Non-Replicable GRR Case Study

June 5, 2002

NON-REPLICABLE GRR CASE STUDY

David Benham, DaimlerChrysler Corporation
with input from the MSA workgroup:
Peter Cvetkovski, Ford Motor Company
Michael Down, General Motors Corporation
Gregory Gruska, The Third Generation, Inc.

ABSTRACT
Gage studies provide an estimate of how much of the observed process variation is
due to measurement system variation. This is typically done by a methodical
procedure of measuring, then re-measuring the same parts by different appraisers.
This cannot be done with a non-replicable (destructive) measurement system
because the measurement procedure cannot be replicated on a given part after it
has been destroyed. Here a method using ANOVA is used in a case study which
demonstrates one possible way to determine measurement variation in a nonreplicable system.

This paper is intended to serve as additional guidelines for the analysis of measurement systems.
www.aiag.org/publications/quality/msa3.html

Non-Replicable GRR Case Study

June 5, 2002

Non-Replicable GRR Case Study

June 5, 2002

Non-Replicable GRR Case Study

BACKGROUND
Gage studies provide an estimate of how much of the observed process
variation is due to measurement system variation. This is typically done
by a methodical procedure of measuring, then re-measuring the same
parts by different appraisers. With a standard 10-3-31 GRR, an
assumption is made that between the trials and when handing off
between appraisers, no physical change has occurred to the part. All
appraisers in the study have the opportunity to examine the same parts.
The measurements can be replicated between appraisers and between
trials. With most situations this is a safe assumption to make.
In Figure 1, there are 10 parts to be measured, numbered from 1 to 10.
As can be seen, the same part is measured by each appraiser and for each
trial. A 10-2-3 format is shown for the sake of brevity and as a lead-in to
the case study which follows; randomization is not shown for the sake of
clarity.
Appraiser 1

Appraiser 2

Part #

Trial 1

Trial 2

Trial 3

Trial 1

Trial 2

Trial 3

1
2
3
4
5
6
7
8
9
10

10

10

10

10

10

10

Figure 1: Example of a Standard GRR Layout (10-2-3)

However, there are certain instances where the measurements cannot be
replicated between trials or appraisers. The part is destroyed or
somehow physically changed when it is measured that characteristic
cannot be measured again. That measurement characteristic is said to be
non-replicable. An example might be a destructive weld test where a
weld nut is pushed off a part and the peak amount of pushout force
before destruction is measured; the weld is destroyed in the process, so it
cannot be measured again. So, how can a measurement systems analysis
be conducted when the part is destroyed2 during its measurement?
1
2

10 parts, 3 appraisers, 3 trials.

Not all non-replicable MSA studies necessarily involve destroying parts. In fact, the measurands need not be parts per se, either. For
example, when tests are done on chemical processes, the chemical sample (part) used for testing may have been altered by the test itself
and the solution it was drawn from may have been from a dynamic process where the solution is in constant motion therefore it cannot be
precisely re-sampled.

