Академический Документы
Профессиональный Документы
Культура Документы
11
Measurement Systems
Analysis
■ Measurement Systems Analysis Overview, 11-2
■ Gage R&R Study, 11-4
■ Gage Run Chart, 11-23
■ Gage Linearity and Accuracy Study, 11-27
accurate and precise precise but not accurate accurate but not precise not accurate or precise
Precision
Precision, or measurement variation, can be broken down into two components:
■ repeatability—the variation due to the measuring device. It is the variation observed
when the same operator measures the same part repeatedly with the same device.
■ reproducibility—the variation due to the measurement system. It is the variation
observed when different operators measure the same parts using the same device.
To examine your measurement system’s precision, see Gage R&R Study on page 11-4.
To look at a plot of all of the measurements by operator/part combination, and thus
visualize the repeatability and reproducibility components of the measurement
variation, see Gage Run Chart on page 11-23.
MINITAB provides two methods for assessing repeatability and reproducibility: X and R,
and ANOVA. The X and R method breaks down the overall variation into three
categories: part-to-part, repeatability, and reproducibility. The ANOVA method goes
one step further and breaks down reproducibility into its operator, and
operator-by-part, components.
Overall Variation
Repeatabilit Reproducibilit
Operator Operator by
The ANOVA method is more accurate than the X and R method, in part, because it
considers the operator by part interaction [3] and [4]. Gage R&R Study (Crossed) allows
you to choose between the X and R method and the ANOVA method. Gage R&R Study
(Nested) uses the ANOVA method only.
If you need to use destructive testing, you must be able to assume that all parts within a
single batch are identical enough to claim that they are the same part. If you are unable
to make that assumption then part-to-part variation within a batch will mask the
measurement system variation.
If you can make that assumption, then choosing between a crossed or nested Gage R&R
Study for destructive testing depends on how your measurement process is set up. If all
11-4 MINITAB User’s Guide 2
Data
The Gage R&R studies require balanced designs (equal numbers of observations per
cell) and replicates. You can estimate any missing observations with the methods
described in [2].
Note If you use destructive testing, you must be able to assume that all parts within a single
batch are identical enough to claim that they are the same part.
In the example on the right, PartNum1 for Daryl is truly a different part from PartNum1
for Beth.
PartNum Operator Measure PartNum Operator Measure
1 Daryl 1.48 1 Daryl 1.48
1 Daryl 1.43 1 Daryl 1.43
2 Daryl 1.83 2 Daryl 1.83
2 Daryl 1.83 2 Daryl 1.83
3 Daryl 1.53 3 Daryl 1.53
3 Daryl 1.52 3 Daryl 1.52
4 Beth 1.38 1 Beth 1.38
4 Beth 1.78 1 Beth 1.78
5 Beth 1.33 2 Beth 1.33
…
…
The Gage R&R studies require balanced designs (equal numbers of observations per
cell) and replicates. You can estimate any missing observations with the methods
described in [2].
2 In Part or batch numbers, enter the column of part or batch names or numbers.
5 If you like, use any of the options described below, then click OK.
Options
X and R method
MINITAB first calculates the sample ranges from each set of measurements taken by an
operator on a part. The sample ranges are then used to calculate the average range for
repeatability. The variance component for reproducibility is calculated from the range of
the averages of all measurements for each operator. Reproducibility, in this case, is the
same as the variance component for operator. The variance component for parts is
calculated from the range of the averages of all measurements for each part.
ANOVA method
When both Parts and Operators are entered
When you enter Operators as well as Parts, your data are analyzed using a balanced
two-factor factorial design. Both factors are considered to be random. The model
includes the main effects of Parts and Operators, plus the Part by Operator interaction.
(When operators are not entered, the model is a balanced one-way ANOVA with Part as
a random factor, as described in the next section.)
MINITAB first calculates the ANOVA table for the appropriate model. That table is then
used to calculate the variance components, which appear in the Gage R&R tables.
Note Some of the variance components could be estimated as negative numbers when the
Part by Operator term in the full model is not significant. MINITAB will first display an
ANOVA table for the full model. If the p-value for the Part by Operator term is > 0.25, a
reduced model is then fitted and used to calculate the variance components. This
reduced model includes only the main effects of Part and Operator.
■ With the full model, the variance component for Reproducibility is further broken
down into variation due to Operator and variation due to the Part by Operator
interaction:
– The Operator component is the variation observed between different operators
measuring the same part.
– The Part by Operator interaction is the variation among the average part sizes
measured by each operator. This interaction takes into account cases where, for
instance, one operator gets more variation when measuring smaller parts, whereas
another operator gets more variation when measuring larger parts.
Use the table of variance components to interpret these effects.
■ With the reduced model, the variance component for Reproducibility is simply the
variance component for Operator.
ANOVA Method
When you use Gage R&R Study (Nested), your data are analyzed using a nested design.
