+ All Categories
Home > Documents > Design Search and Optimization_mynotes

Design Search and Optimization_mynotes

Date post: 14-Apr-2018
Category:
Upload: adam-mahmood
View: 216 times
Download: 0 times
Share this document with a friend

of 59

Transcript
  • 7/29/2019 Design Search and Optimization_mynotes

    1/59

    Design Search and Optimization (DSO) (1 lecture)

    First what is designsynthesis vs analysis

    What is optimal design? Are all designs optimal (best)? Might we ever deliberately accept

    sub-optimal design? To answer this we must be able to compare competing designs and say

    which we prefer and hopefully why. (Examples of cars, fridgescost vs performance, trade

    offs).

  • 7/29/2019 Design Search and Optimization_mynotes

    2/59

    What is design and how does design search & optimisation fit into it?

    Engineering Design is the use of analysis to support synthesis (creation) so as to adequately

    define a product (or process).

    Note difference between analysis and synthesis. Synthesis involves decisions whereas

    analysis does not (ie decisions about the product) how big, what material, what

    manufacturing method. Analysis provides the information for rational decisions to be made.

    DSO is a formalism for carrying forward such decision making and is normally thought of as

    an automated activity controlled by a computer code. It involves postulating (guessing) a

    design, analysing it, deciding if the results are acceptable and if not deciding how to change

    it. If it is to be changed the process is repeated until we have an acceptable design or we run

    out of effort/time.

    Design search and optimization is the term used to describe the use of formal optimization

    methods in design.

    Optimization concerns the finding of the inputs to functions that cause those functions to

    reach extreme values.

    Designers are often not interested in finding optimal designs in the strictest sense: rather, they

    wish to improve their designs using the available resources within bounds set by the desire to

    produce balanced, robust, safe and well-engineered products that can be made and sold

    profitably. It is never possible to include all the factors of concern explicitly in computational

    models and so any results produced by numerical optimization algorithms must be tempered

    by judgments that allow for those aspects not encompassed by the models used. Moreover,

    there will often be multiple ways of solving design issues that all have benefits and

    drawbacks and no single clear winning design will emerge: instead, human judgment will be

    needed. Thus, it is more accurate to say that designers use optimization methods in the search

    for improved designshence the term Design Search and Optimization.

    To set this up we adopt the ideas of:

    Objectives

    Design variables and their bounds

    Constraints and their limits

    Fixed parameters

    External noise/uncertain parameters

    Methods of analysis

  • 7/29/2019 Design Search and Optimization_mynotes

    3/59

    Schemes for linking design variables to analysis

    Schemes for linking analysis to objectives and constraints

    Discuss the Nowacki beam problemdistribute his papergive reference to books.

    Nowacki Beam (See Engineering Design via Surrogate Modelling) :

  • 7/29/2019 Design Search and Optimization_mynotes

    4/59

    TAXONOMY OF OPTIMIZATION (METHODS)Optimization may be defined as the search for a set of inputs x (not necessarily numerical) that

    minimize (or maximize) the outputs of a functionf (x) subject to inequality constraintsg(x) 0 andequality constraints h(x) = 0. The functions may be represented by simple expressions, complexcomputer simulations, analog devices or even large-scale experimental facilities.

    Inputs can be either nonnumeric or numeric.Example of nonnumeric inputs, can be the selection of a compressor type as being either axial or

    centrifugal or the selection of a material between aluminium alloy and fibre-reinforced plastic for usein wing construction.

    If all the inputs are numeric, the next division that occurs in problems is between those with

    continuous variable inputs and those that take discrete values (including the integers).Examples of discrete variables are things such as commercially available plate thicknesses,standardized bolt diameters, or the number of blades on a compressor disk. Continuous variableswould include wing-span and fan-blade length.

    The outputs can also be categorized. There can be a single goal to be maximized or minimized (the

    so-called objective function) or are there multiple objectives?If there are multiple goals, thesewill lead to what is termed a Pareto fr ont: a set of solutions that are all equally valid until some

    weight or preference is expressedbetween the goals. For example, if it is desired to reduce weightand cost, these aims may pull the design in opposite directionsuntil it is known which is moreimportant or some weighting between them is chosen, it will not be possible to decide on the best

    design.Next, one can categorize by the presence or absence of constraints. Such constraints may simplyapply to the upper and lower values the inputs may take (when they are commonly known as bounds)or they may involve extremely complex relationships with the inputs (as in the limits typically foundon stress levels in components).The next categorisation is the type offunctional relationship between the inputs and outputs.These can be linear, nonlinear or discontinuous. They can be stationary, time dependent or

    stochastic in nature.o Discontinuous functions make optimization most difficult, and are the normo Linear relationships may lead to some very efficient solutions.

    We can therefore categorize the Nowacki beam design problem as having continuous inputs,

    one or two outputs, various constraints and stationary nonlinear (but not discontinuous)

    relationships linking these quantities.Methods themselves may also be classified in a number of ways. The first division that can bemade is between methods that deal with:

    1. Optimal selection2. Solve linear equations, and3. The rest.

    Optimal selection routines commonly stem from the operational research (OR) community andtypically are set up to deal with idealized traveling salesman or knapsack problems.

    Linear problems are nowadays almost universally solved using linear programming and the so-

    called simplex or revised simplex methods, which can efficiently deal with thousands of

    variables.

    Optimal selection methods and linear programming, while valuable, essentially lie outside the scopeof this book, as they do not find much practical application in aerospace design, which is dominatedby large scale Computational Fluid Dynamics (CFD) and Computational Solid Mechanics (CSM)

    models; they are not discussed further here.

  • 7/29/2019 Design Search and Optimization_mynotes

    5/59

    The remaining methods that deal with nonlinear numeric problems form much the largest collectionof approaches. At the most basic level, such searches may be divided between those that need

    information on the local gradient of the function being searched and those that do not. Searchesthat will workwithout any gradient information may be termed zeroth order while those needing

    the first derivatives are first order, and so on. A further distinction may be made betweenapproaches that can deal with constraints, those that cannot and those that just need feasiblestarting points.Methods may also be categorized by whether they are deterministic in natureor have somestochastic element.

    o Deterministic searches will always yield the same answers if started from the same initialconditions on a given problem; stochastic methods make no such guarantees.

    o You may think that variation of results from run to run ought to be avoided but it turns outthat stochastic search methods are often very robust in nature: a straightforward random walkover the inputs is clearly not repeatable if truly random sequences are usednonetheless,such a very simple search is the only rational approach to take if absolutely no information isavailable on the functions being dealt with.

    A random search is completely unbiased and therefore cannot be misled by features in the problem.

    Although it is almost always the case in design that some prior information is available on thefunctions being dealt with, the pure random search can be surprisingly powerful and it also forms a

    benchmark against which other methods can be measured: if a search cannot improve on a randomwalk, few would argue that it was a very appropriate method!

    A distinction may further be made between searches that work with one design at a time and thosethat seek to manipulate populations or sets of designs.

    Population-based search has gained much ground with the advent of cluster and parallel-basedcomputing architectures since the evaluation of groups of designs may be readily parallelized on suchsystems. Perhaps, the most well-known of such methods are those based on the so-called evolutionarymethods and those that use groups of calculations to construct approximations to the real objectivesand constraints, the so-called Design of Experiment and Response Surface methods.