Non-Replicable GRR Case Study

June 5, 2002

Study Approach

The first thing that must be done before tackling a non-replicable GRR
study is to ensure that all the conditions surrounding the measurement
testing atmosphere are defined, standardized and controlled appraisers
should be similarly qualified and trained, lighting should be adequate and
consistently controlled, work instructions should be detailed and
operationally defined, environmental conditions should be controlled to
an adequate degree, equipment should be properly maintained and
calibrated, failure modes understood, etc. Figure 2 in the MSA manual,
Measurement System Variability Cause and Effect Diagram, p. 15, and
the Suggested Elements for a Measurement System Development
Checklist, pp. 36 38, may assist in this endeavor.
Second, there is a good deal of prerequisite work that must be done
before doing a non-replicable study. The production process must be
stable and the nature of its variation understood to the extent that units
may be appropriately sampled for the non-replicable study where is the
process homogeneous and where is it heterogeneous?
Another
consideration: if the overall process appears to be stable AND
CAPABLE, and all the surrounding pre-requisites have been met, it may
not make sense to spend the effort to do a non-replicable study since the
overall capability includes measurement error if the total product
variation and location is OK, the measurement system may be considered
acceptable.
Standard GRR procedures and analysis methods must be changed and
certain other assumptions must be made before conducting a nonreplicable measurement systems analysis. The plan for sampling parts to
be used in a non-replicable GRR needs some structure. Since the original
part cannot be re-measured due to its destruction, other similar
(homogeneous) parts must be chosen for the study (for the other trials
and other appraisers) and an assumption must be made that they are
duplicate or identical parts. In other words, as the duplicate parts are
re-measured across other trials and by other appraisers, we will pretend
that the same part is being measured. Refer to Figure 2. Part 1 is now
Part 1-1, 1-2, 1-3, 1-4, 1-5, 1-6, for this 10-2-3 layout. Six very similar,
assumed to be identical, parts are used to represent Part 1, and so on for
all 10 parts. The assumption must be made that all the parts sampled
consecutively (within one batch) are identical enough that they can be
treated as if they are the same. If the particular process of interest does
not satisfy this assumption, this method will not work.

Non-Replicable GRR Case Study

June 5, 2002

Appraiser 1

Appraiser 2

Part #

Trial 1

Trial 2

Trial 3

Trial 1

Trial 2

Trial 3

1A1F
2A2F
3A3F
4A4F
5A5F
6A6F
7A7F
8A8F
9A9F
10A10F

1-1

1-2

1-3

1-4

1-5

1-6

2-1

2-2

2-3

2-4

2-5

2-6

3-1

3-2

3-3

3-4

3-5

3-6

4-1

4-2

4-3

4-4

4-5

4-6

5-1

5-2

5-3

5-4

5-5

5-6

6-1

6-2

6-3

6-4

6-5

6-6

7-1

7-2

7-3

7-4

7-5

7-6

8-1

8-2

8-3

8-4

8-5

8-6

9-1

9-2

9-3

9-4

9-5

9-6

10-1

10-2

10-3

10-4

10-5

10-6

Figure 2: Non-Replicable GRR Layout3

Care must be taken in choosing these duplicate parts. Typically for the
parts that represent part number 1 in a study, each duplicate is selected
in a way that it is as much alike the original part as possible. Likewise
for part number 2, and number 3, 4, 5, etc. These parts should be
produced under production conditions as similar as possible. Consider
the 5 Ms +E4 and make them all as alike as possible. Generally, if
parts are taken from production in a consecutive manner, this
requirement is met.
However, the parts chosen to represent part number 2, for example, must
be chosen to be unlike part number 1, part number 3, 4, 5, etc. So
between part numbers, the 5 Ms +E must be unlike each other. These
differences must be forced to be between part numbers. The total
number of duplicate parts selected for each row must equal the number
of appraisers times the number of trials.5 In Figure 2, groups of parts
within each row are assumed to be identical, but groups of parts between
rows are assumed to be different.
Part variation may be expressed as part-to-part, shift-to-shift, day-to-day,
lot-to-lot, batch-to-batch, week-to-week, etc. With parts the minimum
variation would be part-to-part this represents the minimum possible
amount of time between each part. When parts are not sampled
consecutively (i.e., part-to-part), there is more opportunity for variation
to occur different production operators, different raw material, different
components, changes in environment, etc.
So, within a row it is desirable to minimize variation by taking parts
consecutively, thus representing part-to-part variation. Between rows it
is desirable to maximize variation by taking parts from different lots,
batches, etc. There may be economic, time or other constraints involved
which will impose limits on the length of time we can wait to take
3
4
5

The necessary randomized presentation to the appraiser is not shown here for the sake of clarity.
Man, Machine, Material, Method, Measurement plus Environment. Measurement may seem redundant here, but there may be times where
two or more identical measurement systems are used to gain the same information and this should be considered in any study.
When the source of measurement variation is thought to be due to equipment only, using different appraisers may not be required.

