Lecture - 1
Reference: Book Chapters 3 and 4
Op#miza#on
in
Prac#ce
with
MATLAB:
for
Engineering
Students
and
Professionals
Op#miza#on
in
Prac#ce
with
MATLAB:
2
for
Engineering
Students
and
Professionals
§ In
prac#ce,
doing
something
as
well
as
possible
within
prac#cal
constraints
is
very
sa#sfactory.
§ Op#miza#on
provides
us
with
the
means
to
make
things
happen/
work
in
the
best
possible
prac#cal
way.
§ Brute-‐force
or
manual
op#miza#on
is
done
by
trial
and
error,
and
using
past
experience,
which
for
most
prac#cal
problems:
§ Will
lead
to
highly
sub-‐op#mal
solu#ons,
since
only
few
trials
can
be
performed
in
a
limited
#me,
and/or
§ Is
too
#me
consuming
to
be
prac#cally
feasible.
§ This
is
where
quan(ta(ve
op(miza(on
comes
in:
§ Uses
mathema#cal
strategies
to
provide
an
efficient
and
systema#c
way
to
op#mize.
§ Using
the
capabili#es
of
modern
day
compu#ng,
these
mathema#cal
strategies
become
all
the
more
powerful
in
implementa#on.
Op#miza#on
in
Prac#ce
with
MATLAB:
3
for
Engineering
Students
and
Professionals
§ Box
A
shows
two
inputs
–
1.
the
dream
design
and
2.
the
ini#al
design.
§ Box
B
shows
the
analysis
phase.
§ Box
D
is
where
the
design
is
improved
in
a
very
systema#c
way.
§ Box
E
shows
manual
op#miza#on
by
a
human
being
(trial
&
error/intui#ve)
Op#miza#on
in
Prac#ce
with
MATLAB:
4
for
Engineering
Students
and
Professionals
Might
take
forever
Not
an
organized
without
reaching
approach
desired
performance
Tradi#onal
Non-‐Opt:
Do
trial
and
error.
Point
A
Point
B
(Ini#al
bad
design)
(Great
op#mal
design)
Op#miza#on
Yield
op(mal
designs
in
a
reasonable
(meframe
A
systema(c
approach
Op#miza#on
in
Prac#ce
with
MATLAB:
5
for
Engineering
Students
and
Professionals
§ Undergraduates:
To
acquire
the
ability
to
op#mize
designs
yourself,
and
feel
comfortable
and
confident
with
the
results
§ As
a
student,
in
the
classroom
seXng
(e.g.,
Capstone
Design
project);
§ As
an
engineer
(post
gradua#on),
in
the
industry
seXng.
§ Graduate
Students:
To
be
able
to
use
op#miza#on
to
find
be\er
ways
to
proceed
with
your
experiments,
system
modeling,
or
designs,
in
the
course
of
your
ongoing
and
future
research.
§ Industry
Personnel:
To
acquire
the
ability
to
leverage
the
immense
poten#al
of
op#miza#on
in
different
real-‐life
projects.
§ Learn
so^ware
tools
to
readily
apply
op#miza#on
in
your
projects;
§ Acquire
knowledge
to
be
able
to
cri#cally
verify
the
op#mal
designs
§ Acquire
knowledge
to
be
able
to
iden#fy
the
challenges
in
op#mizing
a
system,
and
to
know
where
to
look
for
the
solu#ons.
Op#miza#on
in
Prac#ce
with
MATLAB:
6
for
Engineering
Students
and
Professionals
§ Analysis
and
Op*miza*on
are
two
core
components
of
a
systems
design
process.
Op#miza#on
in
Prac#ce
with
MATLAB:
7
for
Engineering
Students
and
Professionals
§ Engineering
Analysis
can
be
defined
as:
The
applica*on
of
scien*fic
principles
and
processes
to
reveal
the
proper*es
and
the
state
of
a
system,
and
also
understand
the
underlying
physics
driving
the
system
behavior.
§ Analysis
generally
demands
disciplinary
knowledge
per#nent
to
the
system
or
mechanism
being
analyzed.
§ Prac#cal
systems
involve
mul#ple
disciplines,
e.g.,
designing
an
aircra^
requires
structural,
aerodynamic,
and
control
analyses.
§ If
disciplinary
understanding
has
reached
certain
level
of
maturity,
mathema#cal
analysis
tools
might
be
readily
available.
§ On
the
other
hand,
in
the
case
of
mechanisms
or
phenomena
that
are
not
yet
well
understood,
in-‐depth
and
fundamental
analysis
might
be
required
thereby
demanding
the
involvement
of
a
disciplinary
expert.
