Michael J. White
Master Black Belt NA OSS
Unpublished work TRW Automotive 2006
8/21/06
MTB mike
GR&R v3
Definitions:
MS; Measurement System Variation (Minitab will label this Total Gage R&R)
PTP; Part-To-Part variation
GR&R; Gage Repeatability and Reproducibility study
TV; Total Study Variation
Caution
The calculations for the GR&R study are not dependent upon whether the
parts are in specification or not. Whether the process is in control or not.
When you do a linearity/accuracy study you select or create parts than span
the entire range of the specification. If you use these same parts for the
GR&R study you will artificially impact the %GR&R and %PT results.
It assumes that accuracy and stability is not an issue.
It is possible to calculate a zero for Reproducibility see example in
appendix.
Unpublished work TRW Automotive 2006
8/21/06
MTB mike
GR&R v3
Caution
AIAG Measurement Systems Analysis MSA 3rd Edition; page vi
MTB mike
GR&R v3
O v e r a ll
V a ria t io n
P a rt-T o -P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e ra to rs
" R e p r o d u c ib ilit y "
O p e ra to r
O p e ra to r
B y P a rt
" I n t e r a c t io n "
When you do a GR&R study the overall variation in the data can be
decomposed into two categories. Variation due to Part-To-Part differences
(PTP) and variation due the the measurement system (MS). The variation
due to MS can further be decomposed into variation due to the gage
(repeatability) and variation due to differences in operators (reproducibility). If
a significant interaction exists, then the reproducibility can be decomposed
into operator and operator-by-part.
Minitab Users Guide Number 2
Unpublished work TRW Automotive 2006
8/21/06
MTB mike
GR&R v3
O v e r a ll
V a ria t io n
P a rt-T o -P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r i a t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
O p e ra to r
B y P a rt
" I n t e r a c t io n "
MTB mike
GR&R v3
O v e r a ll
V a ria t io n
P a rt-T o -P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r i a t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
O p e ra to r
B y P a rt
" I n t e r a c t io n "
MTB mike
GR&R v3
O v e r a ll
V a ria t io n
P a rt-T o -P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r i a t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
O p e ra to r
B y P a rt
" I n t e r a c t io n "
The purpose of the GR&R study is to quantify how large the variation
contributed to the MS (repeatability & reproducibility) is relative to a standard.
The standard is usually the print tolerance and the study variation (process
measurements). Our training material refers to the print Tolerance as %PT,
the study variation as %GR&R. Please be cautious when referring to
them, this distinction may not be clear to everyone. Your customer may
refer to the %GR&R as the one relative to the print tolerance.
Obviously, if you randomly select parts from your process the %GR&R has
meaning relative to your process and any studies completed in the future. If
you arbitrarily select/create parts either ones that are too close to each other
or too far apart it will impact the %TP. If too close the %PT will be large, if too
far apart the %PT will be small.
Unpublished work TRW Automotive 2006
8/21/06
MTB mike
GR&R v3
Lets look at the results of a GR&R study
that is part of our Green Belt training. Data
in appendix. Tolerance = 4
O v e r a ll
V a ria t io n
P a r t - T o - P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a ria tio n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e ra to rs
" R e p r o d u c ib ilit y "
O p e ra to r
O p e ra to r
B y P a rt
" I n te r a c t io n "
Gage name:
Date of study:
Reported by:
Tolerance:
Misc:
By Unit Number
Percent
100
%Contribution
%Study Var
%Tolerance
13
50
12
11
Gage R&R
Repeat
Reprod
Unit Number 1
Part-to-Part
Joe
0.2
Sally
0.1
UCL=0.2053
R=0.09
0.0
12
LCL=0
11
Operator Joe
Joe
13
Sally
UCL=12.09
Mean=12.02
LCL=11.96
12
11
Operator
Joe
Sally
13
13
Average
Sample Mean
By Operator
X-Bar out of
control indicating
that the parts have
more variability
than the MS.
R Chart by Operator
Sample Range
Range chart in
control is a
measure of MS.
Opacity Meter
8/19/01
GRR
4
12
11
0
Unit Number
MTB mike
GR&R v3
O v e r a ll
V a r ia t io n
P a r t- T o - P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
Gage R&R
O p e ra to r
B y P a rt
" I n t e r a c t io n "
Source
VarComp
%Contribution
(of VarComp)
0.00705
0.00183
0.00521
0.00521
0.55042
0.55747
1.26
0.33
0.94
0.94
98.74
100.00
Source
StdDev
(SD)
Study Var
(5.15*SD)
%Study Var
(%SV)
%Tolerance
(SV/Toler)
0.083950
0.042835
0.072199
0.072199
0.741905
0.746640
0.43234
0.22060
0.37183
0.37183
3.82081
3.84519
11.24
5.74
9.67
9.67
99.37
100.00
10.81
5.52
9.30
9.30
95.52
96.13
No significant
interaction in
this example.