Non-Replicable GRR Case Study

June 5, 2002

samples for the between-row data the process must be run, the PPAP
must be submitted, etc. When constraints arise and interfere with doing
things the right way, the results may be subject to modified
interpretation.
Another statistical assumption that must be made for this type of study is
that the measurement error is normally distributed. This is a prerequisite
for any ANOVA (Analysis of Variance).
ANOVA is a better analysis tool for a non-replicable measurement
systems analysis than the average and range method. ANOVA has the
power to examine interactions that the average range method will not
catch.
As a precautionary note, the results from this type of study will
contain some process variation because the identical parts are
not really identical. This may come into play when interpreting the
results in terms of the error percentage related to process variation
or tolerance. The better the methodology used to achieve an
understanding of the production process and its corresponding
measurement system, the more meaningful this non-replicable
measurement study will be.

It is critical in a non-replicable GRR study that the parts be clearly

identified and saved after testing. If any issues arise after the standard
analysis, these parts may be needed for further exploration e.g.,
microscopic examination.
A case study may serve to better illustrate this methodology.

CASE STUDY
A stamped part goes through a critical weld assembly process which
must be destructively tested on an ongoing basis. The process has an inhouse progressive stamping die which produces the steel stampings.
This process is followed by robotic MIG welding (attaching an outside
purchased steel rod to the stamping) in one of 6 different weld stations
each of which has 4 parallel weld fixtures (only one of which will be
used per assembly). Each weld fixture is assigned a letter designation, A
through X. This process has been in production long enough that it has
been studied and analyzed for stability and capability. Each of the 24
weld stations is providing a stable and capable process, however some
are better than others. In an effort to improve the overall process, the
measurement system would be analyzed using this non-replicable MSA
methodology which had been recently introduced to the supplier.

Study Format

This study used the 10-2-3 format 10 parts, 2 appraisers, 3 thus

requiring 60 parts total for the study. Given the complexity of the
process, it was felt that the 10-2-3 would be more manageable than a 103-3. Although there are 24 weld fixtures, only 10 of them were used for
this study and they were chosen using previously gathered data to

Non-Replicable GRR Case Study

June 5, 2002

represent the full range of the process. There was enough early
confidence in the measurement system to make this judgment.
Similarity (homogeneity) within each row was created by taking 6
consecutively produced stampings (6 is chosen to meet the 2 appraisers x
3 trials requirement), then welding those 6 parts consecutively through
the same weld fixture. Dissimilarity (heterogeneity) between rows was
created by taking groups of 6 consecutive stampings from different coils
of steel at a time separated by a few hours, then running them
consecutively through a different weld fixture at a different time.
The rod component which is welded to the stamping is received in bulk
and has already been determined to not play a major role in pull test
variation. Therefore, in this study there was no effort made to maintain
similarity and dissimilarity issues with the rod component.
Previous studies using common problem solving tools had shown that a
manual positioning and clamping system used on the testing machine
was appraiser dependent, so a new and better positioning system with
hydraulic clamps was installed. Parts are located into the machine with
positive locators and hydraulic clamps. A hook on the testing machine
grabs the rod and mechanically pulls on the rod to destruction. A digital
readout on the machine displays the peak pulloff force in pounds and
reads to one decimal place. From this readout, the data is recorded and
the failure mode noted (weld must pull metal from the stamping).
Although the appraiser dependency was assumed to be resolved, this
study still used two appraisers to verify that assumption.
A total of 60 parts were required to do this study. There were 10 groups
of similar parts, 2 appraisers and 3 trials; 10 x 2 x 3 = 60. Parts were
first gathered off the stamping operation, carefully numbered and
quarantined until all 60 parts had been collected. These 10 groups of
parts were selected at 3 hour intervals, over 3 days of production, in
order to force some difference between each group of parts.
Then, each similar stamped group of parts was run through a different
weld fixture. Parts introduced to each weld fixture were presented in
random order within each group of 6.
Appraiser 1
Part #

