You are on page 1of 27

MME445: Lecture 16

Multiple Constraints and


Conflicting Objectives
Part 2: Selection with multiple objectives
Learning Objectives

Knowledge & Knowledge on graphical trade-off methods


Understanding and penalty functions

Ability to select systematically when design


Skills & Abilities
objectives conflict
Appreciation of the value of compromise in
Values & Attitudes
engineering design

Resources
• M F Ashby, Materials Selection in Mechanical Design, 4th Ed., Ch. 07
Lecture Outcome
At the end of this lecture, students should be able to
• analyse the process of materials selection for problems having
multiple objectives,
• examine how exchange constant can be determined or estimated
and be used to calculate the penalty function, and
• analyse how we can select a better material for an existing
application using relative penalty function

3/27
1. Introduction & Synopsis

❑ Common design objectives:


1. Minimising mass (sprint bike, satellite components)
2. Minimising volume (mobile phone, minidisk player)
3. Minimising cost (everything)
4. Maximising energy saving
5. Minimising environmental impact (packaging, cars)

❑ Each defines a performance metric

❑ Some objectives may conflict with another


We wish to minimise both !!
(meeting all constraints, of course!)

4/27
❑ Conflict arises, because the choice that optimizes one objective
will not, in general, do the same for the others

❑ The best choice is thus a compromise, optimizing none but pushing


all as close to their optima as their interdependence allows.

❑ And this highlights the central problem:


• How is mass to be compared with cost, or volume with environmental impact?
• Each is measured in different units; they are incommensurate.
• We need strategies to deal with this.

5/27
Trade-off strategies

❑ Consider the choice of material to minimize both cost


(performance metric P1) and mass (performance metric P2) while also
meeting a set of constraints (e.g., maximum service temperature,
corrosion resistance in a certain environment)

❑ Following the standard terminology of optimizations theory,


we define a solution as a viable choice of material, meeting all the
constraints but not necessarily optimal by either of the objectives.

6/27
2. Multi-Objective Optimisation
The terminology

Expensive
• Solution:
a viable choice, meeting
constraints, but not necessarily

Metric 2: Cost, C
optimum by either objective
criterion.

• Express objectives to be
minimised
while plotting all viable solutions as
Cheap

function of performance metrics.


Light Metric 1: Mass, m Heavy

7/27
2. Multi-Objective Optimisation
The terminology

• Dominated solution:

Expensive
one that is definitely non-optimal
(as point A) A
Dominated solution
(obviously, there are solutions better

Metric 2: Cost, C
than this one)

B
Non-dominated
• Non-dominated solution: solution

one that is optimal by one metric Trade-off


(as point B) but not necessarily surface
by both metric
Cheap

Light Metric 1: Mass, m Heavy


• Trade-off surface:
the surface on which the non-dominated solutions lie
(also called the Pareto Front) (In our case a 2-dimensional curve) 8/27
Finding a compromise: Strategy 1
compromise by intuition and experience

Heavy
1. Make a trade-off plot
2. Sketch a trade-off curve
3. Use intuition to select a
solution on the trade off curve

Mass, m
• “Solutions” nearest the surface
offer the best compromise
Trade-off
between mass and cost curve
Light

• 8 out of 50
Choose from among these !! Cheap Cost, C Expensive

• Which part of the trade-off curve is the best?


The strategy is to use in selecting materials from among the non-dominated (red-colored) set
is intuition (experience, good judgment, common sense – call it what you like).
9/27
Finding a compromise: Strategy 2
treating objectives as constraints

Reformulate all but one of the

Heavy
objectives as a constraint,
setting an upper limit of it, and Setting cost
as constraint;
then optimize the other objective e.g., C < 15 $/kg

Mass, m
Trade-off
• Here the cost is set as a curve
constraint and an upper limit
of 15 $/kg (say) is set

• The solution that minimizes Best


the other objective can then choice
Light

be read off.
Cheap Cost, C Expensive

❑ But this is cheating!!


• This is not a true optimisation – cost has been treated as a constraint, not an objective.
• It is an extreme sort of optimization. 10/27
❑ Strategies 1 and 2 help with all trade-off problems
in material selection, but they rely to some extent on judgment.