  • 7/29/2019 Design Search and Optimization_mynotes

    6/59

    SUMMARRY OF TAXONOMY OF OPTIMIZATION

    inputs

    some non-numeric all numeric

    continuous some discrete

    problems

    outputs

    Single goal Multiple goals (Pareto fronts)

    constraints

    unconstrained bounds constrained

    methods

    LINEAR PROBLEMS

    LINEAR PROGRAMMING

    OPTIMAL SELECTION

    INTEGER PROGRAMMINGTHE REST

    operational research

    sorting/exchange methods

    methods

    simplex methods

    search over vertex space

    NOT DISCUSSED

    FURTHER

    NOT DISCUSSED

    FURTHER

    No gradients needed

    (zero order)

    gradients needed

    (first or second order)

    cope with constraints directly only unconstrained

    deterministic stochastic

    population based one at a time

    the rest of the

    course!

    subject to

    functions

    linear non-linear discontinuous

  • 7/29/2019 Design Search and Optimization_mynotes

    7/59

    Generally, optimization algorithms can be divided in two basic classes: deterministic and

    probabilistic algorithms. Deterministic algorithms (see also Definition 30.11 on page 550) are

    most often used if a clear relation between the characteristics of the possible solutions and

    their utility for a given problem exists. Then, the search space can efficiently be explored

    using for example a divide and conquer scheme. If the relation between a solution candidateand its fitness are not so obvious or too complicated, or the dimensionality of the searchspace is very high, it becomes harder to solve a problem deterministically.

  • 7/29/2019 Design Search and Optimization_mynotes

    8/59

    Brief History of Optimization

    Newton and classical gradients

    144,42 2 xxdx

    dyxxy

    42

    2

    dx

    yd 2min y

    Pattern search and which way is down methods

    Stochastic searchpopulation

    DoE based search and RSMS - HYBRIDS

    Nonlinear numerical optimization methods are a very large field, but may be grouped into

    three main themes:1. The classical gradient-based methods and hill-climbers2. the evolutionary approaches3. The adaptation of Design of Experiment and Response Surface methods

    First came classical calculus and Newtons method for dealing with functions where we

    cannot solve explicitly. Calculus we are familiar with 0)(' xf and VExf )('' for a

    minimum etc. Newton is basically root searching for 0)(' xf and is covered in a

    subsequent lecture.

    Next come Cauchy and steepest descentie find the downhill direction and move in that

    direction until we start to go uphill againslow in valleys:-

  • 7/29/2019 Design Search and Optimization_mynotes

    9/59

    Classical hill-climbers all seek to move uphill (or downhill) from some given starting point in

    the direction of improved designs. Sometimes, they use simple heuristics to do this (as in the

    Hooke and Jeeves search (Hooke and Jeeves) and sometimes they use the local gradient

    (gradients obtained from finite differencing) to establish the best direction to move in (as in

    quadratic programming).

    Then came conjugate gradient methods and quasi-Newton methods that exploit local

    curve fitting based on a curvature. Various ways of holding information on the local shape.

    (The Hessian, or its inverse) ie approximation based on CxbAxxxfTT

    21)( where x is

    a vector and A is the Hessian.

    Following these gradient based approaches were a series of Pattern searches which use

    Heuristics. These include Hook & Jeeves and the simplex method (see plots of searches).

    Then a series ofstochastic methods including SA, GA, ES and EP all using sequences of

    random moves and schemes to exploit any gains made. Often working with population of

    designs. The main benefit is that such methods can avoid becoming stuck in local basins of

    attraction and can thus explore more globally. The downside is that they are often slower to

    converge to any optima they do identify.

    Finally come the explicit curve fitting methods based on designed experiments such as

    polynomial curve fits, RBF schemes, Kriging etc. These can be either global with updatesor local with move limits and trust regions.

    Design of Experiment and Response Surface methods are not really optimizersper se: rather,

    they are techniques that allow complex and computationally expensive optimization problems

    to be transformed into simpler tasks that can be tackled by gradient and evolutionary

    methods. Essentially, these are curve fitting techniques that allow the designer to replace

    calls to an expensive analysis code by calls to a curve fit that aims to mimic the real code.

    Finally hybrids and meta searches built on these elements.

    Key considerations are:

    Staircase effect etc

    Can be used for analytical forms

    or numerically.

  • 7/29/2019 Design Search and Optimization_mynotes

    10/59

    Ability to work with black box codes

    Need for gradient information

    Robustness to poor, noisy or incomplete data or badly shaped functions

    Speed of convergence Ability to run calculations in parallel

    Repeatability/average performance of stochastic methods

    The Place of Optimization in Design Commercial Tools

  • 7/29/2019 Design Search and Optimization_mynotes

    11/59

    GEOMETRY MODELLING & DESIGN PARAMETERIZATION-The Role of Parameterization in Design

    -Discrete and Domain Element Parameterizations

    -NACA Airfoils

    -Spline Based Approaches

    -Partial Differential Equation and Other Analytical Approaches

    -Basis Function Representation

    -Morphing

    -Shape Grammars

    -Mesh Based Evolutionary Encodings

    -CAD Tools v's Dedicated Parameterization Methods

    The purpose of design parameterization is to define geometric parameters, as well as to

    collect a subset of these parameters as design variables that can vary during the design

    process. During the design process, the design engineer will vary the design variables in order

    to improve structural performances.

    The need for parameterization design variables/design intent the things we are free to

    change.

    Flexibility versus no of variables & tacit knowledge of workable designs (SECT 2.1 of

    book). Some exampleslook at the Nowacki beam & ask what are design choices? How do

    we encapsulate (capture) them? What about choice of section? (Want a section that gives a

    high second moment area). How do we parameterise a cross section? How many variables do

    we need.

    1) a circle - just need radius1 variable

    2) a squarejust need side length1 variable

    3) /4) an elipse or rectangle2 variables

    5) an I section symmetric about N/A (neutral axis) or a box L or T sections in all cases

    we need overall width and depth plus thickness of web and flange4 variables

    Can we make a single parameterisation span square, rectangle, box, L, T and I? If so how

    do we use an integer variable as an index plus 4 numbers or can we use the 4 numbers

    themselves? Clearly square/rectangle/box can be linked:-

  • 7/29/2019 Design Search and Optimization_mynotes

    12/59

    Not obvious how to deal with L, T or I together. If use an index then there is no obvious

    ranking so search not simple. We can of course do

    As the solids are when tf= depth/2

    tw= width/2

    or or

    As continuous sets Simplest combined form is

    for etc

  • 7/29/2019 Design Search and Optimization_mynotes

    13/59

    LECTURE 2

    To make this generate all shapes we consider 4 rectangles

    We make the widths of :

    T and B to equal width overall

    L and R to equal webt

    Height of T equal toflangetop

    t

    Height of B equal to flangebott

    Height of L and R equal to height flangetopt flangebott and then add offsets of L and R from the

    outer edges as offsetl or offsetr .

    This needs seven variables but can now describe all our shapes as to get L or T we just set

    fangebottomt to be zero.

    Talk through the various figures in sect 2.1 and see PPT of External shape of UAV

    Internal structure of UAV

    Evolutionary optimisation of beams

    T

    L R

    B

  • 7/29/2019 Design Search and Optimization_mynotes

    14/59

    DEMO 1Excel & Nowacki Beam

    How do we set up searches in Excel?