Trial 1

Trial 2

1-1R1-6R
2-1P2-6P
3-1H3-6H
4-1G4-6G
5-1E5-6E
6-1F6-6F
7-1M7-6M
8-1O8-6O
9-1Q9-6Q
10-1T10-6T

1-6R53
2-2P27

Appraiser 2
Trial 3

Trial 1

Trial 2

Trial 3

1-5R25

1-3R7

1-4R40

1-1R34

1-2R32

2-6P35

2-5P57

2-4P12

2-3P43

2-1P17

3-1H21

3-2H36

3-6H1

3-5H56

3-4H10

3-3H26

4-4G46

4-6G42

4-3G8

4-2G28

4-1G55

4-5G30

5-4E5

5-3E20

5-1E13

5-6E54

5-2E39

5-5E50

6-1F52

6-3F3

6-4F37

6-5F29

6-2F51

6-6F45

7-6M16

7-4M11

7-1M23

7-2M6

7-3M15

7-5M14

8-6O49

8-3O60

8-1O33

8-5O41

8-2O44

8-4O19

9-5Q31

9-6Q59

9-3Q24

9-2Q4

9-4Q9

9-1Q2

10-2T22

10-5T18

10-3T47

10-4T58

10-1T48

10-6T38

Figure 3: Layout Used for Case Study

5

Non-Replicable GRR Case Study

June 5, 2002

Referring to Figure 3, each row shows the similar parts. 1-1R stands
for Stamping #1 of the first group of 6 stampings, which was run through
weld fixture R. 1-2R stands for Stamping #2 of the first group of
stampings, which was run through weld fixture R. 10-1T stands for
Stamping #1 of the tenth group of stampings, which was run through
weld fixture T.6
The parts were numbered with the identification shown in Figure 3.
Parts within each row were presented in a random order to both the weld
assembly operation and to the weld test operation, each with a different
random order. The order shown above is for the weld test operation; the
order for the weld assembly operation within each row was a different
random order and is not shown here. Such randomization reduces the
possibility of any bias present in the order of manufacturing and/or
testing.
Once all assembly was completed, the parts were presented to the
appraisers for destructive testing and the data were recorded. Parts were
saved, preserving the original part numbers, in case any post-analysis
needed to be done.

RESULTS
The data were put into a Minitab Gage R&R (nested) routine which
generated a nested ANOVA. A nested (vs. crossed) ANOVA is required
for this type of study because all parts are not tested by (crossed with) all
appraisers across all trials they cannot be because they are destroyed
after one test. Each appraiser cannot be crossed with each part. Other
charts were also generated by Minitab.
The first thing to look at is the Gage Run Chart.

For the sake of clarity here, the randomized order of total presentation to the appraisers running the test machine is shown as a subscript.

Non-Replicable GRR Case Study

June 5, 2002

10200

Appr1

PULLOFF

Appr2

9200

8200
"Part"

10

PULLOFF

10200

9200

8200
"Part"

Figure 4: Gage Run Chart Weld Tester, Rod Pulloff

The Gage Run Chart in Figure 4 shows plotted points that represent
individual measurement values made on each part by each appraiser
during each trial. The data is grouped by appraiser and presented in the
order in which the trials occurred. The horizontal dashed line represents
the overall mean of all the individual values displayed on the chart. The
abscissa shows each part number (a group of 6 parts in this case).
This type of chart provides visual clues as to the presence of any patterns
in the data. It is desirable that there be no particular pattern within
parts.
If undesirable patterns are present, then further, more
sophisticated investigation may be required. It is also desirable that there
be differences shown between parts. If these differences do not occur
then the measurement system cannot distinguish between the parts used
in the study. Since there are no control limits or other statistical guides
displayed on this chart, some guesswork and common sense may be
required when reviewing it.
In this particular study, there are no significant patterns within parts
and there is some difference between parts.

