Advanced Optimisation

Approaches

DR. PRASHANT PAWAR

pawarpm@sveri.ac.in
The World Around Us

“All modern products are designed as a SYSTEM”

The World Around Us
AIRCRAFT SPACECRAFT AUTOMOBILES BUILDINGS
Aerodynamics Astrodynamics Engines Structure & Seismology

Propulsion Structures Body/chassis Space and Aesthetics

Structures Communications Aerodynamics Aerodynamics/Wind

Controls Payload & Sensor Electronics HVAC

Avionics/Software Optics Hydraulics Networking

Manufacturing Guidance & Control Industrial design Fire & Safety

Others Others Others Others

The World Within Us
SYSTEMS?
Relationships System

Components

A Component can itself be a SYSTEM.

System .

Subsystem
SYSTEM COMPLEXITY
6

Analysis codes should reside

with experts ?
System analysis should execute
analysis codes on experts’
Aerodynamics Expert
computers or … ?

Structures Expert System Analysis Controls Expert

 RS Model

MODELING
7

Meta Model
The Modeling Space

2
y = b0 + ∑ bi xi + ∑ bii xi + ∑∑ bii xi x j

f(t) + m!x! 1 !x! 1 x! 1 x

x(t)
- m s s
input - output
Model c!
x x!

kx x

m!x! + cx! + kx = f (t )
World

Physical system
EVOLUTION OF DESIGN PROCESS
8

Trial & Error Empirical Mathematical Probabilistic

Deterministic Stochastic
(Factors of Safety) (Risk Quantified)

Random
Experimentation

``
Experience-based
Graphical Approaches Computer models Computer
based on system Simulations based on
Systematic physics
 system physics

Experimentation

Point estimates Robust Solutions
EVOLUTION OF DESIGN PROCESS
9

CONCEPTUAL PRELIMINARY
DESIGN DESIGN DETAIL
DESIGN
PRODUCT
ATTRIBUTES
100:1
10:1
1:1

Conceptual design is crucial to the success of the overall design process and
resulting system. It has been estimated that “at least 80% of a Mission’s life-
cycle cost is locked in by the concept that is chosen” and “conceptual design
decision have a 100:1 leverage on end product quality and cost”
EVOLUTION OF DESIGN PROCESS
10

CONVENTIONAL OPTIMAL
1. Specification 1. Specification
2. Baseline design 2. Baseline design
3. Analysis (or experiment) 3. Analysis
4. Check performance or failure criteria 4. Check constraints
5. Does design satisfy the optimality
5. Is design satisfactory?
conditions? (If yes, then stop)
(If yes, then stop)
6. Change design parameters using an
6. Change design parameters based on optimization strategy, return to 3.
intuition and experience, return to 3.
 WHAT IS OPTIMIZATION?
● “Making things better”
● “Generating more profit”
● “Determining the best”
● “Do more with less”
 WHAT IS OPTIMIZATION?
“The determination of values for design variables
which minimize (maximize) the objective, while
satisfying all constraints”
Principles of Optimal Design: Modeling and Computation
2d Ed. by Panos Y. Papalambros and Douglass J. Wilde, Cambridge University Press, New York, 1988,
2000.
 OPTIMIZATION

Design Space: The space of working (Hill in this case)

Objective: Find the Highest Point.
Design Variables: Longitude and latitude.
Constraints: Stay inside the fences.
OPTIMIZATION

Objective Function
Constraints
Bounds

Design Variables
LOCAL AND GLOBAL OPTIMA
LOCAL maxima
OPTIMA

Local Local
maxima minima
minima
GLOBAL
MINIMA
Optimization Problems
Optimization Problems
MDO: Overview
Multidisciplinary design optimization (MDO) deals with the optimization of several
engineering disciplines simultaneously.

MDO gives the engineer the opportunity to find the optimal solution of some
system accounting for the interactions between the different disciplines.

It should be noted that the multidisciplinary solution might not be the solution for
any one discipline analyzed separate from the other disciplines, but is the best
solution accounting for the interactions.

The MDO field has become vital in design environments in the past decades as
designs are becoming more and more complex.
MDO: Framework
MDO: Overview
AIRCRAFT DESIGN
Problem formulation is not obvious
and requires engineering judgment.

One can only make one thing best

at a time.”
MSDO: Terminology
IMPORTANCE OF DESIGN: XEROX COPIER
Controlled market in 60s and 70s. Shares fell significantly in 80s.
Competition with 40 other companies. They sold copier at less than
manufacturing cost of copier by Xerox.
Comparing plastic parts with Japanese and Europeans, they found that
Japanese firms could produce a part for 50% less than American and
European firms.
Xerox attributed the cost difference to three factors;
¤ Material costs were 10% less in Japan.

