8/18/2019 Quality Tools - Yanez
1/49
Measuring System Analysis
Case Study for the Automotive
Industry
Alberto A. Yáñez-Moreno
TMAC/UTA ASQ Houston 2013
8/18/2019 Quality Tools - Yanez
2/49
Learning Objectives
• Understand why we measure
• MSA basic concepts
•
Components of measurement errors
•
Understand how to conduct a MSA
• Data analysis theory
•
Automotive case studies
8/18/2019 Quality Tools - Yanez
3/49
!
Measurement Systems
•
Measurement systems are like eyeglasses, when thelenses are incorrect, the vision is blurred.
• A measurement system allows us to “see” the process.
When a measurement system is poor, we lose the ability to
make good decisions about how to improve the process• In the Measure Phase of the DMAIC process, the MSA
should be conducted on the “Y” or KPOV
• In the Control Phase of the DMAIC process, the MSA
should be conducted on the critical few “X’s” or KPIV
8/18/2019 Quality Tools - Yanez
4/49
#
Basic Concepts
•
Every process produces a “product” or “service”•
Every product or service possesses requirements
• Every requirement can be measured
• The total observed variation is equal to the real product
variation plus the variation due to the measurement systemNote: We want most of the variation coming from theProduct/Part and very little coming from the Measuring
System
Total
2 Process/Part/
Service
2 2 Measurement System
8/18/2019 Quality Tools - Yanez
5/49
$
How might measurementvariation affect these decisions?
What if the amount ofmeasurement variation is
unknown?
Process
Measurement
Process
Measurement
Measurement variation can make our processcapabilities appear worse/better than they are
Why Worry About Measurement
Variation?Consider the reasons why we measure:
Verify product/process conformity tospecifications
Assist in continuous improvementactivities
8/18/2019 Quality Tools - Yanez
6/49
%
Components of Measurement ErrorThe sources of variability for the measurement systemcontaining continuous data are:
8/18/2019 Quality Tools - Yanez
7/49
&
Measurement Properties to Study
•
The two most common key measures associated with ameasurement system are accuracy and precision
• Accuracy and precision are different, independent
properties. You may encounter a data set that is accurate,yet not precise or precise, yet inaccurate
• Sometimes you may encounter a data set that is neither accurate nor precise. Obviously, we desire to have data
that exhibits both properties
8/18/2019 Quality Tools - Yanez
8/49
'
Measurement Properties to Study
•
Not only do we want our measurement systems to beaccurate and precise, but!
–
We want the measurement system to be able to detect smallchanges to the process (good discrimination)
–
When applied to the same items of interest, the measurementsystem should produce the same results in the future that it did inthe past (stability)
–
We want the system to be linear. Linearity concerns the behaviorof the measurement process across a wide spectrum of
applications
8/18/2019 Quality Tools - Yanez
9/49
())*+,)- ./0 1+2)3/345 677*/8+,829
:
Desired
Desired
Current
Current
USL
USLLSL
LSL
Target
Target
())*+,)-
1+2)3/345
8/18/2019 Quality Tools - Yanez
10/49
;<
Accuracy and Bias
•
!""#$%"& 3/ 8=2 2>8258 84 ?=3)= 8=2 %'($%)( 4@ 8=2A2,/*+2A258/ 92.3,82 @+4A 8=2 *$#( '%,#(0 65 /3AB72 82+A/C 38
,99+2//2/ 8=2 D*2/E45C !"# %&'(%)'* +, - )'. ./' 0(1)/.2
%#34'(56 6@ 8=2 ,5/?2+ 3/ -2/C 8=25 ?2 /,- 8=,8 8=2
A2,/*+2A258 /-/82A 3/ ,))*+,820 6@ 8=2 ,5/?2+ 3/ 54C 8=25 ?2
=,.2 ,5 35,))*+,82 A2,/*+2A258 /-/82A• -.%/ 3/ 8=2 82+A F3.25 84 8=2 0./*%1"( G28?225 8=2 4G/2+.29
,.2+,F2 A2,/*+2A258 ,59 *2( *$#( '%,#(C 4+ H+3F=8I ,5/?2+
8/18/2019 Quality Tools - Yanez
11/49
;;
Accurate vs. Inaccurate
Mean Mean
TrueValue
TrueValue
J3,/
8/18/2019 Quality Tools - Yanez
12/49
;K
Instrument 2
Instrument 1
Potential Bias Problems• Average of measurements are different by a fixed
amount. Consider the manufacture of first-article. Biaseffects include: –
Operator Bias – Different operators get detectable differentaverages for the same value
–
Instrument Bias – Different instruments get detectable differentaverages for the same measurement
–
Others – Day-to-day (environment), fixtures, customer andsupplier (sites)
8/18/2019 Quality Tools - Yanez
13/49
;!
