CHAP 8 STRUCTURAL DESIGN USING

FINITE ELEMENTS

FINITE ELEMENT ANALYSIS AND DESIGN

Nam-Ho Kim

1
 INTRODUCTION
• FEA: determining the response of a given structure for a
given set of loads and boundary conditions
– Geometry, material properties, BCs and loads are well defined
• Engineering design: a process of synthesis in which parts are
put together to build a structure that will perform a given set of
functions satisfactorily
• Analysis is systematic and can be taught easily; design is an
iterative process

• Creative design: creating a new structure or machine that

does not exist
• Adaptive design: modifying an existing design (evolutionary
process)

2
 INTRODUCTION – STRUCTURAL DESIGN
• Structural design: a procedure to improve or enhance the
performance of a structure by changing its parameters
• Performance: a measurable quantity (constraint and goal)
– the weight, stiffness or compliance; the fatigue life; noise and vibration
levels; safety
• Constraint: As long as the performance satisfies the criterion,
its level is not important
– Ex: the maximum stress should be less than the allowable stress
• Goal: the performance that the engineer wants to improve as
much as possible
• Design variables: system parameters that can be changed
during the design process
– Plate thickness, cross-sectional area, shape, etc

3
 SAFETY MARGIN
• Factor of Safety (stress performance)
– Structures should satisfy a constraint:
– σi(x): the calculated stress at a point x in the i-th component
– σallowable: allowable stress from material strength (failure strength)

– Factor of safety: effect of material variability, experimental errors, etc

• For general type of performance, use response and capacity
– Response Ri: calculated value of performance
– Capacity Ci: allowed value of performance

– When multiple loads are applied simultaneously

4
 SAFETY MARGIN cont.
• Safety Margin: the excess capacity compared with the
response

– The member can afford additional Zi stress before it fails

• Sufficiency Factor Si: ratio of the allowable capacity to the

response

– Normalized in terms of the capacity

5
 LOAD FACTOR
• Instead of dividing the capacity by SF, increase applied loads
• Load factor λ: the factor by which the applied loads must be
multiplied just enough to cause the structure to fail
• Let a set of loads be F = {F1, F2, …, FN}T. For a given failure
strength Ci, the load factor is defined by

• For proportional loading with linear system

• Load factor is the ratio between capacity and response

• The structure should be designed in order to satisfy a given
level of load factor
6
 EXAMPLE
• Design the height h of cantilevered beam with SF = 1.5
– E = 2.9×104 ksi, w = 2.25 in.
1) Allowable tip deflection Dallowable = 2.5 in. (No need SF)
– FE equation after applying BCs

– FE solution

L = 100 in
h
w

F = 2,000 lb
7
 EXAMPLE cont.
2) Failure strength = 40 ksi (Need SF)
– Supporting moment at the wall

– Maximum stress at the wall

– Height calculation with the factor of safety

8
 FULLY-STRESSED DESIGN
• For complex structure with different failure criteria needs an
iterative process to find the best design – Intuitive design
• Fully-Stressed Design: Very useful with stress and minimum
gage constraints
For the best design, each member of the structure that
is not at its minimum gage is fully stressed under at
least one of the design load conditions

– From the observation that all members will be in the allowable stress
level unless they reach the minimum gage
– It is based on an assumption that the effect of adding or removing
material from a member is to change the stresses in that member
– Works well with statically determinate structure (member force does
not change when the cross-section increases)

9
 FSD cont.
• Assume the loads carried by members remained constant
• Stress in each member is calculated
• Member is resized to bring the stress to the allowable value
• Truss example (member force F = σ·A)

• If member forces are constant (statically determinant system),

one iteration to come up FSD
• Repeated calculations are required when member forces vary
• If Anew is less than the minimum gage, choose min. gage

10
 FSD cont.
• Beam element: Use section modulus
– Maximum bending stress

Section modulus

– Update section modulus

– If M remains constant, one iteration yields FSD

• The section modulus in beam corresponds to cross-sectional
area in truss

11
 FSD EXAMPLE – CANTILEVERED BEAM
• Initial w = 2.25, h = 3.5 in. Determine new height using FSD
• Section modulus and max. stress at the initial design