¤ Tooling and processing costs were 15% less.

¤ Remaining 25% (Half of the difference) was attributed to how the parts
were DESIGNED
 What Is Probabilistic Engineering Design?
23
 What Is Probabilistic Engineering Design?
24
 Effect of Uncertainty
25
 How to Deal with Uncertainty
26
 Understand your System
27

The Blind Man and the Elephant

What we learn from an experiment may depend on where we

look, how we look, and the scope of our view!
Cost of Experimentation

Resources (people, equipment, machinery, computers etc.)

Time
Material (unprocessed or unusable product)

28
Crash Simulation tests: An example

Each simulation
takes 36 -160 hrs
for
Ford Motor Inc.
 Design of Experiments
30

DoE provides a strong tool to design and analyze

experiments:
¤ Maximizes the knowledge
¤ Eliminates redundant observations
¤ Reduces the time
¤ and Resources to make experiments
 Major Approaches to DOE
31

Factorial Design
Taguchi Method
Response Surface Design
 Design of Experiments
32

The design of experiment is the sampling plan in design

variable space.
Many researchers advocate the use of “space filling”
designs when sampling deterministic computer analyses to
treat all regions of the design space equally.
 Meta Model Developing Sequence
33

Initialize Design Space SELECT EXPERIMENTAL DESIGN

_____________________ _____________________________________
Lower bound ≤ xi ≤ Upperbound •Central Composite Design
•Orthogonal Array
•Latin Hypercube Sampling
Pruning
_____________________
Removal of unreasonable designs

Sample Design Space CONSTRUCT SURROGATE MODEL

__________________________ _____________________________________
Training Data •Response Surface
•Kriging
•Least Square – Support Vector Machine

Validate Approximations
α k, b
_____________________
Using validation data
METAMODELING Approaches

Regression models
Support Vector Machines
Artificial neural networks
Kriging
Many others
 Meta-Model based Multidisciplinary Design and Optimization Strategy
35

Multidisciplinary Design Design Variables (X)

Analysis Module OPTIMIZATION MODULE
___________________ ______________________
DESIGN OF EXPERIMENTS
______________________
Latin Hypercube Sampling Find: Optimum Design
Variables (X*)
Vehicle Configuration
Satisfy: Constraints

Weight Analysis
Minimize: GLOM, Mg
META-MODELING MODULE (SVM)
Propulsion Analysis ______________________
Support Vector Regression
Trajectory Analysis
Meta Model based Optimal Design
(X*SVM)
 Input Response

x1 Black Boxed
System y
x2

Original System DOE and Experiments RS Model

x2 0

-1 0
-1
1
x1

2
y = b0 + ∑ bi xi + ∑ bii xi + ∑∑ bii xi x j
 Optimization Problem

Variable Selection Sensitivity

Approximation Domain

Analysis
DOE & Experiments
Sensitivity
If RSM using Sensitivity
Construct RSM

No Reliable
RSM ?
Test Criteria Yes

Optimization Using RSM Optimizer

Estimated Opt Response

No
Reliable OPT? 1 Analysis
Calc Error at opt from RSM
Yes

Final Optimal Solution

Example - AUTOMOTIVE SIDE IMPACT

Example by K.K.Choi, U. of Iowa, “Moving Least Square Method for Reliability-Based Design
Optimization”, WCSMO4, 2001
 What is Robust Design?
39

Example: We want to pick x to maximize F

Simply doing a trade study to optimize the value of F would

F lead the designer to pick this point
What if I pick this
point instead?
This means that
values of F as low
as this can be
expected!

Robust design: a design whose performance is insensitive to variations.

Uncertainty Analysis
Uncertainty Analysis
Methods

Non-Probabilistic Probabilistic
Methods Methods

Interval Fuzzy Sensitivity Moment Polynomial Monte Carlo

Analysis Logic Derivatives Methods Chaos Methods
 ROBUST DESIGN Strategy
41

Multidisciplinary Design Variables (X) OPTIMIZATION

Design Analysis MODULE
Module ______________________
___________________ Find: Control Variables (X)
DESIGN OF Satisfy:
EXPERIMENTS Constraints
______________________ Minimize:
Vehicle Configuration Latin Hypercube Sampling Weighted Sum of Mean and
Worst Case Deviation

Weight Analysis

UNCERTAINTY
Propulsion Analysis
PROPAGATION
______________________
Worst Case Variation
Trajectory Analysis

Robust Design (X*Robust)