❑ A more systematic method is possible – based on Penalty


Function.

11/27
Finding a compromise: Strategy 3
penalty functions and exchange constants

❑ The trade-off surface identifies the subset of solutions that offer the best
compromises between the objectives.

❑ Ultimately, though, we want a single solution.

❑ One way to do this is to aggregate the various objectives into a


single objective function, formulated such a way that its minimum
defines the most preferable solution.

❑ To do this we define a locally linear penalty function


(or, value function or utility function) :
a = exchange constant
𝑍 = 𝛼1 𝑃1 + 𝛼2 𝑃2 + 𝛼3 𝑃3 + ... P = performance metric

❑ The best choice is the material with the smallest value of Z.


12/27
❑ The a’s are called exchange constants
(or, equivalently, utility constants or scaling constants)

❑ They convert the units of performance into the units of Z,


which are usually currency ($).

❑ The exchange constants are mathematically defined by


𝜕𝑍 𝑍 = 𝛼1 𝑃1 + 𝛼2 𝑃2 + 𝛼3 𝑃3 + ...
𝛼𝑖 =
𝜕𝑃𝑖 𝑃𝑗 ,𝑗≠𝑖

• They measure the increment in penalty for a unit increment in a given


performance metric, all others being held constant.
• If, for example, the metric P2 is mass m, then a2 is the change in Z associated
with a unit increase in m, taking other metric, cost for example, is constant.
𝜕𝑍
𝛼2 =
𝜕𝑃𝑚 𝑃
𝑐

13/27
❑ Frequently one of the objectives to be minimized is cost, C, so that P1 = C.

❑ Then it makes sense to measure Z in units of currency.

❑ With this choice, a unit change in C gives a unit change in Z,


with the result that a1 = 1 and the equation of Z becomes
𝑍 = 𝐶 + 𝛼2 𝑃2 + 𝛼3 𝑃3 + ...

❑ Consider now the earlier example in which


P1 = cost C, P2 = mass m, and a2 = a, so that
𝑍 = 𝐶 + 𝛼𝑚
1 1
𝑚=− 𝐶+ 𝑍
𝛼 𝛼
❑ Then a is the change in Z associated with a unit increase in m.
The second equation defines linear a relationship between m and C.
14/27
Summary of Strategy 3:

Heavy
1. Define locally-linear penalty Z1 Z2 Z3 Z4
Contours of
function Z = C + a m constant Z
2. Seek solution with smallest Z
Trade-off

Mass, m
curve
2 ways to do this:
Optimum solution,
❑ Evaluate Z analytically for minimising Z

each solution and rank them


Decreasing
values of Z
Light
−1/a
❑ Make trade-off plot
Cheap Cost, C Expensive
▪ Plot on it contours of Z
1 1
𝑚=− 𝐶+ 𝑍 ❑ The best choices lie near the point at which
𝛼 𝛼
▪ Lines of constant Z have slope -1/a one of these lines is tangent to the trade-off
curve, since this minimizes Z.
▪ Read off solution with the lowest Z,
which lies at the bottom left corner of
the plot 15/27
❑ Almost all the material selection charts use logarithmic scales.
❑ A linear m-C relationship, plotted on log scales, appears as a curve
❑ But the procedure remains the same:
The best candidates are those nearest the point at which one of these curves
just touches the trade-off surface.

16/27
But what is the meaning of a ?

❑ The quantity a is called an “exchange constant” -- it measures the value


of performance, here the value of saving 1 kg of mass ($/kg).

❑ It quantifies the effect of a material substitution on the total value,


or the (value) penalty involved in the substitution.

How to get a ?
◼ market survey (perceived value)
◼ full life cost (engineering criteria)

17/27
Estimation/determination of exchange constants

❑ Values base on engineering criteria

• e.g., approx. derivation from technical models for the lifetime cost of a system
(e.g., weight saving in transport system)

• affects dramatically the application of materials


(e.g., aluminium alloys in aircraft, not in car bodies; titanium alloys in military
than civil aircraft)

• values of exchange constant varies with time


(change in fuel price, legislation, etc.)

