Questions and Answers
How Relevant Questions
Obtain
Useful Answers
Judson B. Estes
Fiat Chrysler Automobiles
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 1
Focus on the 4 Cs
Weave the wisdom from many available tool sets into a package of
training, certification and project work
1. Collect
Currently available facts relevant to problem.
Listen for what is already known and suspected.
Communicate to entire team the current facts to get on the same page.
2. Contrast
A Measurable difference in performance.
How do you measure the performance?
How Big is the difference?
3. Converge
Use Logical Strategies to isolate the candidate cause.
What split are you making?
How does that narrow the possible causes?
4. Confirm
Test the candidate cause to prove it is the true root cause.
What is your Statistical Confidence?
When can we implement the fix?
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 2
Collect Phase
Describe Problem
Identify Possible Causes
Evaluate Possible Measurements
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 3
Collect Phase
Describe the Problem
State the Problem naming the deviation for
which you want to find the cause
To help stay on track, ask:
What object (or group of objects) has the deviation?
What deviation does it have?
What do we see, feel, hear, taste, or smell that tells us there is a
deviation?
Write a short statement in Object/Deviation format
Use one object and one deviation
Be specific, separate if needed
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 4
Collect Phase
Specify the Problem
Describe the deviation
factually to increase
understanding of the
deviation
Ask questions in 4 areas:
WHATIdentity
IS IS NOT
WHERELocation
WHENTiming
Describe the Tighten IS data.
problem in Help eliminate
EXTENTSize detail. possible causes.
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 5
Collect Phase Example
PROBLEM SOLVING
[1] PROBLEM AREA
Problem Statement: LX low beam bulb infant failure
Description PROBLEM ARPEA NON-PROBLEM AREA
facts (IS observed / reported) (IS NOT observed/reported)
WHAT:
1. Object low beam bulb # L0009006 Magnums and 300C's. Also WK, PT use
used in LX 300 models same Pt # low beam bulb, All other bulbs
LXCH48, LXCP48, LXFP48 . in the Click here to see ID of vehicle,
Click here to see VIN list subassemblies, bulb and field warranty
performance.
2. Defect LOP 085032 Needs Other potential lab conclusions could have
replacement Some examined been but were not "cold shock", "high
from Field, Hot shock is the voltage" & "wear out"
main conclusion from the
Sylvania lab report
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 6
I often say that when you can measure
what you are speaking about, and express it
in numbers, you know something about it;
but when you cannot measure it, when you
cannot express it in numbers, your
knowledge is of a meagre and unsatisfactory
kind;
-Lord Kelvin
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 7
Contrast Phase
Required
Current
Frequency
0 1 Length (mm) 4
Required
Frequency
Current
-3 -1.5 0 1.5 3
Flushness (mm)
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 8
Contrast Phase
Open seams on DR
seats resulting in high Problem Definition Statement
warranty costs
Cloth seats Leather seats 90% of returns are leather
Front seat Rear seat
83% of returns are front seats
Driver side Passenger side 71% of returns are driver side seats
Seat back Seat cushion 92% of returns and narratives are seat cushions
Examination of returned product shows
Seam A Seam B
seam B accounts for 42 out of 54 claims
BOB/WOW seam Other
WOW cushion strategies See Strategy Diagram
Frequency
Required
Current Find and eliminate the Red X
causing open seams on the DR
front driver side leather seats
0 1 Width (mm) 4
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 9
Contrast Phase
What is a BOB and a WOW?
Best of the Best and Worst of the Worst
Not necessarily a good and bad part but really
parts that are as different as possible in the
way they effect the Customer.
We are looking for contrast in order to see
differences
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 10
Contrast Phase example
Measure twice and
get the same answer
on BOB and WOW
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 11
Converge Phase
Once we make sure our measurement system isnt
fooling us then we start generating clues
We then use certain tools to begin to converge on
the Red X candidate
Concentration diagram
Component search Stage 1 and 2
Operation Search
Paired and Group comparisons
Event to Energy transform
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 12
Reactive Problem Solving Hierarchy
Use the Right Tool for the Problem
Innovation and Evolution
TRIZ and
Systemology
No Strategy and All Tools
Pure Statistics
Simple Strategy and Most Tools
Six Sigma
Classical and Taguchi Multiple Variables and
Interactions
Design of Experiments
Shainin Red X Strategies Multiple Strategies, Easy Statistics
IS/ IS Not Problem Specifications Distinctions and Changes
Ishikawa Fishbone Diagram Organized Brainstorming
5 Whys to the Root Cause Simple Questioning
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 13
Problem Solving Hierarchy
Least Widely Used Hardest to Grasp
TRIZ and
Systemology
Increased
Pure Statistics
More Variables Variation and
and Interactions Environment
Six Sigma Changes
Classical and Taguchi
Design of Experiments
Shainin
Critical Thinking
Ishikawa Fishbone Diagrams
Most Widely 5 Whys to the Root Cause Easiest to
Used Grasp
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 14
Concentration Diagram example
1 2 3 4 5 6 7 8
A
xx
xx
B x
x
C xxx
xx
D xxx
xx
xxx
xxxx
E
x
F
Paint Craters on B pillar
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 15
Component Search Stage 2
.