MTB mike
GR&R v3
O v e r a ll
V a r ia t io n
P a r t- T o - P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
O p e ra to r
B y P a rt
" I n t e r a c t io n "
No significant
interaction in
this example.
Gage R&R
Source
VarComp
%Contribution
(of VarComp)
0.00705
0.00183
0.00521
0.00521
0.55042
0.55747
1.26
0.33
0.94
0.94
98.74
100.00
MTB mike
10
GR&R v3
O v e r a ll
V a r ia t io n
P a r t- T o - P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
O p e ra to r
B y P a rt
" I n t e r a c t io n "
No significant
interaction in
this example.
Please review the Calculating the Variance Components for Gage R&R Study
(Crossed) by hand document in the appendix for details.
MTB uses the following functions to calculate the variance components. Interaction pvalue greater than 0.25 dropped from model.
a = number of parts
b = number of operators
n = number of replicates
MTB mike
11
GR&R v3
O v e r a ll
V a r ia t io n
P a r t- T o - P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
O p e ra to r
B y P a rt
" I n t e r a c t io n "
No significant
interaction in
this example.
DF
Unit Number
4
Operator
1
Operator*Unit Number 4
Repeatability
30
Total
39
Without Interaction
used in the following
calculations.
SS
MS
17.6209
0.1061
0.0088
0.0536
17.7894
4.40522
0.10609
0.00220
0.00179
2005.79
48.31
1.23
0.00000
0.00225
0.31941
Since p-value of
interaction is > 0.25, it
is dropped from the
model.
DF
SS
MS
Unit Number
Operator
Repeatability
Total
4
1
34
39
17.6209
0.1061
0.0624
17.7894
4.40522
0.10609
0.00183
F
2400.86
57.82
P
0.00000
0.00000
MTB mike
12
GR&R v3
O v e r a ll
V a r ia t io n
P a r t- T o - P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
No significant
interaction in
this example.
O p e ra to r
B y P a rt
" I n t e r a c t io n "
DF
Unit Number
Operator
Repeatability
Total
4 17.6209
1 0.1061
34 0.0624
39 17.7894
MS Repeatability = 0.00183
MS Part = 4.40522
Unpublished work TRW Automotive 2006
8/21/06
SS
MS
MS Operator = 0.10609
a = 5 parts
b = 2 operators
n = 4 replicates
MTB mike
13
GR&R v3
O v e r a ll
V a r ia t io n
P a r t- T o - P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
MS Repeatability = 0.00183
MS Operator = 0.10609
MS Part = 4.40522
No significant
interaction in
this example.
O p e ra to r
B y P a rt
" I n t e r a c t io n "
a = 5 parts
b = 2 operators
n = 4 replicates
MTB mike
14
GR&R v3
O v e r a ll
V a r ia t io n
P a rt-T o -P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e ra to rs
" R e p r o d u c ib ilit y "
O p e ra to r
No significant
interaction in
this example.
O p e ra to r
B y P a rt
" I n t e r a c t io n "
Gage R&R
Source
VarComp
%Contribution
(of VarComp)
0.00705
0.00183
0.00521
0.00521
0.55042
0.55747
1.26
0.33
0.94
0.94
98.74
100.00
Authors note; Please forgive round off. It is possible to get a VarComp of Zero for Reproducibility when we have a significant interaction that has a
MS greater than the MS of Operator. See example in appendix.
Unpublished work TRW Automotive 2006
8/21/06
MTB mike
15
GR&R v3
O v e r a ll
V a r ia t io n
P a r t- T o - P a rt
V a r ia t i o n
M e a s u re m e n t S y s te m
V a r ia t i o n
V a r ia t io n D u e
To G age
" R e p e a t a b ilit y "
V a r ia t io n D u e
T o O p e r a to rs
" R e p r o d u c ib ilit y "
O p e ra to r
Gage R&R
O p e ra to r
B y P a rt
" I n t e r a c t io n "
Source
VarComp
%Contribution
(of VarComp)
0.00705
0.00183
0.00521
0.00521
0.55042
0.55747
1.26
0.33
0.94
0.94
98.74
100.00
Source
StdDev
(SD)
Study Var
(5.15*SD)
%Study Var
(%SV)
%Tolerance
(SV/Toler)
0.083950
0.042835
0.072199
0.072199
0.741905
0.746640
0.43234
0.22060
0.37183
0.37183
3.82081
3.84519
11.24
5.74
9.67
9.67
99.37
100.00
10.81
5.52
9.30
9.30
95.52
96.13
No significant
interaction in
this example.
MTB mike
16
GR&R v3
Variance contributed by
each component.
Percent contribution
relative to total.
