Вы находитесь на странице: 1из 25

Multiple Objectives - Outline

• Conflicting objectives
• Multi-objective optimisation

• Reaching a compromise

• Value functions and exchange constants

• Case studies: casing for minidisk player


air tanks for truck

Multiple Constraints and Objectives 1


The problem of conflicting objectives
• Real-life often requires a compromise between
conflicting objectives:
– Price versus performance of a bike or car
– Accumulating wealth versus quality of life
– Health versus the pleasures of rich food.

• Conflict arises because the choice that optimizes one


metric of performance will not in general do the same
for the others.

• Best choice is a compromise, optimizing none but


pushing all as close to optimum as their
interdependence allows.

Multiple Constraints and Objectives 2


Conflicting objectives in design
• Common design objectives that influence the choice
of material, are:
– Minimizing mass (sprint bike; satellite components)
– Minimizing volume (mobile phone; minidisk player)
Objectives – Maximizing energy density (flywheels, springs)
– Minimizing eco-impact (packaging)
– Minimizing cost (everything)

• Each objective defines a performance metric. Take,


as example
– mass, m we wish to minimize both,
– cost, C (all other constraints being met)

• Solutions that minimize mass seldom minimize cost,


and vice versa.
Multiple Constraints and Objectives 3
Multiple Objectives: The Terminology

Heavy
• Solution: a viable choice,
meeting constraints, but not
necessarily optimum by either
criterion. A Dominated

Metric 1: mass m
solution
• Dominated solution: one that
is unambiguously non-optimal B Non-dominated
(as A: some other solution is solution
better by both metrics)

• Non-dominated solution: one Trade-off


surface
that is optimal by one metric (as
B: optimal by one criterion but
Light

not necessarily by both)


Cheap Metric 2: cost C Expensive

• Trade-off surface: the surface on which the non-dominated solutions


lie (also called the Pareto Front)
Three strategies for finding best compromise
Multiple Constraints and Objectives 4
Finding a compromise: Strategy 1
1. Make trade-off plot
• Cost vs. mass of bicycles

2. Sketch a trade-off surface


• Connect non-dominated
solutions

3. Use intuition
• Select a solution on the
trade-off surface

Mass and cost of bicycles:

• Well defined trade-off surface

• “Solutions” on or near the surface offer the best compromise between mass and cost

• Choose from among these; the choice depends on how highly you value a light
bicycle, -- a question of relative values

Multiple Constraints and Objectives 5


Finding a compromise: Strategy 2
1. Make trade-off plot
• Cost vs. mass of bicycles

2. Sketch a trade-off surface


• Connect non-dominated
solutions

3. Reformulate all but one of


the objectives as constraints Upper limit on Cost
• Set upper limits for them
Lightest solution <$1000

Mass and cost of bicycles:

• Good if you have budget limit

• Trade-off surface leads you to the best choice within budget

• But not a true optimisation -- cost has been treated as a constraint, not an objective.

Multiple Constraints and Objectives 6


Cost, Price and Utility
• Any product has a Cost, a Price, and a Utility

• The Cost is the sum of all manufacturing and


distribution costs

• The Price is the selling price

• The Utility is the consumers perceived worth of the


product.

• As long as C < P < U, everyone is happy

Multiple Constraints and Objectives 7


Value Functions
• For each objective, we can derive a performance
equation with the general form:
1
 F  52 ρ
2

P1 = f1 (F )g1 (G )m1 (M ) e.g. mass =   L


δ
1
4
 max  E 2

• The difficulty with competing objectives is that they


are rarely expressed in the same units.
• Define a Value Function:

V = E1$ P1 + E2$ P2 + E3$ P3 K


where E1$ is an exchange constant - the change in
value associated with a change in the performance
measure P1.

Multiple Constraints and Objectives 8


The exchange constant, E$
e.g. mass and cost of an transportation system component:
V = E1$ × mass + E2$ × cost
 ∂V   ∂V 
E =
$
 E =
$
2  = −1
 ∂cost  mass
1
 ∂mass  cost

Value V and exchange constants Ei are all negative numbers.

The best value has the


largest (least negative) V.

