Lecture 4 Robust Design
Lecture 4 Robust Design
page 1
Product Cost and Quality
• The inherent cost to make a product is a
function of its design
• Minimizing the product's cost to the lowest
possible level within the limits set by its
design is largely a matter of avoiding
defects, tolerance deviations, and other
errors during production
page 2
Costs of Quality Deficiencies
• Scrapped parts
• Larger lot sizes for scrap allowances
• Rework, re-inspection,
• Customer complaints and returns
• Warranty costs
• Lost sales
• Lost good will in the marketplace
page 3
The Cost of Poor Quality (COPQ) “Iceberg”
Inspection Traditional Quality Costs
Warranty
Scrap (tangible)
Rework
Rejects
Profit
Cost of Cost of
Poor Quality Poor Quality
COPQ COPQ COPQ
Total Cost to
manufacture
and deliver Theoretical Theoretical Theoretical
products Costs Costs Costs
Uniform Distribution
LS US
Zero Defects!
page 6
Uniform Distribution
Target(customer preference)
C B A A B C
LS Uniform Distribution US
page 7
Adjust to Target
• Target and small normally distributed
variation about the target produce customer
satisfaction.
Target
Normal Distribution
page 8
Use Specifications, but….
• Think small variation
– make performance consistent
– reduce sensitivity to all forms of variation
• Think target
– bring average performance to customer
preference
page 9
Traditional Quality Metric
• All products within specifications equality good.
• All products outside specifications equally bad.
unacceptable
page 10
Continuous Improvement
• Move specifications closer
• Increase cost
• Quality and cost trade-off!
tighter
tolerances
LS US
page 11
Taguchi Methods
• G. Taguchi has had an important influence
on the development of quality engineering,
especially in the design area ‑ both product
design and process design
• Taguchi’s contributions include:
1. The Taguchi loss function
2. Robust design
3. Off‑line and on‑line quality control
page 12
Taguchi Methods
• Taguchi advocates a 3 step, off-line quality control method
for product design
Step 1. System Design
– concept design and synthesis
– innovation and creativity
Step 2. Parameter Design
– parameter sizing to ensure
robustness to variations
Step 3. Tolerance Design
– establish product and process
tolerances to minimize costs
page 13
The Taguchi Loss Function
page 14
Taguchi Loss Function - continued
page 15
Taguchi Loss Function - continued
• Loss occurs when a product's functional
characteristic differs from its nominal or target
value
• When the dimension of a component deviates
from its nominal value, the component's
function is adversely affected
– No matter how small the deviation, there is
some loss in function
– The loss increases at an accelerating rate
as the deviation grows, according to
Taguchi
page 16
Determining
Quality Loss Function
• Specifications are set
• Specify a target with minimal variation
• Increase in variation cause loss to society
page 17
Taguchi Loss Function
Lower Upper
Spec Limit Spec Limit
.500
.498 .502
page 18
Traditional Approach to Quality Control
• If the product dimension is within the
tolerance limits, it is acceptable
– Whether the dimension is close to the
nominal value or close to one of the
tolerance limits, it is acceptable
• The reality is that products closer to the
nominal specification are better quality
– In order to improve quality, one must
attempt to reduce the loss by designing the
product and process to be as close as
possible to the nominal value
page 19
Taguchi Loss Function
Lower Upper
Spec Limit Spec Limit
.500
.498 .502
Loss Loss
Lower Upper
Spec Limit Spec Limit
.500
.498 .502
page 22
Specify a target with minimal variation
page 23
Specify a target with minimal variation
page 24
Taguchi Loss Function
L=k(y-m)2 L
A0
k= 2
∆0 A0
page 26
Robust Design
page 27
Robust Design - continued
A robust design is:
• A design in which the product's function and
performance are relatively insensitive to
variations in design and manufacturing
parameters
– Involves the design of both the product and
process so that the manufactured product will
be relatively unaffected by all noise factors
page 28
Robust Design Concept
• Traditional approach:
– Minimize noise & variation by better control of
design parameters of product/process
– Often expensive and may or may not work
page 30
Off‑Line and On‑Line Quality Control
• Off‑line quality control - concerned with
design issues, both product design and
process design
– It precedes on‑line control
• On‑line quality control - concerned with
production operations and customer
relations after shipment
– Objective is to manufacture products
within the specifications defined in
product design, using methods and
procedures developed in process design
page 31
Two Stages in Off‑Line Quality Control
1. Product design stage - involves development of
a new product or a new model of an existing
product
– Goals: to properly identify customer needs
and to design a product that meets those
needs but can also be made consistently and
economically
2. Process design stage - the manufacturing
engineering function
– Concerned with specifying the processes
and equipment, setting work standards,
documenting procedures, and developing
clear and workable specifications for
manufacturing page 32
Three Step Approach in Product Design and
Process Design
• System design - application of engineering
knowledge and analysis to develop a prototype
design that will meet customer needs
• Parameter design - determining optimal
parameter settings for the product and process
– This stage is where a robust design is
achieved
• Tolerance design - attempts to achieve a
balance between setting wide tolerances to
facilitate manufacture and minimizing
tolerances to optimize product performance
page 33
The Quadratic Loss Function
L=k(y-m)2 L
A0
k= 2
∆0 A0
page 34
Loss Function
• The steeper the slope, the more important the
loss function
• Assuming that the functional tolerance range is
m-∆, m+∆ and we know the consumer loss as
A($) we calculate:
A = k∆2 or k = A/∆2
So
L(y) = (A/∆2)(y-m)2
page 35
Expected Loss
• Expected loss is the mean loss over over n products
• The expectation is taken with respect to the
distribution of the quality characteristic y
E[L(y)] = E[k(y - m)2]
= k(variance of y + squared bias of y)
= k[Var(y) + (µ – m)2]
= k (MSD)
Where Mean Square Deviation - MSD is given
by
n 2
MSD = ∑ ( yi − m) / n
i =1 page 36
Example
• The customer tolerance for the height of a steering
mechanism are 1.5±0.02mm. For a product that
just exceeds these limits, the cost to the consumer
for getting it fixed is $50.
page 38
Example
E(y – 1.5)2 is estimated as
10 2
= ∑ ( yi − 1.5) / n
i =1
= 0.0049 / 10
= 0.00049
Hence, expected loss per unit is