Op#miza#on
in
Prac#ce
with
MATLAB:
8
for
Engineering
Students
and
Professionals
§ Analysis
need
not
be
a
purely
mathema#cal
or
theore#cal
process,
especially
in
designing
new
systems.
§ Analysis
could
involve
“experiments–tes#ng–mathema#cal
inference”
as
an
itera#ve
process.
§ Such
an
approach
is
necessary
due
to
any
of
the
following
reasons:
1. The
underlying
physics
is
not
well
understood;
the
fundamental
disciplinary
principles
or
theory
do
not
directly
apply;
2. There
are
geometrical
complexi#es
and
inherent
uncertain#es;
or
3. There
is
a
lack
of
knowledge
of
the
material
proper#es
(e.g.,
thermodynamic
or
structural
proper#es).
Op#miza#on
in
Prac#ce
with
MATLAB:
9
for
Engineering
Students
and
Professionals
§ Design
in
general
terms
can
be
defined
as
the
crea*on
of
a
plan
and/or
strategy
for
construc*ng
a
physical
system
or
process.
§ Based
on
the
“object
of
design”,
engineering
design
could
be
classified
into:
1. product
design
2. systems
design
3. industrial
design,
and
4. process
design
Op#miza#on
in
Prac#ce
with
MATLAB:
10
for
Engineering
Students
and
Professionals
§ Prac#cal
engineering
design
can
be
perceived
as
a
mul#-‐stage
process.
§ Overlap
or
itera#on
among
the
different
stages
is
o^en
prevalent
and
necessary.
Op#miza#on
in
Prac#ce
with
MATLAB:
11
for
Engineering
Students
and
Professionals
§ Mathema*cal
op*miza*on
is
the
process
of
maximizing
and/or
minimizing
one
or
more
objec*ves
without
viola*ng
specified
design
constraints,
by
regula*ng
a
set
of
variable
parameters
that
influence
the
objec*ves
and
the
design
constraints.
§ The
three
types
of
quan##es
in
op#miza#on:
1. Objec(ves:
The
quan##es
that
you
would
like
to
improve
–
e.g.,
fuel
efficiency
of
an
aircra>
(to
be
maximized),
or
manufacturing
cost
of
the
aircra>
(to
be
minimized).
2. Constraints:
The
quan##es
or
criteria
that
your
design
needs
to
sa#sfy
–
e.g.,
the
cargo/payload
capacity
of
an
aircra>.
3. Variables:
The
quan##es
that
you
can
directly
change
to
improve
the
design,
where
the
values
of
the
objec#ves
and
constraints
are
regulated
by
these
quan##es
–
e.g.,
the
dimensions
and
the
material
of
the
aircra>
wing.
Objec*ve
func*ons
and
constraint
func*ons
are
o>en
together
called
criteria
func*ons,
since
one
can
be
converted
into
another.
Op#miza#on
in
Prac#ce
with
MATLAB:
12
for
Engineering
Students
and
Professionals
§ There
does
not
exist
a
unique
structure
to
how
they
are
related
−
the
structure
generally
depends
on
the
available
human,
computa#onal,
and
physical
resources
and
on
the
choices
of
decision-‐makers.
Op#miza#on
in
Prac#ce
with
MATLAB:
13
for
Engineering
Students
and
Professionals
§ Design
in
general
is
the
enveloping
process
that
includes
analysis
and
op*miza*on
as
sub-‐processes.
§ Primary
steps
in
design
include
(i)
defining
the
design
objec#ves,
constraints,
and
variables,
(ii)
performing
or
using
analysis,
and
(iii)
performing
op#miza#on,
(iv)
verifying
the
op#mum
design.
§ Analysis
provides
you
with
the
knowledge
to
relate
the
variable
values
to
the
criteria
func#ons
of
interest
(through
models).
§ Op(miza(on,
which
is
the
main
driver
for
improving
the
design,
uses
the
knowledge
(model)
obtained
through
analyses.
§ Analysis
also
provides
the
opportunity
to
inves#gate
the
performance
of
the
final
op#mum
design.
§ On
the
other
hand,
op(miza(on
could
provide
food
for
further
analysis
–
e.g.,
which
region
of
the
design
space
is
to
be
analyzed
in
more
detail.
Op#miza#on
in
Prac#ce
with
MATLAB:
14
for
Engineering
Students
and
Professionals
§ Engineering
design
could
involve
qualita#ve
elements
that
go
beyond
quan#ta#ve
“analysis”
and
“op(miza(on”,
while
o^en
impac#ng
these
two
quan#ta#ve
ac#vi#es.