Gage R&R
Source
VarComp
%Contribution
(of VarComp)
0.00705
0.00183
0.00521
0.00521
0.55042
0.55747
1.26
0.33
0.94
0.94
98.74
100.00
Source
StdDev
(SD)
Study Var
(5.15*SD)
%Study Var
(%SV)
%Tolerance
(SV/Toler)
0.083950
0.042835
0.072199
0.072199
0.741905
0.746640
0.43234
0.22060
0.37183
0.37183
3.82081
3.84519
11.24
5.74
9.67
9.67
99.37
100.00
10.81
5.52
9.30
9.30
95.52
96.13
MTB mike
17
GR&R v3
Gage R&R
Source
VarComp
%Contribution
(of VarComp)
0.00705
0.00183
0.00521
0.00521
0.55042
0.55747
1.26
0.33
0.94
0.94
98.74
100.00
Source
StdDev
(SD)
Study Var
(5.15*SD)
%Study Var
(%SV)
%Tolerance
(SV/Toler)
0.083950
0.042835
0.072199
0.072199
0.741905
0.746640
0.43234
0.22060
0.37183
0.37183
3.82081
3.84519
11.24
5.74
9.67
9.67
99.37
100.00
10.81
5.52
9.30
9.30
95.52
96.13
MTB mike
18
GR&R v3
The graph below shows the relative size of the Measurement System Variation,
Study Variation and the Print Tolerance. Does not imply relative position.
GR&R STUDY
%GR&R = 11.24% Relative To Study Variation
%PT = 10.81% Relative to Print Tolerance of 4
MS Variation
Study Variation
Print Tolerance = 4
MTB mike
19
GR&R v3
The general guidelines for interpreting the GR&R percentage is as follows;
0% to 10% is Desired, 10% to 20% is Acceptable, 20% to 30% is
Borderline, and above 30% is Unacceptable. The graphic below shows the
relative size of these percentages (and 50%) to the process variation.
GR&R STUDY
10%, 20%, 30%, 50%
Relative To Study Variation
GR&R% =10% GR&R% =20%
Study Variation
GR&R% =50%
GR&R% =30%
MTB mike
20
GR&R v3
CATEGORIES
You can think of the number of distinct categories as the number of groups
within the process data that your measurement system can distinguish.
Suppose you measure ten different parts, and MINITAB reports the number of
distinct categories as four. This means that your measurement system, over
time, cannot detect the difference between some of the parts. Increasing the
precision of the gage will increase the number of distinct categories.
The Automotive Industry Action Group (AIAG) states:
"If the number of data categories is less than two, the measurement system is
of no value for controlling the process. It is all noise and one part cannot be
said to be different from another. If the number of categories is two, it would
mean that the data can be divided into high and low groups, however, this is
only equivalent to attribute data. The number of categories must be five,
preferably more, for the measurement system to be acceptable for the analysis
of the process."
Automotive Industry Action Group (AIAG) (2002). Measurement Systems
Analysis Reference Manual. Chrysler, Ford, General Motors Supplier Quality
Requirements Task Force.
Unpublished work TRW Automotive 2006
8/21/06
MTB mike
21
GR&R v3
Lets take a look at the categories or resolution of the MS. In this GR&R study the
number of distinctive categories is 12. Lets look at a visual before we go through a
rigorous explanation. Recall that the standard deviation of the MS was 0.08395 (times
5.15 = 0.43234), and the standard deviation of the PTP was 0.74190 (times 5.15 =
3.8208).
GR&R STUDY
MS
5.15 * St Dev = 0.43234
P-T-P
5.15 * St Dev = 3.8208
MTB mike
22
GR&R v3
The graphic below indicates how many times the spread of the MS will go
into the spread of the study variation. It appears to go in 9 times. This
indicates that we can distinguish at least nine groups or categories within
the study variation. A relatively high number.
GR&R STUDY
MTB mike
23
GR&R v3
The categories represent the number of non-overlapping confidences
intervals that span the range of the study variation. Minitab calculates it, by
dividing the standard deviation of the PTP by the standard deviation of the
Total Gage R&R and multiplying by 1.41. Then rounding to the nearest
integer.
MTB mike
24
GR&R v3
The number of categories must be five, preferably more, for the
measurement system to be acceptable for the analysis of the process."
GR&R CATEGORIES
Relative to Study Variaion
MTB mike
25
GR&R v3
The next obvious question is where did the 1.41 come from. The following is
a relationship between variances.
0< r<1
D = sqrt((1+r)/(1-r))
1 < D < infinity
The square root is in the formula to return the value to the original
units of the data (standard deviation units).
