1
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Multidisciplinary System
Design Optimization (MSDO)
Sensitivity Analysis
Lecture 8
1 March 2004
Olivier de Weck
Karen Willcox
2
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Today’s Topics
Sensitivity Analysis
– effect of changing design variables
– effect of changing parameters
– effect of changing constraints
Gradient calculation methods
– Analytical and Symbolic
– Finite difference
– Adjoint methods
– Automatic differentiation
3
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Standard Problem Definition
1
2
min ( )
s.t. ( ) 0 1,..,
( ) 0 1,..,
1,..,
j
k
u
iii
J
gjm
hkm
xxxi n
d
d d
x
x
x
A
For now, we consider a single objective function, J(x).
There are n design variables, and a total of m
constraints (m=m
1
+m
2
).
The bounds are known as side constraints.
4
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Sensitivity Analysis
Sensitivity analysis is a key capability aside from the
optimization algorithms we discussed.
Sensitivity analysis is key to understanding which
design variables, constraints, and parameters are
important drivers for the optimum solution x*.
The process is NOT finished once a solution x* has
been found. A sensitivity analysis is part of post-
processing.
Sensitivity/Gradient information is also needed by:
– gradient search algorithms
– isoperformance/goal programming
– robust design
5
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Sensitivity Analysis
How sensitive is the “optimal” solution J* to
changes or perturbations of the design
variables x*?
How sensitive is the “optimal” solution x* to
changes in the constraints g(x), h(x) and
fixed parameters p ?
6
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Sensitivity Analysis: Aircraft
Questions for aircraft design:
How does my solution change if I
change the cruise altitude?
change the cruise speed?
change the range?
change material properties?
relax the constraint on payload?
..
7
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Sensitivity Analysis
Questions for spacecraft design:
How does my solution change if I
change the orbital altitude?
change the transmission frequency?
change the specific impulse of the propellant?
change launch vehicle?
Change desired mission lifetime?
..
8
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Gradient Vector – single objective
“How does the objective function J
value change as we change elements
of the design vector x?”
1
2
n
J
x
J
x
J
x
w
a o
? ?
w
? ?
w ? ?
? ?
w
?
? ?
? ?
? ?
w
? ?
? ?
w
? ?
J
#
Compute partial derivatives
of J with respect to x
i
i
J
x
w
w
?J
Gradient vector points normal
to the tangent hyperplane of J(x)
1
x
2
x
3
x
9
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Geometry of Gradient vector (2D)
0 0.5 1 1.5 2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
x
1
x
2
Contour plot
3
.1
3
.
1
3
.1
3
.
2
5
3
.
2
5
3
.
2
5
3.
25
3
.
2
5
3
.5
3
.5
3.5
3
.
5
4
4
4
4
5
5
Example function:
12 1 2
12
1
,Jxx x x
xx
?
2
11
2
21
1
1
1
1
J
xxx
J
J
xxx
w
a o a o
? ? ? ?
w
? ? ? ?
?
w ? ? ? ?
? ? ? ?
w
? ? ? ?
Gradient normal to contours
10
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Geometry of Gradient vector (3D)
222
123
Jxxx
1
2
3
2
2
2
x
Jx
x
a o
? ?
?
? ?
? ?
? ?
increasing
values of J
1
x
2
x
3
x
Tangent plane
123
22260xxx
> @
111
T
o
x
> @
222
o
T
J ?
x
J=3
Example
Gradient vector points to larger values of J
11
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Taylor Series Expansion
,
kk n
Jwherxx 6 \ 6 Taylor Series Expansion of Objective Function
Tangential
hyperplane
at x
o
Effect of curvature
(2nd derivative)
at x
o
000 0 0
1
() () ()( ) ( )()( )H.O.T.
2
T
T
JJ J
a o
?
? ?
0
xx xxx xxHxxx
first order term
second order term
12
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Jacobian Matrix – multiple objectives
If there is more than one objective function, i.e.
if we have a gradient vector for each J
i
, arrange them
columnwise and get Jacobian matrix:
12
11 1
12
22 2
12
z
z
z
nn n
J JJ
xx x
J JJ
xx x
J JJ
xx x
w w w
a o
? ?
w w w
w w w
? ?
w w w
?
? ?
w w w
w w w
? ?
J
"
"
# # % #
"
1
2
z
J
J
J
a o
? ?
? ?
? ?
? ?
? ?
J
#
n x z
z x 1
13
? Massachusetts Institute of Technology - Prof. de Weck and Prof. Willcox
Engineering Systems Division and Dept. of Aeronautics and Astronautics
Normalization
In order to compare sensitivities from different
design variables in terms of their relative sensitivity
it is necessary to normalize:
i
J
x
w
w
o
x
“raw” - unnormalized sensitivity = partial
derivative evaluated at point x
i,o
,
()
io
ii i
x
JJ J
xx J x
'