The model includes the main effects for Operator and Part (Operator), in which part is
nested in operator. Because each operator measures distinct parts, there is no
Operator-by-Part interaction.
MINITAB first calculates the ANOVA table for the appropriate model. That table is then
used to calculate the variance components—Repeatability, Reproducibility, and
Part-to-Part.
Note Some of the variance components could be estimated as negative numbers when the
Part by Operator term in the full model is not significant. MINITAB will first display an
ANOVA table for the full model. If the p-value for the Part by Operator term is > 0.25, a
reduced model is then fitted and used to calculate the variance components. This
reduced model includes only the main effects of Part and Operator.
number, usually referred to as the study variation, estimates the width of the
interval you need to capture 99% of your process measurements.
– %Study Var—the percent of the study variation for each component (the
standard deviation for each component divided by the total standard deviation).
These percentages do not add to 100.
■ Number of Distinct Categories—the number of distinct categories within the
process data that the measurement system can discern. For instance, imagine you
measured ten different parts, and MINITAB reported that your measurement system
could discern four distinct categories. This means that some of those ten parts are
not different enough to be discerned as being different by your measurement
system. If you want to distinguish a higher number of distinct categories, you need a
more precise gage.
The number is calculated by dividing the standard deviation for Parts by the standard
deviation for Gage, then multiplying by 1.41 and rounding down to the nearest
integer. This number represents the number of non-overlapping confidence intervals
that will span the range of product variation.
The Automobile Industry Action Group (AIAG) [1] suggests that when the number of
categories is less than two, the measurement system is of no value for controlling the
process, since one part cannot be distinguished from another. When the number of
categories is two, the data can be divided into two groups, say high and low. When
the number of categories is three, the data can be divided into three groups, say low,
middle and high. A value of four or more denotes an acceptable measurement
system.
5 Click OK.
5 Click OK.
%Contribution
Source Variance (of Variance) A
Total Gage R&R 2.08E-03 6.33
Repeatability 1.15E-03 3.51
Reproducibility 9.29E-04 2.82
Part-to-Part 3.08E-02 93.67
Total Variation 3.29E-02 100.00
a. The measurement system variation (Total Gage R&R) is much smaller than what was found
for the same data with the ANOVA method. That is because the X and R method does not
account for the Operator by Part effect, which was very large for this data set. Here you get
misleading estimates of the percentage of variation due to the measurement system.
B According to AIAG, 4 represents an adequate measuring system. However, as explained
above, you would be better off using the ANOVA method for this data. (See Session window
output on page 11-9.)
%Contribution A
Source Variance (of Variance)
a. A large percentage (78.111%) of the variation in the data is due to the measuring system
(Gage R&R); little is due to differences between parts (21.889%).
B A 1 tells you the measurement system is poor; it cannot distinguish differences between parts.
(See Session window output on page 11-9.)
A
B
a. A low percentage of variation (6%) is due to the measurement system (Gage R&R), and a
high percentage (94%) is due to differences between parts.
B Although the X and R method does not account for the Operator by Part interaction, this
plot shows you that the interaction is significant. Here, the X and R method grossly
overestimates the capability of the gage. You may want to use the ANOVA method, which
accounts for the Operator by Part interaction.
C Most of the points in the X Chart are outside the control limits when the variation is mainly
due to part-to-part differences.
Source DF SS MS F P
Gage R&R
%Contribution
Source VarComp (of VarComp) B
a. When the p-value for Operator by Part is < 0.25, MINITAB fits the full model. In this case, the
ANOVA method will be more accurate than the X and R method, which does not account
for this interaction.
B The percent contribution from Part-To-Part is larger than that of Total Gage R&R. This tells
you that most of the variation is due to differences between parts; very little is due to
measurement system error.
C According to AIAG, 4 represents an adequate measuring system. (See Session window output
on page 11-9.)
Source DF SS MS F P
Source DF SS MS F P
Gage R&R
%Contribution
Source VarComp (of VarComp)
B
Total Gage R&R 7304.7 84.36
Repeatability 7304.7 84.36
Reproducibility 0.0 0.00
Operator 0.0 0.00
Part-To-Part 1354.5 15.64
Total Variation 8659.2 100.00
a. When the p-value for Operator by Part is > 0.25, MINITAB fits the model without the
interaction and uses the reduced model to define the Gage R&R statistics.
B The percent contribution from Total Gage R&R is larger than that of Part-To-Part. Thus,
most of the variation arises from the measuring system; very little is due to differences
between parts.
C A 1 tells you the measurement system is poor; it cannot distinguish differences between parts.
(See Session window output on page 11-9.)
MINITAB User’s Guide 2 11-17
C E
D
a. The percent contribution from Part-To-Part is larger than that of Total Gage R&R, telling you
that most of the variation is due to differences between parts; little is due to the measurement
system.
B There are large differences between parts, as shown by the non-level line.
C There are small differences between operators, as shown by the nearly level line.
