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