Reliability Lessons From The

Automotive Industry

Larry Edson
Larry Edson Consulting Inc.
larryedson@hotmail.com
Cell: 248-207-2187
Home: 248-347-6212
A Special ½ Hour Presentation For The Danish Wind Turbine Industry

The Evolution of The Evolution of

Automotive Reliability YOU Wind Turbine Reliability

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 1

 The “Lessons” In a Nutshell
• Use Reliability Allocation to translate the big “I Want This Wind
Turbine To Be Reliable” into reliability requirements for devices.
• At the device level - Use the three step AQQ process shown below
to develop a mature device on time:
1. Analysis
Develop
2. Qualitative Testing
3. Quantitative Testing Demonstrate and Validate

• Define the “severe customer” or “severe environment” and use this to

quantify “one-life of damage” for each failure mechanism.
• Develop and Demonstrate the needed level of reliability relative
to the “severe level of damage” for each critical failure mechanism for
each device. Combine statistical rigor with engineering physics.
• You should define the requirements and the tests to be conducted by
your suppliers… do not assume that your supplier is the expert and
will do your job for you.
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 2
 Starting Point - The Three “Faces”
of Reliability
• We see “Reliability” as three phenomena that are
often lumped together to form a “bathtub” curve:

Quality Issues – Reliability

Issues – Failure
Failure Rate

Failure Rate
Improves With Rate Gets
Time Worse With
Time

Random Chance Failure

We Control
This With
SPC Or
Screening We Must Engineer
Stress Exceeds Strength – An Umbrella Robustness Into The
May Help But May Not Be Sufficient… Product To Move This
Difficult To Engineer A Solution To The To The Right
Un-Predictable
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 3
 The Problem Addressed by The
Reliability Engineer is…
• The product starts to wear out as soon as you get the
quality issues sorted out….not fair to the customer!
Failure Rate

We Must Engineer
Robustness Into The Product
To Move The Wear Out
Curve To The Right
Random Chance Failure

Lets Focus On When Do I Get

The Chance To
Moving The Wear See The Falling
Out Curve To The Piano?
Right!

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 4

 The Basic Concept of Stress-Strength
Interference
Reliability and Stress- Analytical Qualitative Quantitative
Targeting The Activities Activities Activities
Failure Strength
Severe User
Mechanisms Interference “A” “Q” “Q”

The World Valid for Product

Of Stress Cumulative and Strength
Non-Cumulative
Damage
Beginning
Of One
Life

The World Product Degrades From Fatigue or

Similar
Of Stress I Am The
Extreme
User “One Life”
= 20 Yrs. Valid for
End Of
One
Cumulative
Life Damage

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 5

 How Much Overlap is There When We
Demonstrate Reliability For The Average User?
Reliability is good for the below average
user…. but not for anyone more severe
Legend than average.
Distribution of
Usage Stress
Not Good!
Severe User
Average User
Nominal
Design Point
and Variation
in Product
Strength

Unreliability for the Average User

Stress or Strength
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 6
 Do Not Use MTBF or Cookbook Reliability
Prediction Methods
• MTBF by definition is when 63% of the product has failed
and 37% of the product has not yet failed.
• MTBF is useful for light bulbs but not for wind turbines.
Probability Density Function
• Shown is a typical 0.009 Pdf

“Electrical” failure Data 1

Weibull-2P
RRY SRM MED FM
F=1/S=0

distribution with Pdf L ine

0.007
MTBF = 100.
• The automotive
0.005
industry never uses
f(t)

MTBF.
0.004
• The automotive
industry never uses 63%
cookbook reliability 0.002

prediction methods
L arry Edson

(Mil-Std 217). L arry Edson Consulting

8/31/2009
9:47:48 AM
0.000
0. 000 80. 000 160.000 240. 000 320. 000 400. 000
T ime, (t)
β=1 .8 00 0 , η=1 00 .3 20 8

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 7

 Assessing Reliability of Existing
Wind Turbines
Information taken from the following will be identified with an *

Incremental
Failure Rate*

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 8

 Field Data* Shown As An Entropy
Based Cumulative Hazard Plot
Note: The Cumulative Hazard Plot is appropriate for a “repairable system”
where the failure rate could exceed 100%. The trending of the lines can be
modeled and extrapolated as needed for future projections.