    Try 142 xx (min at 2x )

    4x 42 3x (min at211x )

    Then look at Nowacki beam problem.

  • 7/29/2019 Design Search and Optimization_mynotes

    15/59

    SINGLE VARIABLE OPTIMISERS, LINE SEARCH: - an optimal solution to a problemwhere there is just one real-valued variable are called line search problems

    IntroductionOptimization methods may be classified according to the types of problems they are designed to deal

    with. Problems can be classified according to the nature of the variables the designer is able to control(nonnumeric, discrete, real valued), the number and type of objectives (one, many, deterministic,probabilistic, static, dynamic), the presence of constraints (none, variable bounds, inequalities,

    equalities) and the types of functional relationships involved (linear, nonlinear, discontinuous).Moreover, optimizers can be sub classified according to their general form of approach.They can be gradient based, rule based (heuristic) or stochasticoften, they will be a subtle blend ofall three. They can work with a single design point at a time or try to advance a group or populationof designs. They can be aiming to meet a single objective or to find a set of designs that togethersatisfy multiple objectives. In all cases, however, optimization may be defined as the search for a

    set of inputs x that minimize (or maximize) the outputs of a function f (x)subject to constraints

    gi (x) 0 and hj (x)= 0.

    In essence, this involves trying to identify those regions of the design space (sets of designer chosen

    variables) that give the best performance and then in accurately finding the minimum in any givenregion. If one thinks of trying to find the lowest point in a geographical landscape, then this amountsto identifying different river basins(sinks) and then tracking down to the river in the valley bottom andthen on to its final destinationin optimization texts, it is common to refer to such regions as basinsof attraction since, if one dropped a ball into such a region, the action of gravity would lead it to thesame point wherever in the basin it started. The search for the best basins is here termed global

    optimization, while accurately locating the lowest point within a single basin is referred to aslocal optimization. Without wishing to stretch the analogy too far, constraints then act onthelandscape rather like national borders that must not be crossed. It will be obvious thatmany searcheswill thus end where the river crosses the border and not at the lowest pointin the basin. Of course,in design, we are often dealing with many more than two variables,but the analogy holds and we canstill define the idea of a basin of attraction in this way.

    The aim of optimization is to try and accomplish such searches as quickly as possible in as robust afashion as possible, while essentially blindfoldedif we had a map of the design space showingcontours, we would not need to search at all, of course.Notice that even when dealing with multiple design goals it is usually possible to specify a single

    objective in this way, either by weighting together the separate goals in some way, or by transformingthe problem into the search for a Pareto optimal set where the goal is then formally the production ofan appropriately spanning Pareto set (usually one that contains designs that represent a wide andevenly spread set of alternatives). It is also the case that suitable goals can often be specified bysetting a target performance for some quantity and then using an optimizer to minimize any deviations

    from the target. This is often termed inverse design and has been applied with some success to airfoiloptimization where target pressure distributions are used to drive the optimization process.

  • 7/29/2019 Design Search and Optimization_mynotes

    16/59

    LECTURE 3 CONT

    Problemminimize a problem of a single real variable without constraints.

    Approach 1given a functional form use calculus, ie

    14)( 2 xxxf

    42)( xxf

    2)( xf , ie VE

    Approach 2Bracket the minimum between two values and search inwards.

    QHow do we find a bracket, ie a series of three values of )(xf such that )()( 12 xfxf

    and ?)()( 32 xfxf

    A Guess two values for x, calculate )(xf at these and then head downhill until the

    function starts to rise and we have a bracket. If we have no knowledge then use

    1,0 21 xx and 3x either -1 or 2 depending on the gradient (if )1()0( ff then 23 x

    else -1). Given three points we use a quadratic curve fit and see if a minimum is

    predicted (2nd diff is +VE) and if so jump to the minimum predicted and evaluate there.

    If a maximum is predicted we simply increase the step size (by say a factor of 1.6180

    golden section) and go on downhill, keeping the three lowest values of )(xf in either

    case.

    See code in Numerical Recipes for example. Another approach is just to keep doubling

    the step size in the downhill direction until a bracket appears.

    QGiven an initial bracket how do we trap the minimum efficiently.

    There are a number of schemes for doing this and all aim to replace one or other (or both) of the outerbounds with ones closer to the optimum and thus produce a new, tighter bracketing triplet:

    o Golden Section Search can yield sufficient accuracyo Fibonacci Search Faster, requires number of iterations

    AApproach 3Golden section search (linear convergenceno use of gradients)

    min at

  • 7/29/2019 Design Search and Optimization_mynotes

    17/59

    Given 21, xx and 3x such that )()( 12 xfxf

    )3()( 2 xfxf

    We choose 4x so that it lies in the larger of the two intervals 1x to 2x and 2x to 3x and

    such that either23

    24

    xx

    xx

    = 0.38197 or

    12

    14

    xx

    xx

    = 1-0.38197 = 0.61803

    Even if the initial bracket is not in the ratio 0.38197:0.61803 this process rapidly settles

    on this ratio.

    Note that (3 5)/2(0.38197).

    Provided we start with a triplet where the internal point is 0.38197 times the triplet width fromone end, this will mean that whichever endpoint we give up when forming the new triplet, theremaining three points will still have this ratio of spacing. Moreover, the width of the triplet is

    reduced by the same amount whichever endpoint is removed, and thus the search progresses at auniform (linear) rateConsider points spaced out atx1 = 0.0,x2 = 0.38197 andx3 = 1.0. With these spacings,x4 mustbe placed at 0.61804 (= 0.38197 (1.0 0.38197) + 0.38197) and the resulting triplet will thenbe eitherx1 = 0.0,x2 = 0.38197 andx4 = 0.61804 orx2 = 0.38197,x4 = 0.61804 andx3 = 1.0

    This approach assumes nothing about the shape of the function and does not

    require gradientsit is not quite so good as Fibonacci but does not require us to fix the

    number of function calls apriori (which Fibonacci does).

    X1 X2 X4

    X3

    Ratio 0.38197:0.61803

    X1 X4 X2

    X3

    Ratio 0.61803: 0.38197

  • 7/29/2019 Design Search and Optimization_mynotes

    18/59

    QGiven an initial bracket and assume the function is smooth so that at its minimum it will

    behave quadratically.

    AApproach 4inverse parabolic interpolation, quadratic search or quadratic interpolation.

    Here we fit a parabola to the bracket and use this to estimate the location of the

    minimum.

    ie we assume ,)( 2 CBxAxxf so thatA

    Bx2*

    , Axf 2*)(

    and we know 2211 )(,)( fxffxf and 33)( fxf such that 3212 , ffff and

    321 xxx

    ie CBxAxf 12

    11

    CBxAxf 22

    22

    CBxAxf 32

    33

    We solve these to get

    )))(())(((2

    ))(())((

    *23211223

    2

    2

    2

    321

    2

    1

    2

    223

    xxffxxff

    xxffxxff

    x

    and

    ))(())((

    ))(())((

    23

    2

    1

    2

    212

    2

    2

    2

    3

    23211223

    xxxxxxxx

    xxffxxffA

    (which is always +VE so hence minimum)

    For example, consider the function

    42)( 34 xxxf with initial data at 2,1,2

    1

    x so that

    42,31,8125.32

    1321

    fff , ie a bracket.

    then

    214286.1

    )12)(38125.3()2

    11)(34(2

    )14)(38125.3()4

    11)(34(

    *

    x

    This may be compared to the analytical solution given from

    064)( 23 xxxf when0

    x

    or2

    11x . Note3125.2)5.1(

    f

  • 7/29/2019 Design Search and Optimization_mynotes

    19/59

    01212)( 2 xxxf at 0x inflexion

    =9 at211x minimum

    So the solution is improved

    ,5932164.2)214286.1( f ie less than all points in initial bracket.