D. J. Wheeler and R. W. Lyday (1989). Evaluating the Measurement Process, Second Edition. SPC Press, Inc.
Unpublished work TRW Automotive 2006
8/21/06
MTB mike
26
GR&R v3
TV = PTP + TG
therefore
PTP
1 r
TV
D
PTP
1 r
1
TV
1
TV PTP
TV PTP
TV PTP
TG
PTP TG PTP
TG
2 PTP TG
TG
2 PTP TG
2 PTP TG
2 PTP
1
TG
TG
TG
TG
D 1.41
TV PTP
TV
TV PTP
TV
2 PTP
PTP
PTP
2*
1.41
TG
TG
TG
PTP
PTP
StDev PTP
1.41
1.41 *
TG
StDev TG
TG
MTB mike
27
GR&R v3
In our example;
Gage R&R
Source
VarComp
%Contribution
(of VarComp)
0.00705
0.00183
0.00521
0.00521
0.55042
0.55747
1.26
0.33
0.94
0.94
98.74
100.00
Source
StdDev
(SD)
Study Var
(5.15*SD)
%Study Var
(%SV)
%Tolerance
(SV/Toler)
0.083950
0.042835
0.072199
0.072199
0.741905
0.746640
0.43234
0.22060
0.37183
0.37183
3.82081
3.84519
11.24
5.74
9.67
9.67
99.37
100.00
10.81
5.52
9.30
9.30
95.52
96.13
D 1.41 *
StDev PTP
.741905
1.41 *
12.46
StDev TG
.083950
MTB mike
28
GR&R v3
MTB mike
29
GR&R v3
GR&R data
MTB mike
30
GR&R v3
MTB hand
calculations
MTB mike
31
GR&R v3
Example of Reproducibility = Zero
MTB mike
32
GR&R v3
Example of Reproducibility = Zero
Gage name:
Date of study:
Reported by:
Tolerance:
Misc:
By Part
Percent
100
1.5
%Contribution
%Study Var
%Tolerance
1.4
1.3
50
1.2
1.1
0
Gage R&R
Repeat
Reprod
Part
Part-to-Part
R Chart by Operator
Sample Range
0.3
CD
DD
1.5
UCL=0.2631
1.4
0.2
1.3
R=0.1022
0.1
0.0
1.2
1.1
LCL=0
0
Operator
CD
DD
1.4
Operator*Part Interaction
UCL=1.379
1.3
Mean=1.275
1.2
LCL=1.170
1.1
Operator
1.5
DD
Average
1.5
Sample Mean
By Operator
CD
DD
1.4
1.3
1.2
1.1
Part
MTB mike
33
GR&R v3
Example of Reproducibility = Zero
Two-Way ANOVA Table With Interaction
Source
DF
Part
Operator
Operator*Valve No
Repeatability
Total
4
1
4
20
29
SS
MS
0.368326 0.0920816
0.000488 0.0004880
0.003298 0.0008246
0.073889 0.0036945
0.446002
111.666
0.592
0.223
0.00023
0.48462
0.92226
DF
SS
Part
Operator
Repeatability
Total
4
1
24
29
0.368326
0.000488
0.077188
0.446002
MS
0.0920816 28.6309
0.0004880 0.1517
0.0032162
P
0.00000
0.70031
Reduced
Model
MTB mike
34
GR&R v3
Example of Reproducibility = Zero
Two-Way ANOVA Table Without Interaction
Source
DF
SS
Part
Operator
Repeatability
Total
4
1
24
29
0.368326
0.000488
0.077188
0.446002
MS Repeatability = 0.077188
MS Part = 0.368326
MS
0.0920816 28.6309
0.0004880 0.1517
0.0032162
MS Operator = 0.000488
0.00000
0.70031
a = 5 parts
b = 2 operators
n = 3 replicates
-0.005113
MTB mike
35
GR&R v3
Example of Reproducibility = Zero
Two-Way ANOVA Table Without Interaction
Source
DF
SS
Part
Operator
Repeatability
Total
4
1
24
29
0.368326
0.000488
0.077188
0.446002
MS
0.0920816 28.6309
0.0004880 0.1517
0.0032162
P
0.00000
0.70031
Variance
Sum of
Squares
n 1
Degrees of Freedom
36
GR&R v3
Example of Reproducibility = Zero
Two-Way ANOVA Table Without Interaction
Source
DF
SS
Part
Operator
Repeatability
Total
4
1
24
29
0.368326
0.000488
0.077188
0.446002
MS
0.0920816 28.6309
0.0004880 0.1517
0.0032162
P
0.00000
0.70031
MTB mike
37
GR&R v3
Example of Reproducibility = Zero
%Contribution
Source
Source
17.84
17.84
0.00
0.00
82.16
100.00
0.292063
0.292063
0.000000
0.000000
0.626755
0.691464
42.24
42.24
0.00
0.00
90.64
100.00
36.51
36.51
0.00
0.00
78.34
86.43
MTB mike
38