Next the summary of the GRR graphics is reviewed.

Non-Replicable GRR Case Study

June 5, 2002

By "Part" (Appr No.)

Components of Variation

Percent

100

%Contribution
%Study Var

1
50

10000

9000

4
0
Gage R&R

Repeat

Reprod

8000
Part2
Appr No

Part-to-Part

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Appr1
Appr2

R Chart by Appr No.

Sample Range

1000

Appr1

By Appr No.
Appr2

UCL=928.6

10000

2
500
9000

R=360.7

LCL=0

8000

Appr No

Appr1

Appr2

Xbar Chart by Appr No.

Sample Mean

10000

Appr1

Appr2

3
UCL=9525

9500

Mean=9156

9000

LCL=8787
8500

Figure 5: GRR Summary Graphics GRR Study (Nested), Weld Tester, Rod Pulloff
A graphical summary, generated by Minitab, of this study is presented in
Figure 5. Five graphs are shown and are numbered from 1 to 5.
Presentation and analysis of these charts falls in line with a standard
GRR.
Graph No. 1 shows the components of variation. Four sources are
displayed here

Gage R&R is the variation due to the measurement system.

Repeatability is the variation due to the measuring equipment.

Reproducibility is the variation due to differences in the appraisers.

Part-to-Part is the variation due to measurements taken across

different parts.

Two bars are shown for each source of variation7:

% Contribution is 100 times that variance component source

divided by the total variance ( 2 ).

% Study Variation is 100 times that study variation source divided

by the total study variation ( ).

It is possible in Minitab to also display % Tolerance and % Process but these options were not chosen here.

Non-Replicable GRR Case Study

June 5, 2002

Graph No. 2 is a Range (R) Chart showing the range of the readings for
each appraiser on each part. Note that there is an upper control limit
(UCL) and lower control limit (LCL) and that to be acceptable all points
should be within these limits. Out of control conditions means there is
some sort of inconsistency occurring and should be investigated before
going further.
The data in this case study is all within the control limits and is
considered acceptable with respect to the Range Chart.
Graph No. 3 is an Xbar ( X ) Chart showing the average of each
appraisers 3 readings on each part. Note that there is an upper control
limit (UCL) and lower control limit (LCL) and that to be acceptable,
approximately 50% of the points should be outside these control limits.
The distance between the control limits represents the band of
measurement system variation. If all points were within the control
limits it would mean that no significant distinction can be made between
any of the parts in the study.
In the data for this case study, 11 of 20 points (55%) are out of control
and this is considered acceptable for the Xbar Chart.
Graph No. 4 is a chart of the part-to-part variation across the study. For
this particular study, Graph 4 does not add much value.
Graph No. 5 shows the grand average for each appraiser as well as
points representing the individual readings. The horizontal line between
these points indicates a visual reference for the difference in the grand
average the flatter this line the less difference there is between these
grand averages.
In the data for this case study, the reference line is quite flat indicating
very little difference in the grand average for each appraiser.
Graphs 4 and 5 may be used as clue generators for further analysis.
Also, other graphics, such as histograms, may be used for more detailed
analysis.
While reviewing the graphics it is a good idea to at the same time review
the ANOVA summary of the GRR data. Again, this summary is the
same as a standard ANOVA table and what would be generated by a
standard GRR study.