    So next triple is 1, 1.214286,2

    With function values 3,2.5932164,4

    this gives

    317823.2,465064.1,385534.2,364454.1 *4*

    4

    *

    2

    *

    2 fxfx

    313928.2,482046.1,334451.2,427849.1 *5*

    5

    *

    3

    *

    3 fxfx

    OBSV 1.

    Now you may ask why not start with x=0,1,2. Trouble with this is f=4,3,4 which is

    symmetric so 1*1 x which does not help!!

    OBSV 2.

    This search approaches our goal from one side only and so is rather slow there are better

    methods!

    When dealing with discrete variables we use the integers as pointers and either use integer

    programming or in mixed problems simply round variables to the nearest integer.

    When our discretes have no natural order (ie materials selection) we in the end are forced

    towards enumeration (listing).

    Approach 5Newtons method

    All will be familiar with the Newton-Raphson method (gradient based scheme) for finding

    the root (or zero) of a function. If we apply this to the derivation of a function we can find

    turning points instead, ie

    )(

    )(1

    i

    i

    iixf

    xfxx

    NB this needs the second derivative.

    Example: 42)( 34 xxxf starting at 2x

  • 7/29/2019 Design Search and Optimization_mynotes

    20/59

    23 64)( xxxf

    )1(12)( xxxf

    )1(12

    6423

    1

    ii

    iiii

    xx

    xxxx

    = ix )1(6

    322

    i

    ii

    x

    xx

    6

    682)12(6

    64222 21

    xx

    = 6667.13

    21

    52777.14

    95

    3

    5

    )3

    2(6

    5)925(2

    3

    213

    x

    *50000.1,50097.1 54 xxx

  • 7/29/2019 Design Search and Optimization_mynotes

    21/59

    LECTURES 4&5 (LS110132)

    Multi variable optimizers

    Gradient Methods

    We next consider multiple variables. Here in addition to finding the size of step to make we

    must also fix the direction.

    Perhaps the simplest approach to multi variable optimizing is to identify the direction of

    steepest descent (Cauchy) and go in that direction until the function stops reducing (the

    optimal step length*

    i ) and then recomputed the direction of steepest descent, ie

    )(*1 iiii xfxx where )( ixf are the gradients at ix

    Exampleminimize 22212

    12121 22),( xxxxxxxxf

    starting at 1x =

    0

    0, 0)0,0( f

    1

    1)(

    221

    241

    /

    /1

    21

    21

    2

    1xf

    xx

    xx

    xf

    xff -this the direction vector

    To get the optimal step length (how far you in a particular direction) we minimize

    ))(( 111 xfxf with respect to 1 , ie set 0/ 1 ddf

    2

    1

    2

    1

    2

    1111

    1

    1111 22)1

    1

    0

    0

    ())((

    ffxfxf

    = 12

    1 2

    122*

    11

    1

    d

    df

  • 7/29/2019 Design Search and Optimization_mynotes

    22/59

    so that

    1

    1

    1

    110

    02x , 1)1()1)(1(2)1(211)1,1(

    22 f (f is getting

    smaller)

    now

    1

    1

    221

    241)( 2xf

    1

    1)1

    1

    1

    1())((

    2

    2

    2222

    ffxfxf

    = 22222

    222 )1()1)(1(2)1(2)1()1(

    = 22222422 22

    2

    2

    22

    2

    2

    = 425 22

    2

    51210 *222

    d

    df

    2.1

    8.0

    1

    151

    1

    13x , 2.1)2.1,8.0( f , and so on to

    5.1

    1as the answer.

  • 7/29/2019 Design Search and Optimization_mynotes

    23/59

    LECTURES 4&5 continued

    CONJUGATE GRADIENT

    The problem with steepest descent is that unless our function has circular contours the

    direction of steepest descent never points at the final optimum. The conjugate gradient

    approach seeks to improve over this with improved directions.

    We start as per the steepest descent but at the second step use a direction conjugate to the

    first, ie

    )( 1*

    112 xfxx but afterwards use iiii Sxx*

    1

    NB 0jT

    i ASS if iS and jS are conjugate directions for a quadratic problem of the form

    CxBAxxxf TT 21)(

    Nota Bene -Abbreviation. NB: Used to direct attention to something particularly important.

    where 121

    2

    i

    i

    i

    ii Sf

    ffS

    and )( 11 xfS

    Here iS takes the place of if used in steepest descent. Notice that iS accumulates

    information from all previous steps this is good and bad good as the direction is

    conjugate to all previous steps, bad as it can accumulate round off errors in practice

    we restart from a steepest descent step after m steps where m is one more than the

    number of design variables. If our function is quadratic this process converges in as

    many steps as directions/dimensions in the problem.

    Example: minimize 22212

    12121 22),( xxxxxxxxf starting from

    0

    01x

    Conjugate gradient search example.

  • 7/29/2019 Design Search and Optimization_mynotes

    24/59

    Starting at

    0

    0and apply steepest descent gives as before

    1

    12x and 1

    *

    1 and 21

    1

    )(2

    1

    fxf

    So

    1

    11S , ( )( 11 xfS ) 2

    1

    1)(

    2

    22

    fxf

    2

    0

    1

    1

    2

    2

    1

    12S

    2

    0

    1

    1*

    23 x

    To find *2 we minimise )2

    0

    1

    1()( 2222

    fSxf

    12

    1

    2f

    2

    222 )21()21(22)21(1

    124 22

    2

    So4128

    *

    22

    2

    d

    df

    5.1

    1

    2

    0

    41

    1

    13x which is the solution

    If we try another step we just find2

    3f is zero and so the process stops, ie 3f is zero at the

    minimum.

  • 7/29/2019 Design Search and Optimization_mynotes

    25/59

    LECTURES 4& 5 continued

    NEWTONs METHOD

    Newtons method allows for direction and step size and is built on looking for the roots of

    ).(xf

    First we approximate our function as a Taylor series.

    )()(21)()()( 2 ii

    T

    ii

    T

    i xxHxxxxfxfxf

    Where here iH is the matrix of second partial derivatives and is called the Hessian.

    Now we set 0)(

    jx

    xfj=1,2,..n for n variables

    So this gives 0)( iii xxHff

    or iiii fHxx

    1

    1this requires a non singular Hessian of course. (So the Hessian both

    modifies the search direction and sets the step length).

    Exampleminimize 22212

    12121 22),( xxxxxxxxf starting at

    0

    01x

    1

    2

    2

    2

    12

    221

    2

    2

    1

    2

    1

    xx

    f

    xx

    f

    xx

    f

    x

    f

    H

    =

    22

    24for all ix

    411

    1

    H

    42

    22=

    12121

    21

    ix

    ixf

    xff

    2

    1

    /

    /=

    1

    1

    221

    241

    21

    21

    ixxx

    xx

  • 7/29/2019 Design Search and Optimization_mynotes

    26/59

    So

    1

    2121

    21

    0

    02x

    23

    1

    1

    1

    1

    12f

    2

    4

    0

    0

    3

    3

    Hence MINIMUM in 1 STEP

    This has converged in one step because )(xf is quadratic and so H is a constant. There are

    however problems with this approach as we have to compute, invert and store H at each

    step and this is fraught with difficulties on real problems. The most serious issue is obtaining

    the second derivatives as these are very rarely available directly.