18/27
❑ Values base on perceived worth
• More difficult to assess

• e.g., weight/cost trade-off for a bicycle ($20/kg - $2000/kg)

• Family car? Titanium watch?

• One aim of advertising is to increase the perceived worth of a product,


thereby increasing its value without increasing its cost

❑ Environmental impact ??
• How much is unit decrease in eco-impact worth?

• Until an exchange constant is agreed on or imposed, it is difficult


for the designer to respond

19/27
Finding a using engineering criteria
example: materials for transport systems

▪ Mass, in transport systems, means fuel


$ kg
▪ Life cost = Initial cost, C + Fuel cost over life, scaling with mass m
$/kg
▪ Penalty function Z = C + αm

▪ Must establish exchange constant, α

Choice of material depends on system

Steel Steel / Al Al / (composite) Al / Ti / composites Composites


a 3–6 6 – 20 100 – 600 600 – 2,000 (?) 5,000 – 10,000
($/kg)
Penalty function in transport system
Mass vs. cost of beam for given stiffness
System a ($US per kg)

c/E1/2 Family car 0.5~6


1e6 Trade-off Exchange Truck 5 to 20

P2= Cost surface constant


Cobased superalloys Civil aircraft 100 to 500
Ni-based superalloys Military aircraft alloys
Tungsten 500 to 1000
for given a = 500 $/kg Bicycle frame 20-4000
stiffness 100000 Ti-alloys Exchange
Epoxy/HS Carbon Space vehicle 3000 to 10000
constant
Density x Price /Sqrt Modulus

weave Lead alloys

CFRP epoxy a = 50 $/kg Bronze


10000 laminate Exchange
constant
GFRP
1000 a = 5 $/kg Copper alloys
Exchange
Zinc alloys
MAGNESIUM alloys
constant
100
a = 1 $/kg
ALUMINUM alloys
 P 
10 a = −  2  P 2 HSLA steels CAST IRONS
 P 
 1
Engineering 50 100 200 500 1000 2000
/E1/2 5000
P1=Mass Density/Sqrt Modulus
definition of a
for given
stiffness
21/27
FIGURE 7.10
It is often the case that a single material (or subset of materials) is optimal over a wide
range of values of the exchange constant. Then approximate values for exchange constants
are sufficient to reach precise conclusions about the choice of materials.
22/27
23/27
Relative penalty functions
for substitution of an existing choice

❑ When we seek a better material for an existing application, it is more helpful


to compare the new material choice with the existing one.

❑ To do this we define a relative penalty function:


𝐶 𝑚 • subscript 0 means properties of the existing material
𝑍∗ = + 𝛼 ∗
• asterisk * on Z* and a* means that both are now dimensionless
𝐶0 𝑚0

❑ The relative exchange constant a* measures the fractional gain in value


for a given fractional gain in performance.
❑ Thus a* = 1 means that, at constant Z,
∆𝐶 ∆𝑚 This means that, halving the mass
=− is perceived to be worth twice the cost
𝐶0 𝑚0
24/27
❑ Figure 7.8 shows the relative
trade-off plot on linear scales.
❑ The axes are C/Co and m/mo.
❑ The material currently used in
the application appears at the
coordinates (1,1).

Sector A : both lighter and cheaper


Sector B : cheaper but heavier
Sector C : lighter but more expensive
Sector D : uninteresting

❑ The contour of Z* that is tangent to


the relative trade-off surface again
identifies the optimum search area.

❑ So if values for the exchange


constants a are known, a completely
systematic selection is possible.
25/27
Summary & Conclusions

❑ Real design involves conflicting objectives


– often technical performance vs. economic performance (cost).

❑ If the exchange constant is known


– penalty function allows unambiguous choice.

❑ The penalty function technique can be applied to bar charts,


or bubble charts for interactive and visual selection.

26/27
Next Class
MME445: Lecture 17
Multiple Constraints and
Conflicting Objectives
Case studies with conflicting objectives

You might also like