Mean
Bias
TrueValue
1+2)3/345
•
In addition to accuracy, it is important for a measurementsystem to be precise
Precision Precision is the extent towhich we are able to get
the same data values
when independentmeasurements are made
on the same entity
Using the mean of
repeated measurements
improves precisionbecause the dispersion of
the averages is always
less than the dispersion of
the individual data points
8/18/2019 Quality Tools - Yanez
14/49
;#
RulerCaliper
Micrometer
.28
.279
.2794
.28
.282
.2822
.28
.282
.2819
.28
.279
.2791
Discrimination• Discrimination is the capability of detecting small changes in the characteristic
•
The instrument may not be appropriate to identify process variation or quantify
individual part characteristic values if the discrimination is unacceptable
• If an instrument does not allow differentiation between common variation in the
process and special cause variation, it is unsatisfactory
• A common cause to MSA failure can be attributed to rounding up or down
measurements
8/18/2019 Quality Tools - Yanez
15/49
;$
Time One
Time Two
Stability
•
If measurements do not change or drift over time, theinstrument is considered to be stable
8/18/2019 Quality Tools - Yanez
16/49
;%
Linearity• A measure of the difference in bias (or offset) over the range of
the sample characteristic the instrument is expected to seedetermines linearity. If the bias is constant over the range ofmeasurements, then linearity is good
• Over what range of values for a given characteristic can thedevice be used? –
When the measurement equipment is used to measure a wide range of values, linearity is a
concern
Measurement Scale
LowEnd
HighEnd
Measurement Variation
8/18/2019 Quality Tools - Yanez
17/49
;&
Basic Model of Precision
222
rpd rpt MS ! ! ! +=
The Measurement System Variation is equal to the variation due toRepeatability plus the variation due to the Reproducibility and represents
the common cause variation in the Measurement System
2 Repeatability
2 2 Reproducibility Measurement
System
8/18/2019 Quality Tools - Yanez
18/49
;'
Repeatability (Gage precision)
•
Repeatability is the inherent variability of themeasurement system. Used as an estimate of short term
variation. It is the variation that occurs when successive
measurements are made under the same conditions:
–
Same part –
Same characteristic
–
Same person
–
Same instrument
–
Same set-up
–
Same environmental conditions
Repeatability: It takes one to repeat
8/18/2019 Quality Tools - Yanez
19/49
;:
Reproducibility (operators’ precision)
• Reproducibility is the variation in the average of themeasurements made by different operators using the
same measuring instrument when measuring the identical
characteristic on the same part
Operator A
Operator B Operator C
2
rpd
!
Reproducibility: It takes two to reproduce
8/18/2019 Quality Tools - Yanez
20/49
K<
% Repeatability and Reproducibility
Observed Part/Process Variation
MeasurementSystem Variation%R&R = 20%
%R&R = 75%
%R&R = 100% Rule of Thumb:We want
%R&R < 30%
8/18/2019 Quality Tools - Yanez
21/49
K;
Capability of the Measurement
System
• Once precision has been quantified, we now need toassess if the measurement system is precise enough. Todo this we must compare the measurement variation tothe production process. There are many capability indicesbut we will be focusing on the following two indices: –
% R&R
–
Discrimination Index
8/18/2019 Quality Tools - Yanez
22/49
KK
Gage R&R: % R&R
• Assesses what percent of the Total Variation is taken up by
measurement error
•
Includes both repeatability and reproducibility
Care must be taken to use samples representing full, but typical,process variation
100ˆ
ˆ& % !=
Total
MS
R R "
" L/*,77- 2>B+2//29,/ B2+)258
8/18/2019 Quality Tools - Yanez
23/49
K!