    The Quasi Newton methods work with an approximation of either the Hessian or its inverse.

    These are sometimes called variable metric methods

    We already have )(11 iiii xfHxx

    iH is the Hessian

    Which we approximate by

    )(* iiiii xfBxx

    Here iB contains directional information and*

    i the optimal step length. Note that this is the

    steepest descent method if iB = I

    There are then a number of schemes for updating iB without using second derivatives but

    instead using approximations. None is perfect and they are known by the names of those

    who proposed them such as BFGSBroyden Fletcher Goldfarb-Shanno, which is

    iT

    i

    i

    T

    ii

    i

    T

    i

    T

    iii

    i

    T

    i

    ii

    T

    i

    i

    T

    i

    T

    iiiigd

    Bgd

    gd

    dgB

    gd

    gBg

    gd

    ddBB )1(1

    where iii xxd 1

    iii ffg 1

    We do not pursue such methods further here, they are very popular however.

    See Figs 3.6,3.7 in book.

  • 7/29/2019 Design Search and Optimization_mynotes

    27/59

    Noisy/Approximate Function Values- Gradient-based searches can rapidly identify optima in a fitness landscape with high precision.- Gradient-based methods only seek to follow the local gradient to the nearest optimum in thefunction, and so, if they are to be used where there are multiple optima, they must be restarted fromseveral different starting points to try and find these alternative solutions.- One great weakness in gradient-based searches, however, is their ratherpoor ability to deal withany noise in the objective function landscape.

  • 7/29/2019 Design Search and Optimization_mynotes

    28/59

    Non gradient based search methods- establishing the gradient of the objective function in

    all directions during a search can prove difficult, time consuming and sensitive to noise, a variety ofsearch methods have been developed that work without knowledge of gradients. The most commonmethods are pattern/direct searches and stochastic/evolutionary algorithms

    -Pattern or direct search: Hooke and Jeeves, and Nelder and Mead

    Advantages:

    Simple

    Robust (relatively)

    No gradients necessary.

    Disadvantages:

    Can get stuck and special "tricks" are needed to get the search going again.

    May take a lot of calculations

    Nonlinear multi-objective multiplex implementations using pattern search

    algor ithms have been made.

    What do we do if we cannot calculate gradients (or do not wish to use finite differences

    noise/speed).

    This leads to Hook & Jeeves amongst others.

    H&J method:

    1 choose initial step length, set initial point to first base point

    2 increase direction i by step and keep if better else decrease direction i by step and

    keep if better

  • 7/29/2019 Design Search and Optimization_mynotes

    29/59

    3 loop over all directions, if none improve then half step size and repeat unless

    either step too small or run out of time in which case stop

    4 explore must have helped so set current point to new base point

    5 make pattern move equal to vector from previous base point to new base point

    plus any previous successful pattern move still in use

    6 if pattern move helps keep it if not go back to new base point and forget pattern

    move.

    7 repeat from step two

    There are several themes here.

    1 steps change in size for exploration

    2 Directions and steps change for exploitation if the pattern moves help then they

    accumulate so that moves get bigger and bolder until they fail. Siddall provides

    full details and code, as does Schwefel.

    Note that the pattern moves get larger at each exploratory search pattern and slowly alignsitself to the most profitable direction of move. This allows the search to make larger and

  • 7/29/2019 Design Search and Optimization_mynotes

    30/59

    larger steps. Conversely, once a pattern move fails, the search resets and has to start

    accumulating pattern moves from scratch

  • 7/29/2019 Design Search and Optimization_mynotes

    31/59

    -Stochastic/Evolutionary search

    In GA, every run yields a different answer, so need to take averages.

    Run through flying circus slides on simple GAs.

    Use web animation to demo a GA on the bump problem.

    Typical search patterns from a GA, Simulated Annealing, an Evolution Strategy and

    Evolutionary Programming.

  • 7/29/2019 Design Search and Optimization_mynotes

    32/59

  • 7/29/2019 Design Search and Optimization_mynotes

    33/59

    Termination/Convergence

    For local searches we stop at the optimum, ie when no further gains are being made-provided

    we can afford to get that far.

  • 7/29/2019 Design Search and Optimization_mynotes

    34/59

    For global search we use one of

    1 a fixed or limited number of iterations

    2 a fixed or limited elapsed time

    3 when the search has stalled after a given number of iterations

    4 when a given number of basins have been found and searched.

    We rank searches by steepness of gradient/rate of improvement, final result or a balance

    between the two.

  • 7/29/2019 Design Search and Optimization_mynotes

    35/59

    LECTURES 6&7

    CONSTRAINED OPTIMIZATION

    In most real world engineering problems the designer has to satisfy various constraints as

    well as meeting the desire for improved performance. Indeed performance is often set as a

    constraint, ie reduce weight to below x, reduce drag to less than Y etc. Thus we need search

    schemes to deal with constraints, ie

    forxf )(min

    nx

    x

    x

    1

    subject to bounds on ,x xxxL

    and constraints 0)( xgi (inequality constraints)

    0)( xhj (equality constraints)

    Here we describe a number of approaches.

    The simplest is to try and eliminate constraints by construction ie transform problem

    variables using the constraints.

    Example: minimise the surface area of a box of given volume.

    ie min )(2),,( HWWBBHWBHf , whereL,B andH, are the length of the sides

    and WBHv is fixed

    So letBH

    vW

    we have

    Minimize )(2),,( HBBH

    VBHVBHf

    B

    V

    H

    VHB

    222

    So 02

    22

    H

    VB

    H

    fwhen

    B

    VH

    B

    VH 2, or B= 2H

    V

  • 7/29/2019 Design Search and Optimization_mynotes

    36/59

    02

    22

    B

    VH

    B

    fwhen

    HVB or

    HVB 2

    Combining gives

    3

    4

    2

    VHH

    V

    H

    V

    3 VB (Substitute 3 VH into B= 2HV )

    3 VW (Substitute 3 VH & 3 VB intoBH

    vW

    )

    ie all sides of equal length is expected.

    Another way we deal with inequalities is by deciding if they will be active at the

    optimum or not. If so we replace by equality and if not we eliminate them. So if

    inequalities are active then we replace with equality sign.

    Often it is not possible to know which inequalities will be active or to eliminate using algebra

    even if we do! Nonetheless we should not ignore this. It can be done numerically sometimes,

    ie fixed LC calcs when angle of attack is a design variable for a wing or aerofoil.

    GEOM & ALPHA

    CFDOPTIMISE

    with CL= fixed

    CL,CD

    GEOM

    OPTIMISE Iterateon

    alpha

    CFD

    CL

    OR

    CD

  • 7/29/2019 Design Search and Optimization_mynotes

    37/59

    LAGRANGE MULTIPLIERS

    In just the same way as there are formal analytic solutions to unconstrained optimization

    problems the equivalent constrained solutions are based on Lagrange multipliers. This

    approach essentially only works for equality constraints so for inequality constraints a

    precursor step is to decide at any point if an inequality constant will be active, and if so

    replace it with an equality.