Plotting Stage 1 Stage 2
30
WOW
* + *
Green Y = lbs.
* *
20
+
10
+ + +
* +
BOB
0
Orig. 1st D/R 2nd D/R 3rd D/R S1 orig
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 16
Confirmation Phase
Once we identify the Red X candidate it is now time
to use statistics to confirm our candidate.
Some tools types that are used for this:
Six Pack B vs. C
Tukey test
Barrier B vs. C
Spike B vs. C
5 Penny test
Factorial Experiment (DOE)
Binomial probability
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 17
Six Pack B vs. C
B C
B is the Better part or C is the Current part or
process or sub-assembly or process or sub-assembly or
material material
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 18
Six Pack B vs. C Example 1
The required confidence level is 95%, which therefore requires a
sample size of 3 Bs and 3 Cs and an end count of 6.
Run Order B or C Diameter
(mm)
1 C 10.6
2 B 8.3
3 C 11.2
Rank Order 4 C 9.8
5 B 9.1
B 8.3
6 B 8.8
B 8.8
B 9.1 The end count equals 6. Therefore, it
C 9.8 can be stated with 95% confidence that
C 10.6 the Bs are better than the Cs.
C 11.2
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 19
Six Pack B vs. C Test Example
Distribution of two groups looks something like this.
B B B C C C
8.0 8.5 9.0 9.5 10 10.5 11 11.5
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 20
Reliability
by Design
Confidential and Proprietary to Fiat Chrysler
March 2014 Automobiles
Reliability
Prediction of Verification of
Performance Performance
Improvement of Prediction
The best Prediction methods are quantative.
The best Verification is actual parts and systems in real usage.
The best Improvement eliminates all discrepancy between prediction and reality.
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 22
Deterministic Design
Design parameters are deterministic, i.e., they have unique values
CTQs are also deterministic, and are calculated as functions of the design
variables by transfer functions, Y = f (X1, X2, , XN)
Design CTQs (Ys)
Parameters
X1 (Xs)
Y1
Transfer Function .
X2 Y = f (X1, X2, XN) .
. .
XN YN
Most engineering design is deterministic
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 23
Statistical Design
Design parameters are statistical in nature, with mean values and variation
(e.g., standard deviation)
CTQ variations determined by statistical analysis (e.g., Monte Carlo), using the
transfer function and statistical variations in design parameters
Design CTQs (Ys)
Parameters
X1 (Xs)
Y1
Transfer Function .
X2 Y = f (X1, X2, XN) .
. .
XN YN
Noise
Parameters XN1 . . XNn
DFSS uses statistical design to understand and control variation
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 24
Statistical Design
Why Prototyping Doesnt Reveal Problems
Prototyping: Single Product Copy Reality: Multiple Product Copies
X1 LSL USL X1 LSL USL
X2 Y=f(X1,X2...) X2 Y=f(X1,X2...)