D Most of the points in the X Chart are outside the control limits, indicating the variation is
mainly due to differences between parts.
E This graph is a visualization of the p-value for Oper∗Part—0.00016 in this case—indicating a
significant interaction between Part and Operator.
A B
E
C
D
a. The percent contribution from Total Gage R&R is larger than that of Part-to-Part, telling you
that most of the variation is due to the measurement system—primarily repeatability; little is
due to differences between parts.
B There is little difference between parts, as shown by the nearly level line.
C Most of the points in the X chart are inside the control limits, indicating the observed
variation is mainly due to the measurement system.
D There are no differences between operators, as shown by the level line.
E This graph is a visualization of the p-value for Oper∗Part—0.48352 in this case—indicating
the differences between each operator/part combination are insignificant compared to the
total amount of variation.
6 Click OK.
Source DF SS MS F P
Gage R&R
%Contribution
Source VarComp (of VarComp) A
Total Gage R&R 1.28935 82.46
Repeatability 1.28935 82.46
Reproducibility 0.00000 0.00
Part-To-Part 0.27427 17.54
Total Variation 1.56363 100.00
a. The percent contribution for differences between parts (Part-To-Part) is much smaller than
the percentage contribution for measurement system variation (Total Gage R&R). This
indicates that most of the variation is due to measurement system error; very little is due to
differences between part.
B A 1 in number of distinct categories tells you that the measurement system is not able to
distinguish between parts.
a. Most of the variation is due to measurement system error (Gage R&R), while a low
percentage of variation is due to differences between parts.
B Most of the points in the X chart are inside the control limits when the variation is mostly due
to meaurement system error.
Data
Structure your data so each row contains the part name or number, operator (optional),
and the observed measurement. Parts and operators can be text or numbers.
PartNum Operator Measure
1 Daryl 1.48
1 Daryl 1.43
2 Daryl 1.83
2 Daryl 1.83
3 Daryl 1.53
3 Daryl 1.38
1 Beth 1.78
1 Beth 1.33
…
5 If you like, use any of the options described below, then click OK.
Options
a. For each part, you can compare both the variation between measurements made by each
operator, and differences in measurements between operators.
B You can also look at the measurements in relationship to the horizontal reference line. In
this example, the reference line is the mean of all observations.
Most of the variation is due to differences between parts. Some smaller patterns also
appear. For example, Operator 2’s second measurement is consistently (seven times out
of ten) smaller than the first measurement. Operator 2’s measurements are consistently
(eight times out of ten) smaller than Operator 1’s measurements.
a. For each part, you can compare both the variation between measurements made by each
operator, and differences in measurements between operators.
B You can also look at the measurements in relationship to the horizontal reference line. In
this example, the reference line is the mean of all observations.
The dominant factor here is repeatability—large differences in measurements when the
same operator measures the same part. Oscillations might suggest the operators are
“adjusting” how they measure between measurements.
Data
Structure your data so each row contains a part, master measurement, and the
observed measurement on that part (the response). Parts can be text or numbers.
PartNum Master Response
1 2 2.7
1 2 2.5
…
2 4 5.1
2 4 3.9
…
4 If you like, use any of the options described below, then click OK.
Options
Gage Info subdialog box
■ fill in the blank lines on the graphical output label
Method
Both studies are done by selecting parts whose measurements cover the normal range
of values for a particular process, measuring the parts with a master system, then having
an operator make several measurements on each part using a common gage. MINITAB
subtracts each measurement taken by the operator from the master measurement, then
calculates, for each part, an average deviation from the master measurement.
To calculate the linearity of the gage, MINITAB finds the best-fit line relating the average
deviations to the master measurements. Then,
Linearity = slope ∗ process sigma
Generally, the closer the slope is to zero, the better the gage linearity. Linearity can also
be expressed as a percentage of the process variation by multiplying the slope of the
line by 100.
To calculate the accuracy of the gage, MINITAB combines the deviations from the master
measurement for all parts. The mean of this combined sample is the gage accuracy.
Accuracy can also be expressed as a percentage of the overall process variation by
dividing the mean deviation by the process sigma, and multiplying by 100.
a. The variation due to linearity for this gage is 13% of the overall process variation.
B The variation due to accuracy for this gage is less than 1% of the overall process variation.
References
[1] Automotive Industry Task Force (AIAG) (1994). Measurement Systems Analysis
Reference Manual. Chrysler, Ford, General Motors Supplier Quality Requirements
Task Force.
[2] R.J.A. Little and D. B. Rubin (1987). Statistical Analysis With Missing Data, John Wiley
& Sons, New York.
[3] Douglas C. Montgomery and George C. Runger (1993-4). “Gauge Capability and
Designed Experiments. Part I: Basic Methods,” Quality Engineering 6(1), pp.115–135.
[5] S.R. Searle, G. Casella, and C. E. McCulloch (1992). Variance Components, John
Wiley & Sons, New York.
Chapter 11 References