Multiple Constraints and Objectives 9


Finding a compromise: Strategy 3
1. Make trade-off plot Increasing
• Cost vs. mass of bicycles value
2. Sketch a trade-off surface
• Connect non-dominated
solutions E1$
3. Plot contours of constant V 1

V = E1$ × mass + (− 1)cost


cost = E1$ mass − V
Line with slope = E1$ ; y - int = −V
Best solution for E1 = -1000
$ $
kg

4. Select a solution on the trade-off surface that maximizes V

Multiple Constraints and Objectives 10


Case study: Casing for a minidisk player
z Electronic equipment -- portable
computers, players, mobile phones -- all
miniaturized; many now less than 12
mm thick

z An ABS or Polycarbonate casing has to


be > 1mm thick to be stiff enough for
protection; casing occupies 20% of the
volume

z Find best material for a stiff casing of minimum thickness and weight

Objective 1 minimize casing thickness

Objective 2 minimize casing mass

z The thinnest may not be the lightest – need to explore the trade-off

Multiple Constraints and Objectives 11


Performance metrics for the casing
F
Function Stiff casing
w
t
Constraints Deflection of the panel < δmax :
L
3 3
FL wt
δ max ≥ ; I= w = panel width
CEI 12 L = length
Adequate toughness, G1c > 1kJ/m2 ρ = density
t = panel thickness
I = second moment of area
Objective 1 Minimize thickness t E = Young’s Modulus
1/ 3
 S L3  1
Metric 1 t=  ∝
 4E w  E1/ 3
 

Objective 2 Minimize mass m


1/ 3
 12 S w 2   ρ  ρ
Metric 2 m=  L2  1/ 3  ∝ 1/ 3
 C  E  E
 
Multiple Constraints and Objectives 12
Relative Performance Metrics
• We are interested here in substitution. Suppose the
casing is currently made of a material M0.

• The thickness of a casing made from an alternative


material M, differs (for the same stiffness) from one
made of M0 by the factor: 1
t  E0  3
= 
t0  E 

m  ρ   E0 3 
1

• The mass differs by a factor: =  1  ⋅  


m0  E 3   ρ 0 

t m
• Explore the trade-off between and
t0 m0
Multiple Constraints and Objectives 13
The trade-off plot
Elastomers
Trade-off
10
surface
Lead
ABS, m/mo

Cu-alloys
PTFE
Ni-alloys
Ionomer Additional
relative totoABS

Steels ABS constraints:


PE
PC
Ti-alloys
Massrelative

1 PMMA z G1c > 1kJ/m2


Al-alloys PP
Polyester z Wood, leather
Al-SiC Composite Nylon suppressed.
Mass

Mg-alloys
CFRP
GFRP .
Polymer foams

0.1

0.1 1 10
Thickness relative to ABS
Thickness relative to ABS, t/to

z Finding a compromise: CFRP, Al and Mg alloys all offer reduction in mass and thickness
Multiple Constraints and Objectives 14
Postscript
z The four sectors of a trade-off plot for substitution

B. Thinner Elastomers
Trade-off D. Worse by
10
but heavier
surface both metrics
to ABS, m/mo
Lead

Cu-alloys
PTFE
Ni-alloys
Ionomer
relative to ABS

Steels ABS
PE
PC
Ti-alloys
relative

1 PMMA
Al-alloys PP
Al-SiC Composite Polyester
Mass Mass

Nylon

Mg-alloys
CFRP
GFRP .
C.foams
Polymer Lighter
A. Better by
but thicker
0.1
both metrics

0.1 1 10
Thickness relative to ABS
Thickness relative to ABS, t/to

z Is material cost relevant? Probably not -- the case only weighs


a few grams. Volume and weight are much more valuable.

Multiple Constraints and Objectives 15


CES Demo – Minidisk
2
K IC
1. Create Toughness stage: GIC =
E (1 + ν )

2. Get E0 and rho0 for ABS: E0 = 2 GPa; ρ 0 = 1.1 Mg / m 3

3. Create a trade-off plot (linear axes)


 ρ   2GPa 
y − axis :  1  ⋅  
3 
 E   1.1Mg / m 
3

1
 2GPa 
3

x − axis :  
 E 

4. Create a trade-off plot


• Demonstrate linear vs. log axes
• Plot value functions with different exchange constants
• Switch between log and linear axes (unzoomed).