Gage R&R Study - Nested ANOVA

Nested ANOVA Table
SOURCE
Appraiser
Part(Appraiser)
Repeatability
Total

DF
1
18
40
59

SS
15636
13965615
1853410
15834661

MS
15636
775868
46335

F
0.0202
16.7447

P
0.88869
0

Non-Replicable GRR Case Study

June 5, 2002

Gage R&R
SOURCE

VARCOMP

Total Gage R&R

Repeatability
Reproducibility
Part-To-Part
Total Variation

46335
46335
0
243177
289513

SOURCE
Total Gage R&R
Repeatability
Reproducibility
Part-To-Part
Total Variation

STD DEVIATION
(SD)
215.256
215.256
0
493.130
538.064

% CONTRIBUTION
(OF VARCOMP)
16.00
16.00
0
84.00
100.00

STUDY VARIATION
(6 * SD)
1291.54
1291.54
0
2958.78
3228.38

% STUDY VARIATION
(% SV)
40.01
40.01
0
91.65
100.00

CONCLUSIONS
Given all the above information one must make some decisions about the
measurement system is it acceptable, is it useable, is more study
needed, does an appraiser need better training, is customer approval
required for use, etc? For acceptability, generally the GRR% is reviewed
for suitability8.
In this particular case study example the overall GRR% = 40.01% which
does not make for a clean interpretation. Is this acceptable given that the
traditional upper limit for even a marginally acceptable measurement
system is 30%?
Consider:
1. The overall process to which the GRR% is compared is actually
represented only by the parts (and weld fixtures) chosen for this
study. Only 10 of the 24 weld fixtures were included in this
study. Not all of the process is necessarily represented here. If
the process variation due to all the fixtures is much larger than
that of the 10 selected for this study, then the measurement
system may be acceptable based on MSA guidelines.
2. The data in any non-replicable study such as this will necessarily
include SOME process variation. So some portion of the 40.01
GRR% is actually process variation. It is impossible to separate
all process variation from measurement system variation with
this scheme.
3. The machine used to do the destruct pulloffs is a relatively
sophisticated and expensive piece of equipment. How much
8

ndc is also typically reviewed at this time, however for the sake of brevity it will not be shown here.

10

Non-Replicable GRR Case Study

June 5, 2002

more could it be improved and would it be cost effective? These

would be questions for a possible cost study.

Acceptability Issues

9
10
11

Determining the acceptability of results of a non-replicable study

involves many other possible issues which go beyond the immediate
subject of this paper.

Production process stability: In order to properly analyze and

interpret a non-replicable measurement system study, the production
process must be statistically stable. Since the measurement system
stability is contained within the production process data, if the
overall data demonstrates stability, then the measurement system is
theoretically stable. Data collection frequency and sample size must
be balanced between Average Run Length (ARL)9 and cost of
inspection.

Production process capability:

If the production process is
marginally capable, there is not much breathing room between its
natural process limit(s) and the measurement systems gray area
(see discussion in Chapter III, Section C, of MSA-3). Acceptability
limits (for the process data) might be used when the non-replicable
measurement system analysis shows marginal or questionable
acceptability. These acceptability limits would be calculated by
using a 3 or 4 sigma multiplier10 of the GRR , then adding that value
to the lower specification limit and subtracting it from the upper
specification limit. However, for this to work with any degree of
confidence, there must be adequate breathing room between the
production process and these limits. In the example provided, where
there is only a lower specification limit, the resultant value from
using this multiplier would be added to the lower specification limit
only; for a process with only an upper specification limit, the
resultant value from using the multiplier would be subtracted from
the upper specification limit only.

Long-term measurement system stability: Some effort should be

made to study the measurement systems stability on a long-term
basis using techniques similar to what has been described above.
These techniques may be scaled down to mini studies and used on
an ongoing basis.

Ironically, the better the process Cp11, the more difficult it may be to
establish acceptable non-replicable measurement system analysis
results using the methods here. Recall that a major requirement for
this method to work successfully is that one can knowingly produce
similar (homogeneous) parts (to be used within parts and operator
across trials) and dissimilar (heterogeneous) parts (to be used
between parts and operators). If the production process has a very

The average run length (ARL) at a given quality level is the average number of samples (subgroups) taken before an action signal is given.
This multiplier value would be chosen depending on the amount of risk one chooses to accept.
Technically a process with unilateral tolerance (such as demonstrated in the case study here) has no Cp. What is meant here is a process
with a relatively small amount of variation.