Y Defects Y
Xn Xn
Input Realistic
Selected variability Range of distribution
prototype not captured, possible of product
inputs defects masked inputs Y (CTQ)
Prototyping does not verify product robustness
It assesses functionality of a single, often hand-selected, sample
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 25
Statistical Design
Mechanical Example: Simply Supported Box Beam
Performance Requirements: Baseline Design
Applied load: 200 kg/m over 1.5 m
Overhang = LT-L1 = 4.5 m t
h WP
Design margin must be positive,
i.e., yield strength > max stress w P
6 quality F
Low cost
L1
LT
Analysis: Transfer function
Margin = Yield strength - Max stress
= Yield strength - (Max stress from tensile load + Max stress from bending)
F 3hPWp (2LT - 2L1 - Wp)
Margin = Sy - ____________ - ____________________
2ht + 2wt - 4t2 wh3 - (w - 2t) (h - 2t)3
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 26
Statistical Design
Deterministic Design of Beam
F 3hPWp (2LT - 2L1 - Wp)
Analysis: Margin = Sy - ____________ - ____________________
2ht + 2wt - 4t2 wh3 - (w - 2t) (h - 2t)3
Choose values for Design Parameter/Load Value
design parameters Beam length, LT (m) 12
and applied loads: Support length, L1 (m) 7.5
Beam height, h (m) 0.75
Beam width, w (m) 0.25
Section thickness, t (m) 0.05
2
Yield strength, Sy (kg/m ) 89,600
Uniform load density, P (kg/m) 200
Uniform load width, Wp (m) 1.5
Tensile load, F (kg) 100
Substituting: Margin = + 9,726 kg/m2
Baseline design meets positive design margin requirement,
March 2014 but quality level unknown
Confidential and Proprietary to Fiat Chrysler Automobiles 27
Statistical Design
Simply Supported Box Beam
F 3hPWp (2LT - 2L1 - Wp)
Analysis: Margin = Sy - ____________ - ____________________
2ht + 2wt - 4t2 wh3 - (w - 2t) (h - 2t)3
Design parameters & applied loads are statistical in nature
Choose mean values and a variability measure (e.g., std deviation)
Consider tolerances
Design Parameter/Load Mean Std Dev Tolerances
Lower Upper
Beam length, LT (m) 12 0.017 0.05 0.05
Support length, L1 (m) 7.5 0.013 0.04 0.04
Beam height, h (m) 0.75 0.0033 0.01 0.01
Beam width, w (m) 0.25 0.0033 0.01 0.01
Section thickness, t (m) 0.05 0.0025 0 0.01
2
Yield strength, Sy (kg/m ) 89,600 3,200 7,500 0
Uniform load density, P (kg/m) 200 3.3 5 5
Uniform load width, Wp (m) 1.5 0.07 0.2 0.2
Tensile load, F (kg) 100 1.65 5 5
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 28
Statistical Design
Simply Supported Box Beam
F 3hPWp (2LT - 2L1 - Wp)
Analysis: Margin = Sy - ____________ - ____________________
2ht + 2wt - 4t2 wh3 - (w - 2t) (h - 2t)3
Do a statistical analysis (e.g., Monte Carlo), using transfer function
and statistical parameter & load values
.040
Mean = 9,726
Results:
Probability
.030
Margin mean 9,726 kg/m2
.020
Margin std dev 5,466 kg/m2 Defects
Defect probability 3.8% .010
Design 3.3 .000
-5,000 0 10,000 20,000 30,000
Design Margin (kg/m2)
Design margin may be positive or negative!
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 29
Statistical Design
Reaching 6
Design optimization analysis:
Use transfer function to understand the shape of the response surface
and the design margins sensitivity to each design parameter
Reduce defects by shifting mean values or reducing variances of the
most sensitive design parameters
Sensitivities found by partial differentiation of transfer function and
evaluation at design point
Design Parameter/Load Mean Std Dev Sensitivity
Beam length, LT (m) 12 0.017 - 21,003 Margin most
Support length, L1 (m) 7.5 0.013 21,003 sensitive to t,
Beam height, h (m) 0.75 0.0033 180,205 with w and h next
Beam width, w (m) 0.25 0.0033 181,676
Section thickness, t (m) 0.05 0.0025 1,158,739
2
Yield strength, Sy (kg/m ) 89600 3200 1
Uniform load density, P (kg/m) 200 3.3 - 393.8
Uniform load width, Wp (m) 1.5 0.07 - 42,007
Tensile load, F (kg) 100 1.65 - 11.1
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 30
Statistical Design
Reaching 6
Improving the design margin:
In general, design can be improved by shifting means of the most
sensitive parameters or reducing their variabilities
Although t is the most sensitive parameter, we elect to shift the mean of w
(next most sensitive) because box beams come in only a few standard
thicknesses (the next thicker beam would be too costly and heavy)
Design margin results: .040
Probability
Beam width, w 0.25 0.30 0.35 .030
Mean 9,726 17,879 24,518 w = 0.25
.020
Std deviation 5,466 5,124 4,853 Defects
w = 0.30
Defect prob, % 3.8 0.024 0.00002 .010
Design 3.3 5.0 6.5 w = 0.35
.000
-5,000 0 10,000 20,00030,000
Design Margin (kg/m2)
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 31
Problem Solving or Problem Prevention
Discussion and Questions ??