Multiple Constraints and Objectives 16


Case study: air cylinders for trucks
Design goal: lighter, cheap air cylinders for trucks

Compressed air tank

Multiple Constraints and Objectives 17


Design requirements for the air
cylinder
Specification t

Function Pressure vessel Pressure p 2R

• Minimise mass L
Objectives
• Minimise cost

R = radius
• Dimensions L, R, pressure p, given L = length
Constraints • Must not corrode in water or oil ρ = density
p = pressure
• Working temperature -50 to +100°C t = wall thickness
• Safety: must not fail by yielding
• Adequate toughness: K1c > 15 MPa.m1/2

Free • Wall thickness, t;


variables • Choice of material

Multiple Constraints and Objectives 18


Analysis of the air cylinder
Vol of material in cylinder wall t

Objective 1
(
m = 2πR L t + 4πR2t ρ ) Pressure p 2R

 2R 
L
= 2πR L t1 + 
 L 
Aspect ratio Q R = radius
pR σy L = length
Constraint σ= < ρ = density
t S p = pressure
Eliminate t to give: t = wall thickness
σy= yield strength
2 ρ S = safety factor
Metric 1 m = 2 πR L(1 + Q) p S   Q = aspect ratio 2R/L
 σ y 

Objective 2 C = Cm m

2  Cm ρ 
Metric 2 C = 2 πR L(1+ Q) p S  
σ
 y 
Multiple Constraints and Objectives 19
Relative mass and cost
z This is another problem of substitution. The tank is currently made of a
plain carbon steel.
z The mass m and cost C of a tank made from an alternative material M,
differs (for the same strength) from one made of Mo by the factors:

m  ρ   σ y 0  C  Cm ρ   σ y 0 
= ⋅   = ⋅  
 
m0  σ y   ρ 0   
C0  σ y   C m 0 ρ 0 

z For plain carbon steel ρ 0 = 7.85 Mg m 3 Cm 0 = 0.75 CAD $ kg σ y 0 = 320MPa

m C
z Explore the trade-off between and
m0 C0

Multiple Constraints and Objectives 20


Trade-off Plot
Trade-off
Surface

10
Copper alloys

Low C steel Zinc alloys


Mass Relative to Annealed 1020 Steel

Med. C steel
Stainless steel
High C steel Nickel alloys

Titanium alloys

Low alloy steel


0.1
Al alloys Mg alloys

GFRP
Al/SiC Composite
CFRP

0.01
0.1 1 10 100
Cost Relative to Annealed 1020 Steel

Multiple Constraints and Objectives 21


Finding a compromise: the value function
zAluminum alloys and low alloy steels offer modest reductions in
mass at little or no increase in material cost.

z The lightest solutions are GFRP, CRRP and Titanium alloys, but at
a cost penalty -- is it worth it? Define a relative value function:
V E1m − C
V =*
=
C0 C0
E1 m0 m C
if E =
*
1 then V = E *
− *
1
C0 m0 C0

z With mo = 10 kg, Co = $100 and E1 = -$10/kg (trucks), E* = -1 .


z (a) evaluate V* numerically and rank candidates, or
z (b) plot onto relative trade-off plot (lines of slope = -1)

Multiple Constraints and Objectives 22


Value function on trade-off plot

m C Selection for E*= -1 (Em= -10$/kg)


V = (− 1) −
*

m0 C0
m C
10

⇒ =− −V * Mass Relative to Annealed 1020 Steel


Low C steel Zinc alloys

m0 C0 Med. C steel

Stainless steel
High C steel

Value line is curved


because of logarithmic Low alloy steel

scales. 0.1
Al alloys Mg alloys

0.1 1 10
Cost Relative to Annealed 1020 Steel

Multiple Constraints and Objectives 23


Value function on trade-off plot

Selection for E*= -.01 (Em= -1000$/kg)


m C
V =E
*
− *

m0 C0
10

Mass Relative to Annealed 1020 Steel


Low C steel Zinc alloys

Med. C steel

Stainless steel
$1000/kg is typical of High C steel

high-performance 1

aircraft applications.

The added expense of


composite materials is Low alloy steel
warranted in these Al alloys Mg alloys

applications. 0.1
Titanium alloys
Al-SiC Composite

GFRP (isotropic) CFRP (isotropic)


0.1 1 10
Cost Relative to Annealed 1020 Steel

Multiple Constraints and Objectives 24


The main points
• Real design problems involve conflicting objectives
– often technical vs. economic performance (cost).

• Trade-off plots reveal the options, and (when combined


with the other constraints of the design) frequently point
to a final choice.

• If the relative value of the two metrics of performance


(measured by and exchange constant) is known, a value
function allows an unambiguous selection.

Multiple Constraints and Objectives 25

Вам также может понравиться