• New section modulus using stress ratio resizing

Only one iteration

for FSD
L = 100 in
h
w

F = 2,000 lb 12
 DESIGN PARAMETERS
• Selecting design variables – easy for beam and truss, but
more complicated for plane or 3D solids
• Material property design variable
– Varying material properties to find the best material
– Not common, but useful for designing composite materials
• Sizing design variable
– Geometric parameters as design (parametric design variable)
– Appears as a parameter in FEM
– Thickness of plate/shell, cross-sectional geometry of truss/beam, etc

t b b
r b r t

h h
h
t w
w
13
 DESIGN PARAMETERS cont.
• Shape design variable
– Related to the structure’s geometry, which does not appear explicitly
as a parameter
– Beam cross-section is a geometry, but it appears as a moment of
inertia
– Cx, Cy, and r determine the size and location of the hole
– Shape design variables change FE mesh
– Design variables must be limited so that the hole remains inside of the
plate

Cx r
Cx r

Cy
Cy

(a) Initial design (b) Perturbed design

14
 DESIGN PARAMETERS cont.
• Shape design variable cont.
– Shape design variable describe the movement of boundary
– Mesh inside of structure must move accordingly
• Iso-parametric mapping method
– Works well with mapped mesh, maintaining topological mesh
– Convenient on tracking the same elements and nodes
– Possible mesh distortion for large shape-changing design

r r

(a) Initial mesh (b) Perturbed mesh

15
 PARAMETER STUDY – SENSITIVITY ANALYSIS
• Parameter study
– Effect of a design variable on performance (gradual change of DV)
– Cantilevered beam example:

w (in) h (in) σmax (ksi)

2.0 4.0 37.5
Allowable 2.0 4.5 29.6
stress 2.0 5.0 24.0
2.5 4.0 30.0
2.5 4.5 23.7
Acceptable 2.5 5.0 19.2
region 3.0 4.0 25.0
3.0 4.5 19.8
3.0 5.0 16.0

16
 SENSITIVITY ANALYSIS
• Parameter study becomes too expensive with many DVs
• Unable to capture rapid change in performance locally
• Design sensitivity analysis computes the rate of performance
change with respect to design variables
• Sensitivity analysis calculates gradient of performance for
optimization
• Explicit dependence
– Analytical relationship exists between performance and DVs
– Weight of circular cross-section beam

– Sensitivity w.r.t. r:

17
 SENSITIVITY ANALYSIS cont.
• Implicit dependence
– Performance depends on DVs through state variable (displacement)
– Sensitivity of stress:

Difficult to calculate, time consuming

Easy to calculate from given expression of stress
• How to calculate displacement sensitivity?
– Differentiate finite element equation:

– [dK/db] and {dF/db} can be evaluated using either their analytical

expression or numerical differentiation
18
 SENSITIVITY ANALYSIS cont.
• Sensitivity equation must be solved for each design variable
• Sensitivity equation uses the same stiffness matrix with the
original finite element analysis
• Consider RHS as a pseudo-force vector
• Similar to finite element analysis with multiple load cases
• Thus, solving sensitivity equation is very inexpensive using
factorized stiffness matrix
• General form of performance

– Sensitivity Implicit dependent term

Explicit dependent term 19

 FINITE DIFFERENCE SENSITIVITY
• Easiest way to compute sensitivity information of the
performance
• Calculate performance at two different designs
• Forward difference method

• Central difference method

• Consider FEA as a black-box

• Sensitivity computation cost becomes high for many design
variables
– n+1 analyses for forward FDM
– 2N+1 analyses for central FDM
20
 FINITE DIFFERENCE SENSITIVITY cont.
• Accuracy of finite difference sensitivity
– Accurate results can be expected when Δb approaches zero
– For nonlinear performances, a large perturbation yields completely
inaccurate results
– Numerical noise becomes dominant for a too-small perturbation size

(dΗ/db)4

(dΗ/db)3

(dΗ/db)2

(dΗ/db)1

b0 b1 b2 b3 b4 b
21
 EXAMPLE – CANTILEVERED BEAM
• At optimum design (w=2.25 in, h=4.47 in), calculate sensitivity
of tip displacement w.r.t. h
• Exact sensitivity:

• Differentiate [K]