 Monte Carlo Simulations
42
Monte Carlo Analysis
x1
y1

x2

x3 Sources of Variability
y2 Solution:
•
•
Material Properties
Loads Establish tolerances for the
•
•
Boundary and initial conditions
Geometry imperfections input and design variables.
• Assembly imperfections
• Solver
•
•
Computer (round-off, truncation, etc.)
Engineer (choice of element type, algorithm, Measure the system’s
mesh band-width, etc.)
response in statistical terms.
Taguchi : Introduction
Genichi Taguchi (⽥⼝ ⽞⼀)
From 1950s developed a methodology to improve the quality
of products.
Much of his work was carried out in isolation from the
mainstream of Western statistics.
Unknown outside of Japan. Introduced into US in 1980.
Taguchi’s method.
Controversial among statisticians, but many concepts
introduced by him have been accepted
 Overview of Taguchi Parameter Design
Method
45

1. Brainstorming
Design Parameters: Variables under your control
2. Identify Design Parameters
and Noise Factors Noise Factors: Variables you cannot control or
variables that are too expensive
3. Construct Design of to control
Experiments (DOEs) Ideally, you would like to investigate all possible
combinations of design parameters and noise
4. Perform Experiments factors and then pick the best design parameters.
Unfortunately, cost and schedule constraints
frequently prevent us from performing this many
test cases – this is where DOEs come in!
5. Analyze Results
SIMULATED ANNEALING FLOWCHART
Initialisation

Metropolis simulation with fixed

temperature T

Generate new solution

Adjust the solution

Evaluate cost function

Cooling
temperature T
Improvement
Yes No
Accept new Accept new solution
solution with a probability

Check for equilibrium No

Yes
No Stop criteria at outer loop

Yes
Return optimal solution
SA ALGORITHM
SA ALGORITHM
 GA flowchart
Create initial
population Evaluate fitness
of all individuals

Create new population Test termination

criteria
Crossover Mutation Reproduction

Select individuals
for reproduction Quit
Termination criteria can be fixed number of generations, a certain required fitness level is reached, no more improvement in X
generations
Flowchart of a simple GA
Initial population

Evaluation

Selection

Crossover

Mutation/ Inversion
GA: Basic Algorithm
[Start] Generate random population of n chromosomes (suitable solutions for the problem)
[Fitness] Evaluate the fitness f(x) of each chromosome x in the population
[New population] Create a new population by repeating following steps until the new
population is complete
¤ [Selection] Select two parent chromosomes from a population according to their fitness (the better
fitness, the bigger chance to be selected)
¤ [Crossover] With a crossover probability cross over the parents to form a new offspring (children). If
no crossover was performed, offspring is an exact copy of parents.
¤ [Mutation] With a mutation probability mutate new offspring at each locus (position in chromosome).
¤ [Accepting] Place new offspring in a new population
[Replace] Use new generated population for a further run of algorithm
[Test] If the end condition is satisfied, stop, and return the best solution in current population
[Loop] Go to step 2
Nature Vs Computer - Mapping
Nature Computer
Population Set of solutions.
Individual Solution to a problem.
Fitness Quality of a solution.
Chromosome Encoding for a Solution.
Gene Part of the encoding of a solution.
Reproduction Crossover

52
Nature Vs Computer - Mapping
Classical Algorithm Genetic Algorithm
Generates a single point at each Generates a population of points at
iteration. each iteration.

The best point in the

The sequence of points population approaches an optimal
approaches an optimal solution. solution.

Selects the next point in the sequence Selects the next population by
by a deterministic computation. computation which uses random
number generators.
 Optimal Design of the MR Fluid Dampers

● MR dampers are semi-active devices, which are

similar to passive dampers, except the fluid

● Due to controllable viscosity of MR fluids various

applications like breaks, clutches and dampers
[2,3] are developed.

• Detailed numerical and mathematical model and experimental verifications and

prototyping made it possible to implement the MR dampers for various applications in the
mechanical, automobile and aerospace industry.

• However, very few studies have devoted efforts on towards optimal design of MR dampers.

• Optimal design of MR damper by the classical method becomes complicated because of

nonlinear relationship between the variables,
 Parallel plate MR damper model
● Controllable force and dynamic range are
the two most important performance
parameters performance of MR dampers.