Acceptability Summary
% Contribution
1%
10%
% Study
Variation
10%
30%
% Tolerance
10%
30%
Number of
Distinct
Categories
10
5
Desirable to Have All 4 Indicators Say “Go”
“Product
Control”
“Process
Control”
8/18/2019 Quality Tools - Yanez
24/49
K#
Gage R&R•
The objective of a Gage R&R study is to learn as much as possible
about the measurement process in a short amount of time
•
The strategy is to include equipment, operators, parts, and other factors
that will usually be elements of the measurement process
•
Make careful selection of parts (or samples) representing the entire
range of process variation. (Good and Bad to the entire specification)
•
The parts should be labeled in such a way to preclude operator
identification and therefore remove possible operator bias
•
Each part will then be measured multiple times in random sequence byeach operator using the same equipment. This can be replicated for
each equipment set
•
Verify that an on-going calibration, maintenance, and metrology
program exists and is current on the measurement system
8/18/2019 Quality Tools - Yanez
25/49
K$
AIAG Gage R&R Standards
•
The Automotive Industry Action Group (AIAG) has tworecognized MSA standards for Gage R&R: – Short Form – Five parts measured two times by two
different operators
– Long Form – Ten parts measured three times each by three
different operators• For additional insights into Gage R&R, go to
www.aiag.org
• Remember that the Measurement System is acceptable if
the Gage R&R variability is small compared to theProcess or Study Variation seen
8/18/2019 Quality Tools - Yanez
26/49
K%
Running the Gage R&R
•
Each sample should be measured 2 - 3 times by each operator(2 Times is the Short Test)
• Make sure the parts are marked for ease of data collection butremain “blind” (unidentifiable) to the operators
• Be there for the study. Watch for unplanned influences
• Randomize the parts continuously during the study to precludeoperators influencing the test
• The first time evaluating a given measurement process, let theprocess run as it would normally run
• Enter data in appropriate software like: Minitab, QI Macros,SigmaXL, Engine Room, Jump, etc.
8/18/2019 Quality Tools - Yanez
27/49
K&
Minitab Gage R&R “Sixpack”
Let’s look at these sixcharts one at a time
8/18/2019 Quality Tools - Yanez
28/49
K'
Gage R&R Relationships• A measurement process is said to be consistent when the results for
operators are Repeatable and the results between operators areReproducible
•
A gage is valid to detect part-to-part variation when the variability of operator
measurements is small relative to process variability or the tolerance range
• The percent of process variation consumed by the measurement (% R&R) is
then determined once the measurement process is consistent and can detectpart-to-part variation
8/18/2019 Quality Tools - Yanez
29/49
K:
2. R Chart by Operator•
Repeatability is checked by using a special Range Chart where the
differences in the measurements by each operator on each part ischarted. If the difference between the largest value of a measured part
and the smallest value of the same part does not exceed the UCL,
then that gage and operator are considered to be Repeatable
Repeatability is indicated when virtually all of the range points lieunder the upper control limit on the range chart. Any points that
fall above the limit need to be investigated
8/18/2019 Quality Tools - Yanez
30/49
!<
3. Xbar Chart by Operator
•
It is desirable to see plots that consistently go outside the UCL and LCL becauselimits are determined by gage variance and these plots should show that gage
variance is much smaller than variability between the parts
• If the samples chosen do not represent the total variability of the process, the
gage (repeatability) variance may be larger than the part variance and invalidate
the distinct categories calculation
• If the patterns of the operators are not comparable, there may be significant
operator and part interactions (discussed on another slide)
On this chart you want At Least 50% of the points to be
Outside the Control Limits
8/18/2019 Quality Tools - Yanez
31/49
!;
4. Response by Part
•
This graph shows the data for all ten parts for all operators plotted
together. It should show plots that vary from the smallest dimensions
for the parts made by the process to the largest dimensions for thesame parts. Parts should be both in tolerance and out of tolerance if the
process makes them
•
If a part shows a large spread, it may be a poor candidate for the test
because the feature may not be clear on that part
8/18/2019 Quality Tools - Yanez
32/49
!K
5. Response by Operator
•
This graph shows the data for all ten parts for plotted by each operator.