    So consider min ),( 21 xxf subject to 0),( 21 xxg

    ie two variables and one equality constraint.

    At a minimum it may be shown that

    011

    x

    g

    x

    f

    and 022

    x

    g

    x

    f

    and 0),( 21 xxg

    Here is the so called Lagrange Multiplier.

    Now if we write gfL we get

    111 x

    g

    x

    f

    x

    L

    all equal zero at the minimum

    222 x

    g

    x

    f

    x

    L

    from the previous equations

    gL

  • 7/29/2019 Design Search and Optimization_mynotes

    38/59

    Thus if we seek the unconstrained minimun of L (more precisely, turning points of L) we can

    locate the solution to the constrained problem. L is known as the Lagrange function.

    For example minimize ),( yxf2xy

    k

    Subject to 0),( 222 ayxyxg (ie circle of radius a)

    Here )(),,( 2222

    ayxxy

    kgfyxL

    0222

    xykx

    x

    L

    423

    22

    xy

    k

    yx

    k

    2

    yx

    0222

    ayxg

    L

    ,

    3

    20

    2

    222

    ayayy

    Here yax ,3 3

    2a

    Note however that we cannot simply minimize L as the approach would admit of

    saddlepoints or maxima for the gradients of L to be zero.

    PENALTY FUNCTION METHODS

    A more direct approach to dealing with constraints is via the use of penalty functions

    we simply add penalties to the objective function when constraints are violated. There

    are a number of ways of doing this, none of which is perfect:-

    FIXED PENALTIES

    Add a (very) large number to the objective if any constraint is broken

    Add a (very) large number for each broken constraint

    022 31

    yykx

    y

    L

  • 7/29/2019 Design Search and Optimization_mynotes

    39/59

    VARYING PENALTIES

    FUNCTION OF DEGREE OF CONSTRAINT VIOLATION

    Scales the penalties by the constraint violation

    FUNCTION OF HOW LONG WE HAVE BEEN SEARCHING

    Start with low penalties and gradually make more severe so that an essentially

    unconstrained search becomes a fully constrained one

    All these are taken to be exterior penalties, ie they only apply to broken constraints we can

    also use interior penalties which come into effect as the search nears a constraint and then

    gradually remove these as we progress so as to warn the search about nearby problems.

    Sketch Penalty Types

    COMBINED LAGRANGE + PENALTY FN METHOD

    Since many penalty functions introduce discontinuities which make search with efficient

    gradient descent methods difficult, a potentially attractive approach is to combine a penalty withthe method of Lagrange multipliers.

    It is possible to combine the Lagrange scheme with a penalty approach to overcome some of

    the difficulties of pure Lagrange methods. This is sometimes called the Augmented

    Lagrange Multiplier method.

    ie minimize )(xf subject to 0)( xhj pj .. .2,1

    OF

    STEP

    INTERNAL EXTERNALX2

    X1

    OF OF

    X1 X1 X1

  • 7/29/2019 Design Search and Optimization_mynotes

    40/59

    )()(),(1

    xhxfxL j

    P

    j

    j

    is the Lagrangian

    We augment this with an exterior penalty

    )()()(),,(1

    2

    1

    xhrxhxfrxAP

    d

    jkj

    P

    j

    jk

    This model is therefore amenable to search by gradient-based methods provided the originalobjective and constraint functions are smooth.

    It now turns out that minimizing A solves the original problem if we have the correct

    for any kr . Note that kr is some, but not infinite penalty. However we can apply an iterative

    scheme now that will allowj

    andk

    r to converge on a solution providedkk

    rr 1

    and we use

    kjbkjkj xhr *)()1( 2 , note that k is the major iteration counter.

    ie the new s' are added to by the (scaled) amount of violation of the constraints at the

    previous minimum of A

    This approach can also be extended to inequality constraints by setting up as follows; min

    )(xf subject to

    mixgpjxh ij ...1,0),(,...1,0)(

    )()(),,(11

    xhxfrxA j

    p

    j

    jm

    m

    i

    ii

    p

    d

    jk

    m

    i

    ik xhrr1

    )(2

    1

    2

    where i max

    k

    ii rxg2

    ),(

  • 7/29/2019 Design Search and Optimization_mynotes

    41/59

    Sequential Quadratic Programming SQP

    The use of sophisticated Lagrangian processes is now at its most complex and powerful in the

    class of methods known as SQPthese use typically Quasi Newton methods to solve a series

    of sub problems. They are the most powerful methods available for local minimization of

    constrained smooth problems. Academic codes are available from the web. They are less

    good for non-smooth functions and also they are local methods and so cannot find the best

    basin of attraction to search.

    (Chromosone) Repair

    Repair is the process of dealing with a constrained optimization problem by substituting

    feasible designs whenever infeasible ones occur during search. To do this a repair process is

    invoked if any constraint is violated to find the nearest feasible design. Here nearness is

    usually in the Euclidean sense of design variables. Having located such a design (perhaps by

    a local search where the degree of infeasibility is set as a revised objective) the objective

    function of the feasible design is used instead of that at the infeasible point and also

    (optionally) the corrected design vector.

    Replacing the design vector absorbs most information but can cause problems with the search

    engine. This approach is most favoured in evolutionary or other zeroth order (non-gradient)

    methods where gradients are not used at all.

  • 7/29/2019 Design Search and Optimization_mynotes

    42/59

    LECTURE 8

    META-MODELS + RSM

    So far we have considered optimizers working with results coming from the evaluations of

    design & constraint functions that have been presumed to be directly coupled to search codes.

    These codes then build up a picture of how the function is changing with changes in the

    design and seek improvements. Their internal models (we will call them meta-models to

    distinguish from the actual user supplied design models) are implicit in their working.

    We next consider schemes where the building & use of the meta model is explicit and

    directly controlled by the user.

    At its simplest this consists of running a few designs, collecting the results and curve fitting

    to these. Then the curve fits can be used for design search. This would be a natural approach

    for working with data from previous designs or from experiments or field trials it can also

    be used with computer analysis codes.

    We first plan where to run the code to generate data. This can aim to build either a local or a

    global model depending on the range of the design variables. We use formal DoE (Design ofExperiment) methods for this (cf Taguchii). Having run the design points, often in parallel,

    we curve fit. Here again we decide if we need a local (simple) fit or a global (complex) shape

    & also if we need to regress (discuss noisy data). Curve fitting can be fast for simpler models

    or very slow for large accurate ones.

    We call the curve fit a Response Surface Model (RSM) or meta model. Examples include

    Polynomial regression, radial basis functions and kriging and neural nets.

    Having built a model we check its accuracy with test data (separate) or cross validation. We

    then use it to search for a better design. Having found new candidate designs we run the full

    computer code to check if they are good. If so we might stop. More usually we add these to

    the curve fit & iterateupdating, until we run out of effect or we get convergence etc.

    Give examples. Discuss exploit vs explore.

    To summarise: the basic steps are

  • 7/29/2019 Design Search and Optimization_mynotes

    43/59

    1) plan an experiment to sample the design space

    2) run codes & build data-base of results (possibly in parallel)

    3) choose & apply a curve fit, with or without regression

    3a) refine curve fit by some tuning process

    4) predict new interesting design points by searching the meta-model

    5) run codes on new point(s) & update data-base (again possibly in parallel)

    6) check the results from update points against predictions & then either stop or

    move back to step 3)

    Experience shows that for model building it can often take 10n initial designs to build a

    reliable global model where n is the number of variables. There is also a trade between the

    cost of building a meta-model and the usefulness of its predictions. (Show PPT on various

    DoE designs & various RSM types)

    Example 1a local trust region search

    A very simple approach is to evaluate a small local experiments and then shift and shrink it

    until a certain effort is used up.