§ Crea(vity-‐
and
aesthe(cs-‐driven
decisions:
Although
qualita#ve
in
nature,
these
decisions
have
important
quan#ta#ve
implica#ons
for
the
later
stages
of
design
where
analysis
and
op#miza#on
are
involved
(e.g.,
impact
on
material
op*ons).
§ Market-‐driven
decisions:
Design
specifica#on
can
also
be
driven
by
an
understanding
of
the
market,
especially
in
the
case
industrial
and
product
design.
Although,
quan#ta#ve
market
analysis
might
be
available
in
certain
cases,
such
availability
is
not
necessarily
generic
(e.g.,
imagine
the
first
Iphone).
Op#miza#on
in
Prac#ce
with
MATLAB:
15
for
Engineering
Students
and
Professionals
§ In
the
context
of
design,
analysis
and
op#miza#on
are
related
to
each
other
through
modeling.
§ Modeling:
A
process
by
which
an
engineer
or
a
scien*st
translates
the
actual
physical
system/phenomena
under
study
into
a
set
of
mathema*cal
equa*ons
or
opera*ons.
§ Mathema#cally,
modeling
can
be
represented
as:
𝑷=𝒇(𝑿),
where,
P
the
quan#ty
of
interest,
which
is
expressed
as
a
func#on
of
a
vector
of
design
variables,
X.
§ Mathema#cal
models
may
not
be
a
single
analy#cal
func#on.
Op#miza#on
in
Prac#ce
with
MATLAB:
16
for
Engineering
Students
and
Professionals
§ Physics-‐based
Analy(cal
Models:
These
are
theore#cal
models
derived
from
the
physics
of
the
system
(e.g.,
analy#cal
solu#on
of
the
conduc#on
heat
transfer
differen#al
equa#on).
§ Generally
fast
in
execu#on.
§ O^en
inadequate
for
complex
systems
(e.g.,
involving
complex
geometry).
§ Simula(on-‐based
Models:
These
models
generally
leverage
a
discre#zed
representa#on
of
the
system,
in
transla#ng
the
system
behavior
into
a
set
of
algebraic
equa#ons.
These
set
of
equa#ons
are
solved
using
numerical
techniques,
by
harnessing
the
power
of
computers.
Examples
include
CFD.
§ O^en
#me-‐intensive
(could
take
hours/weeks/months).
§ Generally
adequate
in
represen#ng
system
complexi#es.
§ Surrogate
Models:
Surrogate
models
are
purely
mathema#cal
or
sta#s#cal
models,
with
certain
generic
func#onal
forms
and
coefficients
that
can
be
tuned
using
a
set
of
input-‐output
data
that
is
generated
from
physical
or
simula#on-‐based
experiments.
§ Fast
in
execu#on.
§ Can
represent
complex
systems,
generally
at
the
cost
of
accuracy
or
fidelity.
Op#miza#on
in
Prac#ce
with
MATLAB:
17
for
Engineering
Students
and
Professionals
§ The
success
of
op#miza#on
depends
both
on:
§ the
capabili#es
of
the
op#miza#on
method/algorithm;
§ the
effec#veness
of
the
op(miza(on
formula(on.
§ Modeling
the
op#miza#on
or
problem
formula(on
essen#ally
involves
developing
a
clear
defini#on
of
the
design
variables,
design
objec#ves,
and
design
constraints.
§ Problem
formula#on
is
also
strongly
correlated
with
the
choice
of
op#miza#on
algorithms
–
e.g.
During
problem
formula#on,
One
can
convert
equality
constraints
into
inequality
constraints
using
a
tolerance
value,
in
order
to
leverage
powerful
algorithms
that
perform
well
in
the
absence
of
equality
constraints.
Op#miza#on
in
Prac#ce
with
MATLAB:
18
for
Engineering
Students
and
Professionals
§ It
is
important
to
ensure
that
op(miza(on
problem
formula(on
is
coherent
with
the
system
behavior
model.
§ For
example,
the
input
output
exchange
(between
analysis
and
op#miza#on)
demanded
by
the
op#miza#on
formula#on
should
be
sa#sfied
by
the
capabili#es
of
the
models
used
thereof.
§ The
choice
of
analysis
models
also
affects
the
choice
of
op#miza#on
algorithm,
and
vice
versa.
For
example:
§ If
you
choose
an
algorithm
that
generally
requires
a
large
number
of
system
evalua#ons,
then
a
fast
model
of
the
system
behavior
is
needed.
§ If
the
system
behavior
model
is
inherently
highly
nonlinear,
you
will
need
to
formulate
the
op#miza#on
problem
such
that
a
nonlinear
op#miza#on
algorithm
can
be
used
to
solve
the
problem.
Op#miza#on
in
Prac#ce
with
MATLAB:
19
for
Engineering
Students
and
Professionals