11

Non-Replicable GRR Case Study

June 5, 2002

tight Cp, it may not be possible to meet this requirement. However,

if such a process were to demonstrate stability and a high Cpk, then
the measurement system is probably acceptable (provided there are
no bias or linearity issues) because the data that generated this high
Cpk would include measurement variation.

Summary

Bias and linearity are not evaluated by this method. As with any
GRR study, only repeatability and reproducibility are considered.
The measurement equipment calibration plan becomes critical to the
overall acceptability of the non-replicable measurement system.

As can be seen, a non-replicable measurement study may raise as many

questions as it resolves. Due to the relative sophistication and expense of
the measurement equipment used in this non-replicable measurement
process, and due to the factors mentioned above, it was decided that this
measurement process was suitable for gaging continuous improvement of
the manufacturing process.
At the same time, opportunities for
improving the measurement system would also be evaluated in the
future.
This case study is a single event and therefore has been termed a nonreplicable GRR study. The bigger picture with any measurement system
is the long-term, cradle-to-grave concept. The study here may be used to
initially qualify a system, but more work is required to control that
measurement system over time to ensure its stability and usefulness in
making the appropriate process control and/or capability decisions and
continuous improvement. To keep tabs on this measurement process over
time, a control chart could be used to record the results of special,
consecutive samples taken periodically and those results used to
determine the systems stability.
Finally, the methodology above was what was chosen for this particular
situation. There are other methods which will serve to determine nonreplicable measurement error and each must be carefully selected to suit
the particular situation. The methods shown here merely illustrate a
single approach for a single situation and hopefully it will spur the reader
to consider developing an approach for their own non-replicable
measurement error situation.

_____________________________________________________________________________________________
Dave Benham works at DaimlerChrysler Corporation in Auburn Hills, Michigan, as a Senior Consultant in Supplier
Development. He has a B.S. in Psychology and M.A. in Education from Michigan State University. Dave is an
ASQ Certified Quality Engineer and Certified Reliability Engineer. He has worked in the automotive industry in the
field of quality for 25 years.

12

Non-Replicable GRR Case Study

June 5, 2002

ADDITIONAL READING
Bergeret, F.; Maubert, S.; Sourd, P.; Puel, F.; Improving and Applying Destructive Gauge Capability,
Quality Engineering, Vol. 14, No. 1, September 2001, pp. 59-66.
(http://www.asq.org/info/library/faq/gagerr/articles.html#destructive)
Conklin, Joseph D., Assessing Measurement Error in a Destructive Test, 47th Annual Quality Congress,
May 1993, Boston MA, Vol. 0, No. 0, May 1993, pp. 400-405.
(http://www.asq.org/info/library/faq/gagerr/articles.html#destructive)
Ingram, David J.; Taylor, Wayne A.; Measurement System Analysis, Annual Quality Congress
Proceedings, Philadelphia, PA, Vol. 52, No. 0, May 1998, pp. 931-941.
(http://www.asq.org/info/library/faq/gagerr/articles.html#destructive)
Phillips, Aaron R.; Jeffries, Rella; Schneider, Jan; Frankoski, Stanley P.; Using Repeatability and
Reproducibility Studies to Evaluate a Destructive Test Method, Quality Engineering, Vol. 10, No. 2,
December 1997, pp. 283-290. (http://www.asq.org/info/library/faq/gagerr/articles.html#destructive)
Spiers, Beryl, Analysis of Destructive Measuring Systems, 43rd Annual Quality Congress, May 1989,
pp. 22-27. (http://www.asq.org/info/library/faq/gagerr/articles.html#destructive)
Wheeler, D. J., Evaluating the Measurement Process When the Testing is Destructive, TAPPI, 1990
Polymers Lamination & Coatings Conference Proceedings, pp. 805-807 (www.tappi.org;
plc90905.PDF).

13