• Pseudo load vector

22
 EXAMPLE – CANTILEVERED BEAM cont.
• Sensitivity equation:

• Same way of applying BC

• Sensitivity of nodal DOFs

– Same with the exact sensitivity

23
 STRUCTURAL OPTIMIZATION
• What Is Design Optimization?
– To find the best design parameters that meet the design goal and
satisfies constraints.
• Design Parameters: Anything the Designer Can Change
– Thickness of a plate
– Cross-sectional geometry of a beam or truss
– Geometric dimensions
• Design Goal: Objective Function
– Design criterion that will be minimized (or maximized)
– Mass, Stress, Displacement, Natural Frequency, ETC
• Constraint: Conditions that the system must satisfy
– Stress, Displacement, ETC
• Note: Design parameters must affect the design goal and
constraints.

24
OPTIMIZATION FLOW CHART
Physical engineering
problem

Structural modeling

Design parameterization

Performance definition
(cost, constraints)

Structural analysis

model update
(FEM, BEM, CFD…)

Structural
Design sensitivity analysis

Optimized?
No
Yes
Stop
25
 THREE-STEP PROBLEM FORMULATION
w
1. Design Parameterization
• Clear identification t1
• Independence of designs
t2 h

2. Objective Function
• Must be a function of design parameters
• Minimization ( –Maximization)
3. Constraint Functions
• Inequality constraints
• Equality constraints
• Equality constraints must be less than the number of design
parameters

26
 STANDARD FORM
• Standard form of design optimization

• Feasible set: the set of designs that satisfy constraints

27
 BEER CAN EXAMPLE
Design the beer can size so that the D
minimum amount of sheet metal
can be used
(minimize manufacturing cost) H

Constraints
– It is required to hold at least 400 ml of fluid.
– The diameter of the can should be no more
than 8 cm. In addition, it should not be less
than 3.5 cm (shipping & handling).
– The height of the can should be no more
than 18 cm and no less than 8 cm.

28
 BEER CAN EXAMPLE cont.
Standard
Form

Feasible
Domain

29
 GLOBAL OPTIMIZATION
• Global Optimum
– A point b* is called a global minimum for f(b) if

• There is no mathematical method to find the global minimum

• Weierstrass theorem: Existence of global minimum
– If f(b) is continuous and the set S is closed and bounded, then there
is a global minimum
• Local Optimum
– A point b* is called a local minimum for f(b) if
f(b) Feasible set

for all b∈S in a small neighborhood of b*

• Global optimum: b3
• Local optima: b1, b3, b5 b2 b4 b5 b6
b
b1 b3 30
 GRAPHICAL OPTIMIZATION
• Graphical methods can be used to solve the optimization
problem when it has only one or two design variables
• Graphical methods are expensive but help to visualize the
design space and to understand the nature of design problem

1. Draw the design space (lower- & upper-bounds of DVs)

2. Plot constraints on the graph (feasible set)
3. Plot contour lines of objective function
4. Find the optimum point (the objective function has the lowest
value within the feasible set)

31
EXAMPLE – BEER CAN OPTIMIZATION

32
 NUMERICAL METHODS
• In FEA, we don’t know the explicit relationship, f(b), between
design parameters and objective (or constraint) function.
• But, we can calculate f(b) for a given b by solving the finite
element equation.
• Then, how can we find the optimum design?

Use Numerical Method

• Numerical Method
– From the current design, find the next design that can reduce the
objective function and satisfy constraints.
– Repeat the search until there is no way to improve the objective
function further.

33
 NUMERICAL METHOD cont.
• Basic Algorithm
1. Start with b(K) and K = 0.
– Initial design must be given
2. Evaluate function values and their gradients.
– Using finite element analysis
3. Using information from Step 2, determine ∆b(K).
– Design change
4. Check for termination.
– Stop if the problem is converged.
5. Update design

6. Increase K = K + 1 and go to Step 2.

– Design iteration (cycle)

34
 NUMERICAL METHOD cont.
• Change in Design

– αK: Step size

– d(K): Search direction vector
– Design change means the determination of search direction and step
size.

35
 OPTIMIZATION USING EXCEL SOLVER
1
• Four-bar truss optimization L

1 3
L
2
2 F 3
4 4

2F

36