    Step one) choose initial area to search

    Step two) sprinkle in 9 points useLPDoE

    Step three) curve fit with quadratic regression polynomial

    Step four) search within area over RSM to get new candidate design

    Step five) shift search region centre to new candidate point

    to solve the regression we used SVD to get a least squares solution to the over constrained non square matrix

    equation bay where y are the function values, a are the design variable values and their powers

    and b are the polynomial coefficients. So if the SVD of a is '.. VwU then it may be shown that

    }{')]./1(diag.[}{ yUwVb j , see matlab for simple examples of SVD.

  • 7/29/2019 Design Search and Optimization_mynotes

    44/59

    Step six) shrink search region by say 10%

    Step seven) replace oldest design point with new result

    Step eight) go to step three unless run out of time

    Now this simple scheme has a number of faults:

    1) it has no way of expanding the trust region if the data suggests it should be this

    means it may be a), slow and b), fail to find a local minimum of the function.

    2) The point being replaced, the oldest, may not be the most sensible one to discard

    what about discarding the worst point for example.

    Example 2a global RSM search

    1) Here we first use 100 points in an LP array to sample the design space.

    2) Then we construct a krig (stochastic or Gaussian process) RSM which has hyper

    parameters which we tune.

  • 7/29/2019 Design Search and Optimization_mynotes

    45/59

    3) We then search for peaks and return 10 likely locations

    4) We add these to the original 100 pts to get 110 and we rebuild and retune the krig.

    5) We then return to step 3 and repeat 3 times, ending with 130 points and the finalmodel.

  • 7/29/2019 Design Search and Optimization_mynotes

    46/59

  • 7/29/2019 Design Search and Optimization_mynotes

    47/59

    Points to note

    1) the use of a large initial DoE is warranted here because of the multi-modality of

    the problem (the 20 points we might otherwise use).

    2) the updates are added in groups of 10 because we wish to improve the model

    globally and not just in one location, also krig training is costly.

    3) the final surface is reasonable but still far from exact.

    Meta Heuristics

    It is clear from the two previous searches that what we have done is combined components

    such as DoE sampling, RSM building and various searches to build a composite or meta-

    search. It is of course possible to build more and bigger complexities into such approaches

    and this leads to a whole family of meta heuristics

    By way of example we can consider combing various gradient descent schemes & then

    observing which works best and rewarding this with more of our finite budget of compute

    resource. The development of such schemes is an art and also one must bear in mind that the

    no free lunch theorem show that, averaged over all possible problems, all searches are as

  • 7/29/2019 Design Search and Optimization_mynotes

    48/59

    good or bad as each otherso unless our efforts are based on tuning a method to the current

    task they will be futilemoreover there will be a trade between performance on a specialised

    task and general applicability.

    Visualizationshow & describe HAT plots and parallel axis plots.

    LECTURE 9

    Multi-Objective Optimization

    So far we have focussed on problems with a single goal or objective function. This is rarely

    how real design problems occur, although it is quite common, even in industry, to treat them

    this way. In reality most real design problems involve trading between multiple andconflicting goals. We therefore next turn to ways of tackling such problems.

    Perhaps the simplest approach (and that most commonly used in industry) is to set all goals

    except one in the form of constraints, ie instead of aiming for low weight or stress we set

    upper limits on these goals and then ensure that our designs meet them. The difficulty with

    this approach is deciding realistic but demanding targets: if they are two severe we may not

    be able to satisfy them at all, if too loose they may not impact the design at all.

    The next most simple way of proceeding is to use an aggregate or combined objective.

    Typically we add all our goals together with some suitable weighting functions and minimize

    this. This approach is a mimic of the function of money money is societys way of

    allowing completely different things to be balanced against each other (the cost of a holiday v

    a new car for example). It is the function of markets to establish the prices of items and

    hence the weighting between them. Ideally the best approach to balancing competing goals

    to a business engaged in design is to reduce all decisions to their impact on the companys

    profits. Unfortunately this calculation is almost never possible so some surrogate is used.

    This may be completely artificial or it may be some physical quantity such as SFC (aero

    engine makers often use SFC).

    It should be clear that if we consider two goals (say )(1 xf and ))(2 xf then depending on how

    we weight them

    )()()( 21 xBfxAfxf

  • 7/29/2019 Design Search and Optimization_mynotes

    49/59

    then our final optimum will vary. For example consider

    min 21 )( xxf

    min xxxf 4)( 22

    then BxxBABxBxAxxf 4)(4)( 222

    now we only really need one weight hereA

    BC so we get

    cxxcAxf 4)1()( 2

    04)1(2)( cxcAxf when)1(2

    4

    c

    cx

    So for each value of C we get a different solution and two different values of 1fand 2f :

    Methods for combining goal functions

    It will be clear from considering Pareto fronts and simple weighted sums of goals that

    deciding how to combine goals will define the final designs selected. Before looking at more

    advanced schemes for finding the front we briefly explore how a design team might decide

    on a combined objective, either to reduce the problem to a single goal or to allow selections

    to be made from those designs that are found to lie on the Pareto front. All such methods

    attempt to formalise the process of assigning importance to the goals under review.

    C 1x )(1 xf xf2

    0 0 0 0

    1 1 1 -3

    23

    119

    71953

    21 32

    94

    922

    Good

    designsPARETO FRONT

  • 7/29/2019 Design Search and Optimization_mynotes

    50/59

    a) simple voting schemeseach design team member ranks the goals, the goals are then

    given points from 1(lowest rank) to n (highest rank in n goals) and then these are

    summed across the team to result in a weighting scheme. Before application all goals

    are divided by the designers ideal value (or the ideal is subtracted from the goal) to

    allow for the units in use. This simply ensures that all voices are heard.

    b) The eigenvector method. Each pair of goals is ranked on a matrix by being given a

    preference ratio, ie if goal i is three times more important than goal j we set 3ijp .

    Then say goal j is twice as important as goal k we set 2jkp etc. To be consistent

    we should of course say that 6ikp but in fact the method does not require this. In

    any case we then form all the p values into the matrix P and seek W so that

    wPw max , ie the eigenvectors of p are found.

    We then take the eigenvector with the largest eigen value and use this as our weighting

    scheme.

    If we have

    12161

    2131

    631

    p

    We get w=

    111.0

    222.0

    667.0

    The largest eigen value should equal the number of goals if the ps are consistent as

    here (we get a value of 3).

    Say however we were not consistent and used

    12151

    2131

    631

    p

    Then we get a largest eigen value of 3.004 and the weight vector becomes

  • 7/29/2019 Design Search and Optimization_mynotes

    51/59

    122.0

    230.0

    648.0

    w

    That is we decrease the importance of goal 1 and increase that of goals 2 and 3.

    This scheme is simple and easy to use up to around 10 goals at most and is quite useful

    for more than 4 goals where it is difficult to assign numerical values to the weights in a

    consistent fashion. It is still the case however that the aggregate goal is a simple linear

    sum of the individual functions.