• The damper resisting force can be decomposed into a controllable force

due to controllable yield stress and an uncontrollable force.
F Fτ
D= = 1+
Dynamic range Fuc Fη + F f

Controllable Force due to ⎛ 12Qη ⎞ τ 0 LAp

Fτ = ⎜⎜ 2.07 + 2
⎟⎟ sgn (v0 )
to controllable yield stress 12Qη + 0.4 wh τ h
⎝ 0 ⎠

Viscous force
⎛ whv0 ⎞ 12ηQLAp
Fη = ⎜⎜1 + ⎟⎟
⎝ 2Q ⎠ wh 3
 Optimization problem
● Multi-objective optimization function for reference controllable force and
dynamic range

Maximize Fτr D
J =α + (1 − α ) r
Fτ D
Subjected to Rmin < R < Rmax
(h/R)min < (h/R) < (h/R)max
Lmin < L < Lmax

ηmin< η < ηmax

τ 0 Min < τ0 < τ 0 Max
 Dynamic Range Dynamic Range

Gap Ratio

α
Controllable Force Controllable Force
α
Optimal Design Results

Gap Ratio
 Yield Stress Radius of cylinder

α
α

Pole Length Gap Ratio

α
α
Optimal Design parameters
59
Optimal values of design variables obtained for
maximization of surface velocities
Model TS in fl1 fl2 fl3 df1 df2 df3 Objective
(%) (m) (m) (m) (m) (m) (m) function

S-I 2.5 0.37 0.28 0.39 1.071 0.950 2.00 1.700

S-II 5.4 0.39 0.11 0.40 0.005 0.991 2.00 0.292

S-III 7.5 0.38 0.20 0.35 0.980 0.970 1.96 0.164

60
 Velocity profile for TS 2.5% Concentration

Baseline model S-I model

S-II model S-III model

Research Gap

62
 Fuzzy Logic

• Unique soft computing method which simultaneously handles

numerical data and linguistic knowledge.

• Nonlinear mapping of an input feature vector into a scalar

output.

• A typical fuzzy logic system maps crisp inputs to crisp outputs

using four basic components: rules, fuzzifer, inference engine,
and defuzzifer.
 Genetic Fuzzy System

● Hybridization of GA and FL gives genetic fuzzy system

(GFS) in which a GA is used to evolve a fuzzy system by
tuning fuzzy membership functions and learning fuzzy rules.
 Genetic Fuzzy System

● Input and Output: In damage detection problem, the

measurement deltas (change in measurements between
damaged and undamaged model) can be used as inputs (z)
and the combinations of damage levels and locations will be
the output (x) of fuzzy system.
● Fuzzifcation :Structure divided into various locations. Each
of these damage locations is allowed several levels of
damage and split into linguistic variables.
● Fuzzy sets with Gaussian membership functions are used to
define these input variables.
 Rule Generation

● Midpoints of membership function are obtained by numerical

analysis corresponding to a given structural fault.

● For each measurement delta corresponding to given fault, the

degree of membership in the fuzzy set is calculated.

● Each measurement delta is assigned to the fuzzy set with the

maximum degree of membership. The “IF-THEN” rules generated
for each fault relating measurement deltas.

● FS is fine tuned by changing the standard deviation for each set.

 Finding Standard Deviation

● Calculation of standard deviation

● Success Rate

● Noise Model
 Global damage detection

● Physical damage parameters and life consumption

parameters are classified in various zones based on the
displacement measurement deltas.

● Next, the force and moment measurement deltas are also

considered for prediction.
 Local Damage Detection

● Inputs to the GFS are strain-based measurement deltas at

five locations and outputs are physical damage parameters
and life consumption parameter at different locations.
SHM development
 Testing of GFS

● Global Damage Detection

 Testing of GFS

● Local Damage
Detection
 Numerical Analysis of Bearings with Surface Modifications

Modified Partially Textured Cylindrical Shape Journal Bearing

Figure 4.1: Models of bearing systems with different circumferential rectangular shape
grooves (a) bearing with partially grooved (90-360 deg) (b) bearing with partially grooved (90-
270 deg) (c) bearing with partially grooved (90-180 deg)

Figure 4.2: Detailed view of

rectangular groove shape texture on
bearing surface

41
 Comparative Analysis of Modified Groove and
Dimple Shape Textured Journal Bearing

Figure 4.11: Model of bearing systems a) with smooth surface texturing b) with groove
texturing c) with dimple texturing d) detailed view of texturing on bearing surface
e) cut section along axis of bearing for RGST f) cut section along axis of bearing for
EDST 50
 Bearing Surface Feature Optimization
Geometric Model and Surface texturing control factors and their
levels

Factor Level Unit

1 2 3
Groove location 90-155 90-165 90-175 deg
Groove width 2.4 2.5 2.6 mm
Groove height 15 20 25 µm
No. of grooves 7 9 11 -
Spacing bet. grooves 200 300 400 µm 65
 Bearing Surface Feature Optimization
Taguchi’s Design of Experiment

Figure 5.1: Taguchi Process 64

Numerical analysis and observations for total 27 runs

66