

• a
. ( u ,v )   Eijkl ( x)   ij (u)   ij (v) d 


dui du j
 Linearize strain:  ij (u)  0.5  (  )
dx j dxi


 Load linear form: L(u)= L(u )  P  u d   t  u d 




42
Solution algorithm

 The optimization problem is as follows

 Its Lagrange formula is as follows

 The derivative of Lagrange equation w.r.t the design variable r is:

 With switching conditions

43
Solution algorithm

 Lagrange multiplier (λ) is either 0 or >0,

 if λ > 0 it means that the term in the bracket must be =0, which implies:
• r=0 or 1 without intermediate value (black or white mesh)
 If λ=0 it means that the term in the bracket must NOT be 0, which
implies that
• r is intermediate value between 0 and 1 (grey mesh)
 The problem start from the intermediate values and we want to reach 0
or 1. In this case we can assume λ+ & λ- =0, Lagrange equation
becomes as follows:

 Define new parameter Beta :

44
Solution algorithm

 The problem is formulated such that the volume ratio to total domain
volume must be a pre specified constant (i.e. 0.5).
 At first Iteration the volume of Lagrange multiplier “Λ” is assumed
using the bisection method Λ mid between Λ min=0 & Λ max=10^6.
 Then the volume constant is checked, if the value is greater than
the required number (i.e. 0.5), then:
 Λ min= Λ mid & Λ max= Λ max, in another word increase “Λ”
 From b equation if “Λ” increase b decreases
 In this case the function must direct selective r(s)  r(s)-move to
decrease overall volume
 Now we can follow the set of equations for iteration procedure

ζ=move (i.e. 0.2)

45
Solution algorithm

 Then the volume constant is checked, if the value is greater than

the required number (i.e. 0.5), then:
 Λ min= Λ mid & Λ max= Λ max, in another word increase “Λ”
 From b equation if “Λ” increase b decreases
 In this case the function must direct selective r(s)  r(s)-move to
decrease overall volume
 Now we can follow the set of equations for iteration procedure

 If beta is small there is a higher possibility to have r r-move. Then

the overall volume is decreased aiming to find optimum solution
 On the other hand, if volume<volume_required (i.e. 0.5), Λmin= Λmin &
Λmax= Λmid, which means Λ decreases and b increase and probably the
second equation will be triggered and selective r(s)= r(s)+move ,
which increases the volume
 It is worth noting that, depends on the optimization criteria some r46will
increase and other will decrease until we reach the 0/1 state
Solution algorithm

47
TopOpt 88 Code

 nelx: no. of meshes in x-

direction.
 nely: no. of meshes in y-
direction.
 volfrac= volume ratio to
the overall domain.
 Numbering system of
nodes follow the figure, it
starts from the top left
corner moving down, then
to the left direction.
 F=to add forces by node
number and value.
 Fixeddofs= to add support
nodes and direction.
48
TopOpt 88 Code

49
TopOpt 88 Code

 nelx: no. of meshes in x-

direction.
 nely: no. of meshes in y-
direction.
 volfrac= volume ratio to
the overall domain.
 Numbering system of
nodes follow the figure, it
starts from the top left
corner moving down, then
to the left direction.
 F=to add forces by node
number and value.
 Fixeddofs= to add support
nodes and direction.
50
TopOpt 88 Code

51
TopOpt Example

 This is a very simple example to describe the

input of TopOpt.m program 1

52
Assignment

53
Assignment: Special Problems

 SP#1:
 Use fmincon to solve the following optimization problem:

 SP#2:
 Use TopOpt.m software to find the optimum topography of the shown
system. Comment on the results.
 Given:
• Nelx=200
• Nely=160
• VolFrac=0.55
1 1
• Penal=3
Row=80
• Rmin=1.5
Columns
• Ft=1 99,101
54