    One way of combining goals that is more elaborate is via the use of fuzzy logic. Thus

    we define a series of linguistic terms that describe our goal and map these to a score:-

    Thus a bad value scores nothing and a good value scores 1 while those in the indifferent

    range have intermediate scores. If we do this to all functions the resulting scores can then be

    combined by adding (essentially an average) or multiplication (a geometric average). We

    then maximize the combined function. Various shapes for the so called membership

    functions can be used but there seems little to be gained from going above the linear form

    sketched here.

    Score

    1

    0

    BAD INDIFFERENT GOOD

  • 7/29/2019 Design Search and Optimization_mynotes

    52/59

    When used these functions essentially allow non-linear combinations of goal functions which

    clearly allow more complex combined goalshowever if taken too far they can obscure the

    overall problem!

    Example of Fuzzy Logic

  • 7/29/2019 Design Search and Optimization_mynotes

    53/59

    Methods for finding Pareto sets

    Sometimes we do not wish to combine goals without first examining the Pareto front itself.

    Thus we need to construct the front. We then have 3 goals.

    1) the designs we study truly lie on the front, ie they are well optimised

    2) the front has many designs that span its full extent, ie it is well populated

    3) the points are evenly spread on the front, ie we have smooth range of goals.

    This is in fact an optimization task in its own right and may be tabled in a variety of ways.

    A Perhaps the simplest is to construct a family of different combined goals with various

    weighting schemes and then optimize these (including dealing with each goal on its

    own). Although this does not tackle point 3 above it focuses on 1 and gives as many

    points as desired for 2. It is however expensive and known to fail to evenly populate

    the font, especially if the font is concave (using a linear sum of goals is equivalent to

    finding the intersection of a target line and the front and targets only exist for convex

    fronts).

    1

    a b

    xa xb

    f1

    1 0Score

    x

    Score

    1

    0

    f1

    f2

    x

    xa xc xf xp

    score= overall goal

    2

    1

    x

    xa xc xb xd

    AND SIMILAR FOR

  • 7/29/2019 Design Search and Optimization_mynotes

    54/59

    B The next best scheme is to use an optimizer to explore the design space placing any

    new non-dominated points in an archive (and weeding out any dominated ones). Then

    all new design points are given a goal value based on how much they improve the

    archive ie how dominant they are. This means that the objective function is non-

    stationary but provided our search is tolerant of this, this approach works fine (an

    evolution strategy works quite well on this).

    C Use a multi objective population based search (such as a GA). Here we aim to advance

    the whole front in one go so the points in the population are scored against each other

    and those that dominate most score mostthis is capable of meeting all three goals if

    some pressure to spread out points is included. Its weakest aspect is probably in

    finding the extreme ends of the front but these can be found from single objective

    searches directly on each goal.

    It is also possible to use response surface schemes to help reduce run times when finding

    Pareto fronts but this is not covered here.

  • 7/29/2019 Design Search and Optimization_mynotes

    55/59

    LECTURE 10

    ROBUSTNESS IN OPTIMIZATION

    When doing any kind of design work one must bear in mind that

    1) the design as made will differ from nominal

    2) the operating conditions may be unknown and differ from those used in design

    3) the product may change/deteriorate in service

    All these aspects are to some extent stochastic and thus not amenable to deterministic

    approaches to design. One way of dealing with this uncertainty is to try and make designs

    robust to such uncertainty, ie such that their performance changes little as the uncertain

    quantities move over their likely ranges.

    This kind of lack of robustness in design variables (type 1 above) in common when dealing

    with constrained - designers typically stay well away from critical stress limits if they can for

    example to avoid unexpected structural failures.

    f(x)

    vA

    vB

    x

    xB

    Design A is much more sensitive

    than B even though f(xA) is

    better than f(xB)

    xA

  • 7/29/2019 Design Search and Optimization_mynotes

    56/59

    More controlled ways of working all involve trying to simulate uncertainty in the design

    calculations being used. A prerequisite is to know something of the real uncertainties

    anticipated. This requires DATA.

    If however we know something of the uncertainties inherent in the design, manufacture and

    operation we can attempt to account for this.

    The most obvious, simple and direct scheme is the Monte Carlo approach. We simply

    generate a series of scenarios using suitably biased random numbers and run our design

    calculations at each before working with mean or worst case designs unfortunately

    generating such means or worst cases requires 100s of simulations this is usually way too

    expensive.

    The next approach is to replace full Monte Carlo sampling with limited size DoE variations

    around each design point to gain some idea of local sensitivities. Typically one use between

    10 and 30 variations at each design to try and characterize the issues.

    A third approach is the so called noisy phenotype. In this case a normal design search is

    carried out but at each iteration noise is added to the design variables. Then the resulting

    perturbed design results are used to characterise the design. This makes all derived qualities

    non-stationary during the search so a suitable method must be used that is tolerant of this.

    Sometimes the nominal and perturbed designs are both evaluated and the worst used as the

    characterising design. When used with a GA for example this tends to mean only robust

    designs survive the evolutionary process.

    A method increasingly popular in industry is to run a medium sized DoE and then build a

    global response surface through the resulting data. This response surface is then used for

    large scale Monte Carlo sampling to build models of robustness. The main weakness of this

    scheme is that the more unusual design and events that lead to extremes of behaviour may not

    be captured by this process. Therefore as design decisions focus in on promising areas

    update points should be run and the RSM rebuilt and re-explored so that the surrogate model

    is well set up where it needs to be.

    The most complex approach to robustness is to build so called stochastic solvers. These arecodes like finite element codes which instead of reading in deterministic problem statements

  • 7/29/2019 Design Search and Optimization_mynotes

    57/59

    for geometry and loadings can accept these specified in probabilistic form. They then

    directly compute probability measures for the response quantities of interest. Such methods

    are currently in their infancy but can be expected to become prevalent over the next 10-20

    years.

    In whatever form robustness is considered it invariably leads to a multi-objective design

    problem because the designer will desire good performance for the nominal geometry AND

    robustness to likely variations. This tends to lead to Pareto fronts with mean and standard

    deviation as axes.

    Dotted lines show variance

    f(x)

    xa xb

    Range at xb

    ROBUST

    Non dominated

    designs

    PARETO FRONT

    Locus of expected value and variance as x

    increases

    E[f(x)]

    VAR[f(x)]

    Range

    at x1

    Fragile

  • 7/29/2019 Design Search and Optimization_mynotes

    58/59

    Construction of Pareto front from variance analysis.

    Getting started

    Assuming one has a reasonable toolkit of search methods, a parameterisation scheme and an

    automated (or at least mechanistically repeatable) design analysis process it is then possible

    to make some plans as to how to proceed. These will be dominated by the number of

    designer chosen variables and the run time to evaluate a design. Other important aspects will

    be the number of goals, the number and type of constraints and whether or not stochastic

    measures of merit must be constructed using an essentially deterministic code.

    The two flow charts in the PPT deck then give some advice on how to begin

    Describe the various types of optimizers available to tackle non-linear

    search problems and the range of typical problem types encountered in

    design. Pay particular attention to speed accuracy, robustness and usability

  • 7/29/2019 Design Search and Optimization_mynotes

    59/59

    Describe the ways in which optimization tools may be used to tackle

    design problems with multiple goals, paying particular attention to how

    goals may be combined or dealt with simultaneously


Recommended