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 32
Statistical Design
Electronics Example: Switching Power Supply
Vo = 5 Vdc, +/-5%
Vin = 85 - 275 Vac
Input Filter Isolated Switching
Converter
Performance Requirements
Output voltage, Vo: 5 V, +/-5% Feedback
Input voltage, Vin: 85 - 275 V
6 quality
Low cost
V
Baseline Design o
Isolated switching converter/ OPTO R2
feedback section
Baseline design combines Vref
PWM IC
power MOSFET & control circuit R1
CTRL
R1 I
in a 3-pin package
March 2014
b 33
Confidential and Proprietary to Fiat Chrysler Automobiles
Statistical Design
Deterministic Design of Power Supply
Vref
Analysis: Transfer function Vo = Vref + R2 ( ____ + Ib )
R1
Choose values for Design Parameter Value
design parameters:
LM 431I ref voltage, Vref (volts) 2.495
R1 (ohms) 10,000
R2 (ohms) 10,000
Bias current, Ib (amps) 5.0E-06
Substituting: Output voltage = 5.04 volts
Baseline design meets 5V, +/- 5% performance requirement,
but quality level unknown
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 34
Statistical Design
Switching Power Supply
Vref
Analysis: Transfer function
(unchanged)
Vo = Vref + R2 ( ____ + Ib )
R1
Design parameters are Design Parameter Mean Std Dev Tolerances
Lower
statistical in nature. Choose Upper
mean values and a variability LM 431I Vref (volts) 2.495 0.0283 0.085 0.085
measure (e.g., std deviation): R1 (ohms) 10,000 33.33 1% 1%
R2 (ohms) 10,000 33.33 1% 1%
Do a statistical analysis Bias current, Ib (amps) 5.0E-06 1.15E-06 2.00E-06 2.00E-06
(e.g., Monte Carlo), using the
.035
transfer function and the
.026
Probability
statistical parameter values
.017
Results:
.009
Vo mean 5.04 volts
Vo std dev 0.059 volts .000
4.75 4.875 5.00 5.125 5.25
Defects/million 188 (5.06) Volts
Baseline design meets 5V, +/- 5% performance
March 2014 requirement, but quality level is not 6
Confidential and Proprietary to Fiat Chrysler Automobiles 35
Statistical Design
Reaching 6
Design optimization analysis:
Use transfer function to understand the shape of the response surface
and the output voltages sensitivity to each design parameter
Reduce defect rate by shifting mean values or reducing variances of
design parameters
Design Parameter Mean Std Dev Sensitivity
LM 431I Vref (volts) 2.495 0.0283 2
R1 (ohms) 10,000 33.33 -0.0002495
R2 (ohms) 10,000 33.33 0.0002545
Bias current, Ib (amps) 5.0E-06 1.15E-06 10,000
Design Mod 1: Center distribution Base Centered
.038
by increasing R1 to 10,160 ohms
.028
Probability
Results:
.019
Vo mean 5.00 volts
.009
Vo std dev 0.058 volts
Defects/million 20 (5.61) .000
4.75 4.875 5.00 5.125 5.25
March 2014 Confidential and Proprietary to Fiat Chrysler Automobiles 36
Volts
Statistical Design
Reaching 6 (contd)
Design Mod 2: Mod 1 plus 0.1% Design Mod 3: Mod 2 plus LM 431AI
resistors to reduce resistor variance MOSFET to reduce Vref variance
Centered 0.1% Resistors Base 0.1% Resistors
MOSFET Upgrade
.038 .050
Probability
Probability
.028 .037
.019 .025
.009 .012
.000 .000
4.75 4.875 5.00 5.125 5.25 4.75 4.875 5.00 5.125 5.25
Volts Volts
Mean Std Dev DPMO Z ST Cost
Summary Baseline Design 5.04 0.059 189 5.06 100%
Mod 1: Centered via R 1 5.00 0.058 20 5.61 100%
Mod 2: 0.1% Resistors 5.00 0.057 13 5.7 101%
Mod 3: LM 431AI 5.00 0.041 ~0 7.58 105%
Statistical design enables performance, quality
March 2014 & cost prediction during the design process
Confidential and Proprietary to Fiat Chrysler Automobiles 37