Increasing
Failure Rate

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 9

Failures by Sub-System In The Current
Products*

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 10

 Average Downtime Per Failure by
Sub-System In The Current Products*

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 11

 OK… So We See That Things Could be
Better…and Electrical is a Big Contributor
• Strategy of Solution:
– You cannot jump from today’s biggest problem to
tomorrow’s biggest problem as an effective strategy.
– You must have good reliability requirements for all of
the components in a wind turbine as they are all
interconnected.
– Electrical is a major contributor
• Electrical devices fail for mechanical reasons.
• The Electrical Engineering community has historically
not been trained in dealing with mechanical issues.
– Reliability Apportionment, AQQ, and Accelerated
Testing for “Quick Learning Cycles” is a good
strategy……
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 12
 We Look to One of The Greatest Engineers
In History For Guidance – Leonardo Da Vinci
• What would Da Vinci tell us to do? (Draw The Big Picture)

From the book: “How To Think Like Leonardo da Vinci” by Michael J. Gelb
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 13
 The Big Picture of Tonight's “Dinner”

From the book: “How To Think Like Leonardo da Vinci” by Michael J. Gelb

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 14

 The Reliability Engineer Responds With
“Reliability Allocation”
Complete Wind Turbine With Reliability = 5% (3 Failures In 20 Years)

Blades & Generator Gearbox System Converter Nacelle Tower Foundation

Rotor 56% 56% Electrical 56% 95% 99.99% 99.99%
56% 56%

Blades Hub Slip Ring Speed Sensor Pitch System

99% 85% 90% 90% 90% *Thermal Fatigue
98.9%

*Vibration
98.9%
Pitch Motor Pitch Gearbox Pitch Bearings Pitch Bearing Lube System *Pitch Controller
98% 98% 98% 98% Box 98%

PLC Drive *Power Battery

99..5% 99.5% Supply Charger
99.5% 99.5%
*These are the two Reliability Tests on
this electronic device
(thermal cycling and vibration).
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 15
 Improved Reliability At The System Level

Complete Wind Turbine

R = 63.7% at 20 years (1/2 of a Failure In 20 Years)

Blades & Generator Gearbox System Converter High Nacelle Tower Foundation
Rotor 90% 90% Electrical 90% Voltage Cover & 99.99% 99.99%
and Main 90% Transformer Structure
Shaft 99% 99%
99%

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 16

 OK…So Now That I Have A Reliability
Requirement For a Device…What’s Next?

• We turn to one of the world’s greatest quality leaders…

What would Dr. W. Edwards Deming tell us to do?

“By What Method Will You

Make Your Engineering ‘Wants’
Become Reality?”

• “Just saying the words won’t make it happen.”

• We need methods and processes by which we
“engineer reliability into the product.”
• “Quick Learning Cycles” must be the “religion” of
thought during product development.
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 17
 By What Method?....AQQ
• Identify critical failure mechanisms
• “Design In” adequate design margins during the Analytical Step.
• Remove un-wanted surprises with Qualitative Testing
• Validate the existence of adequate design margins with
accelerated testing relative to the severe user or severe
environment.
tive
hase alita e
is P u tiv
lys se= Q ti ta
na Ph a
an
Underestimation A en
t
= Qu
pm o se
ha
evel n P
D atio
lid
Weakness
Va
Weakness

AQ2 as a Strategy
Squeeze Out Time
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 18
 1. Analysis – Selecting The Design Life Target
That Will Result In My Desired Reliability?

99.90
Weibull Probability Plot
Weibull
Slope = 3 Data 1
Unreliability, F(t)

P=2, A=RRX-S
F=1 | S=0

Slope = 1.9

50.00 50%
If my life
requirement is
507 thermal
10.00 cycles then I
need to design
5.00 to at least
97% 1500 thermal
(Example) cycles
1.00
1.00 ≅3 ≅5 10.00

Life Lives Lives

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 19
 Thermal Cycling Example (part 1)
• We have a requirement for this device to withstand 507
thermal cycles with a reliability of 97%. We will assume
a Weibull Slope value of 3. We want to design our joint to
last for 1500 thermal cycles. Will our intended design
pass this requirement?
L=.75 inches
Solder Joint Solder Joint
Ceramic Semiconductor CTE =α1

Epoxy-Fiberglass FR-4 Circuit Board CTE = α2

Modulus of Elasticity For Y = ∆D = ( α 2 − α1) × L × ∆T

Fiberglass

Ep = 2 x 106 lb/in2 = K
CTE Ceramic = 6.7x10-6 (α1)
CTE Fiberglass = 17.5x10-6 (α2)
∆T = -40 to +85 = 1250C
10/6/2009
( )
Y = ∆D = 17.5 × 10 −6 − 6.7−6 × .75 × 125 = .001
P =K×Y (spring equation)
P = 2 × 106 × .001 = 2000 lbs / in2
2000 lbs/in2 is pulling the solder joints apart
Copyright 2009 - Larry Edson Consulting Inc. 20
 Thermal Cycling Example (part 2)
.25 inches wide
L=.75 inches .125 inches tall
Ceramic Semiconductor CTE =α1

Epoxy-Fiberglass FR-4 Circuit Board CTE = α2

• Area of soldered end surface:

Area = height × width

Area = .125 × .25 = .03125 inches2

lbs. 2
Force = 2000 2
× .03125 inches = 62.5 lbs
inch
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 21
 Thermal Cycling Example (part 3)
• The 2000 lbs/in2 is applied through the two solder
joints, thus each solder joint experiences this force.
The equation for the Life-Stress for Solder
life-stress line:
m
N2  S1 
= 
N1  S2 
2.5
80000  2000 
= 
N1  200 
80,000
2.5
= N1 = 253 Cycles
 2000 
 
 20 0  Our current design will fail in 253 cycles and we need it to last for
1500 thermal cycles. We need to redesign this joint. We have
not wasted our money by prototyping a joint that would
fail!
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 22
 2. Qualitative Testing - Used to Confirm Failure
Modes and Remove Outlier Weaknesses
HALT Testing is a good example of Qualitative Testing

Conducted As An Accelerated Test To Produce A Quick Learning Cycle

Limit of
Patch It Up And Keep Technology Equals
Testing Design Maturity
Delamination Strong
Weak

Broken Seal
Time on Test

Intended Design
Errors That Crept Into The Design Margins By
Analysis
If you do not fix these then you waste
the money you spent to get these

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 23

 3. Quantitative Testing – Connecting
Statistical Rigor to Engineering Physics
The Five Critical Parameters That Pin-down The Reliability Requirement

R C Stress Time T Definition

Condition of Failure
and
•Probability •Confidence • Time after Success
of proper in the above •Failure
Mechanism
which the
function at statement R&C
the end of apply •Technical
time T •50% or •Vibration failure
•70% or •Thermal
Fatigue • “One Life” •Perceived
•95% or •90% • 10 years failure
•99%
• 20 years

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 24

 Road Map of Reliability Statistical
Methods
Failures Are Required! No Failures Permitted!

Testing To Failure Success-Run

Test To Failure
At Normal Normal
Stress With Sample Size
Weibull Plotting and Normal
Test Duration
Sudden Death Testing
Reduced
Sample Size
CALT With
Not Extended Test
Traditional
Degradation Duration
Analysis

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 25

 A Wind Turbine Vibration Example

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 26

 Quantify 20 Years Worth of Damage

• The damaging energy is obtained by integration under the

blue curve = .45 Grms
• This level of vibration will occur for 175,200 hours
• The fatigue exponent used is based upon the material being
stressed (a value of 5 is used for steel)
• The “damage” from this is calculated below:
m
Damage = N × ( Stress )
5
3191.27 = 175,200 × (.45 )

• An “equivalent-damage test” is 8 hours at 3.31 Grms:

5
3191.27 = 8 × ( 3.31)
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 27
Applying Success-Run Statistics to Our
Vibration Example
• Using Success-Run and Weibayes statistics, we need to test:
– 6 parts to 1.56 “lives or 12.48 hours at 3.31 Grms
or
– 1 part to 2.83 “lives” or 22.67 hours at 3.31 Grms

10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 28

 Mechanical Cycling Example – CALT Testing to
Quantify Reliability and The Fatigue Exponent
1.00E+5 IPL/Weibull
Data 1
Time
Eta
(hours)
Hours to
10000.00 Failure Force

Life Test Two or More

Requirement m
1000.00
Available
Time Test Two
CALT
Calibrated
Accelerated
Third Stress Level Test Two Life
Identified Test
10.00
100.00 300.00
Stress Foolish Stress Level
Normal Stress Level
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 29
All Data Points Translated to Normal Stress and
Plotted on Weibull Paper
99.00
90.00
IPL/Weibull Collapsing
Data 1 our 2+2+2
50.00 100 data points
F=6 | S=0
into a single
plot of 6
10.00
5.00
data points
by
transposing
Unreliability

1.00
the failures
0.50
at higher
stress to
0.10
the normal
0.05 Requirement stress level.
0.01
5.00E-3

m
1.00E-3  Stress accelerated 
5.00E-4 Life normal stress = Data Valueaccelerated × 
 Stress normal 
1.00E-4
100.00 1,000.00 10,000. 100,000.
Beta=3.9678, K=1.1493E-14, m=4.9162
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 30
 Summary – Coaching The Supplier
• “Reliability” is just another form of engineering made
consistent through a process. No magic here.
• The AQQ process is well proven and effective…most
people only make use of the last letter.
• Learn to focus on the concept of defining “one-life.”
• Focus on the “severe user” or “severe environment.”
• Design accelerated tests based upon the concept of
“Damage equals Damage.”
• Apply the right kind of statistical methods to best fit
the needs of the situation.
• Reliability Apportionment can help rationalize the
reliability requirements being used (sanity check).
10/6/2009 Copyright 2009 - Larry Edson Consulting Inc. 31