The line connecting the averages (shown as the circle with crosshairs)
of all 10 parts measured by each operator should be horizontal
•
Any significant slope is an indication that this operator has a general
bias to measure large or small when compared to the other operators
8/18/2019 Quality Tools - Yanez
33/49
!!
6. Operator * Part Interaction
• Operator Influence: If the lines connecting the plottedaverage points diverge significantly, then there is a
relationship between the operator making the
measurements and the part that the operator is measuring.
This is not good and needs to be investigated
8/18/2019 Quality Tools - Yanez
34/49
!#
Minitab Sixpack
Questions on the graphical output?
8/18/2019 Quality Tools - Yanez
35/49
For %Contribution the guidelines are different:=10% : Unacceptable
!$
Gage R&R: The Numerical Output
The first column represents the source of variation, the second column is an estimateof the actual variation for each source (factor). The third column is the linear % that
each represents of the total variation. It is depicted as the red bar (left-most) on thePareto in the six-pack diagram
8/18/2019 Quality Tools - Yanez
36/49
!%
Gage R&R: The Numerical Output
• This tabulation from Minitab builds the % of Study Variation that each source contributes to a calculatedpotential Total Variation seen in the study. The 6.0 * SD is how statistically 99.73% of the Total Variation
is calculated and this is assumed to equal 99.73% of the true process variation unless the Historical
Sigma is input
•
The %’s are used to grade the validity of the measurement system to perform measurement analysisusing %’s already taught. If the process is performing well, the % Tolerance is then important. The sum
of the %’s may add to more than 100% due to the math
• The Number of Distinct Categories represents the number of non-overlapping measurement groups that
this measurement system can reliably distinguish in the Study Variation. We would like that number to be
5 or higher. Four is marginal. Fewer than 4 implies that the measurement system can only work withattribute data
M=2/2 /=4*79 ,77G2
7'33 8/%# 9:;
M=3/ .,7*2 /=4*79 G2
< ,( =,('
8/18/2019 Quality Tools - Yanez
37/49
37
Case Study # 1•
Selection of parts to be tested
–
10 parts randomly selected – Good, bad & average parts used
• Instructions to operators – Perform tests as normal
– Report variance
•
Tests conducted –
10 parts tested
–
2 operators
– 3 checks on each of the 10 parts
given to the operator in random order
8/18/2019 Quality Tools - Yanez
38/49
NO( M+22
!'
NO(
1,+8 ; 1,+8 K 1,+8 ! 1,+8 # 1,+8 $ 1,+8 % 1,+8 & 1,+8 ' 1,+8 : 1,+8 ;<
PB2+,84+
;
PB2+,84+
K
Q2B2,8 ; Q2B2,8 K Q2B2,8 !
PB2+,84+
;
PB2+,84+
K
Q2B2,8 ; Q2B2,8 K Q2B2,8 !
8/18/2019 Quality Tools - Yanez
39/49
MSA – Gage R&R for Value
39
8/18/2019 Quality Tools - Yanez
40/49
MSA Session - Gage R&R
40
8/18/2019 Quality Tools - Yanez
41/49
#;
Case Study # 2
MSA Bolt Torque
8/18/2019 Quality Tools - Yanez
42/49
#K
Electronic Torque Wrench
Bolt Torque Wrench Operator
8/18/2019 Quality Tools - Yanez
43/49
#!
!"#$%&'() +,$-". /(#0,$'$1&"" +-&%2-%&"
!+/
3#&- 4 3#&- 5 3#&- 6 3#&- 7 3#&- 8 3#&- 9 3#&- : 3#&- ; 3#&- < 3#&-
44
=>"-?&
4
=>"-?&
5
+#.>0"
4
+#.>0"
5
3#&-
45
3#&-
46
3#&-
47
3#&-
48
3#&-
49
3#&-
4@
+#.>0"
4
+#.>0"
5
8/18/2019 Quality Tools - Yanez
44/49
##
Minitab Data
8/18/2019 Quality Tools - Yanez
45/49
#$
Minitab Data Session Data
8/18/2019 Quality Tools - Yanez
46/49
#%
Minitab Data
8/18/2019 Quality Tools - Yanez
47/49
#&
Minitab Data
8/18/2019 Quality Tools - Yanez
48/49
#'
Minitab Data
8/18/2019 Quality Tools - Yanez
49/49
R S ( O2//345