Understanding Measurement System Analysis (MSA) also
known as Gage R&R analysis for Instron® Testing Systems
Author: Rajiv Iyer
Introduction:
Measurement system analysis (MSA), also known as a gage R&R (GRR) study, is a critical tool in understanding the capabilities
of any system used to measure a part or a specimen. In the process of measuring with manually operated equipment, common
sources of variation observed are: the specimen or part, the measurement device, and the operator.
Fundamentally, GRR, where the “RR” stands for “repeatability and reproducibility”, is essential to; (a) determine the total
observed variability due to the gage or instrument, (b) isolate the components of variability in the system, and (c) assess if the
instrument is suitable or capable for the intended application. The two “R’s” used in this study indicate if the gage/instrument is
capable of recording the same observed value if we measure the same part several times under ideal conditions (repeatability)
and how variation is observed when parts are measured under different conditions such as different operators and/or time
periods (reproducibility).
The core question answered by this study is whether the gage or instrument is capable of distinguishing between good and bad
units? The cause and effect diagram below identifies majority source of variations with a gage system or instrument.
Figure 1: Cause and effect diagram (fishbone diagram) on multiple sources of variation [2]
[Link] | Page 1 of 5
As presented in Figure 1, there are multiple factors from various sources such as environment, operator, workpiece/part, and
gage that can introduce measurement variability.
Gage R&R Analysis:
There are two types of GRR analysis: the analysis of variance (ANOVA) method and the average and range method (x̄ and R).
ANOVA Method:
The ANOVA method is a powerful technique to estimate the variability of each component or factor in the experiment while
understanding the effect of each factor on the response (measurement). In a typical GRR study the user is interested in
understanding the effect of factors such as the parts and/or operators. Therefore, ANOVA can be extended to apply in GRR
studies to learn the sources of variation. The analysis estimates p-value (probability value) that is used to statistically determine
if a factor has a significant effect on the experiment or not. If the p-value is less than 0.05 (based on 95% confidence interval)
then that factor is highly significant in the study. Furthermore, the contribution of variation from each factor and the total %R&R
can be estimated.
x̄ and R Method:
The x̄ and R method is an alternate method for conducting a GRR analysis. This method is used when a lower sample size is
preferred for the number of trials per part per operator. A set of mathematical equations are defined to estimate %R&R, making
this method easy for calculations.
The x̄ and R method, unlike the ANOVA method, does not have the ability to estimate interactions between parts and operators
and does not estimate errors from the experiment. Therefore the ANOVA method is preferred for GRR studies with the
availability of powerful statistical software packages.
Figure 2 presents a sample graphical summary from a GRR experiment conducted. All the components variations are presented
in this study.
Figure 2: Gage R&R graphical results on variation contribution to the measurement by various factors [3]
[Link] | Page 2 of 5
Figure 2 shows the part-to-part as high source of variation compared to operators (reproducibility) and the gage itself. This
means the majority of the variation comes from the parts used in the study, resulting in low total gage variations (%R&R). From
Figure 2, one can interpret that the gage is capable of measurements.
The guidelines for GRR studies were defined originally by Automotive Industry Action Group (AIAG) back in the 1990s. An MSA
reference manual was created and approved by Chrysler Motors, GM Corporation, and Ford Motors Company. The guidelines are
defined for the %R&R estimates that will help determine if a gage is capable or not. They are as presented in Table 1.
Table 1: AIAG guidelines for GRR acceptance
GRR DECISION COMMENTS
Generally considered to be an Recommended, especially useful when trying to sort or classify parts
Under 10%
acceptable measurement system. or when tightened process control is required.
Decision should be based upon, for example, importance of
May be acceptable for some
10% to 30% application measurement, cost of measurement device, and cost
applications.
of rework or repair. Should be approved by the customer.
Every effort should be made to improve the measurement system.
This condition may be addressed by the use of an appropriate
Over 30% Considered to be unacceptable. measurement strategy; for example, using the average result of
several reading of the same part characteristic in order to reduce final
measurement variation.
How GRR applies to Instron ® Testing Systems?
Instron systems are commonly used to test properties and performance of various materials, components, and structures
composed of plastics, metals, and composites, among others. Since accurate measurement is an integral part of Instron
systems, it is important to understand the repeatability and reproducibility capabilities. With respect to materials testing, it is
critical GRR considers a range of specimens that represent all materials that may be tested. Fundamentally, the study is
designed to determine if a system has the capability to identify good parts from bad parts. Therefore, when applied to Instron
systems, it is essential that the measurement intended varies significantly to identify part to part variation.
GRR can be classified into two types: non-destructive GRR and destructive GRR. Non-destructive testing is an easier approach
when conducting GRR, assuming the specimens in the study do not change in physical properties that can affect the
measurements over time. For example, specimens such as springs can be used in the GRR study because springs are stable
under ambient conditions over time and do not change physical properties over multiple measurements. Therefore, the
measurement intended for springs which is the spring stiffness should vary amongst the specimens used (ex. springs of
stiffness 0.1 kN/mm, 0.5 kN/mm, and 1 kN/mm are recommended).
However, as a part of GRR experiment planning, it is recommended to use specimens that mimic the system’s intended
applications with a distinct difference in measurements between specimens. This will best evaluate the measurement
capabilities of the Instron system under study. For example, if an application involves testing plastics, it is recommended to
consider using a variety of plastic specimens such as rigid, semi-rigid, and soft plastic materials to introduce part variation for
distinct measurements.
[Link] | Page 3 of 5
Destructive GRR studies are more rigorous when compared to non-destructive GRR because specimens are tested until they fail
or break and therefore cannot be re-used in the experiment. Although rigorous, this study adds value to the measurement
evaluation of Instron systems which are widely used in destructive testing applications. The fundamental concept of destructive
GRR is the same, where the measurement intended to be studied should be distinctly different between the specimens used in
the study. A key assumption introduced in the destructive GRR is that all the specimens obtained from one single batch are
almost identical to each other.
To plan a destructive GRR experiment it is most important to consider specimens that are relevant to the system’s intended
application. For example, if an application involves testing metal specimens, then it is important to use metals of distinctly
different strengths as part variation in the GRR planning. Since the test involves breaking metal specimens, a part can be used
only once in the study and therefore when running multiple trials, it is highly recommended to consider using metals from a
single batch so they are almost identical to each other. Significant variations in measurements of metals from a single batch
can add bias to the study and mislead the GRR results by masking the variations of different types of metals used in the same
study.
Introducing only one part among different operators is not a recommended test methodology for above GRR study. In that case,
a different format of GRR study is recommended known as Type I gage study or precision to tolerance ratio (P/T ratio).
Type I Gage Study of Precision to Tolerance Ratio (P/T Ratio)
In simple terms, a Type I gage study is used to understand the precision and accuracy of measurement system without
considering other sources of variation such as operators or parts. The Type I gage study is typically used to check measurement
capabilities of old or pre-installed systems or when only a single part is available. The comprehensive GRR methods detailed in
previous sections are commonly used for brand new or newly installed systems.
P/T ratio is used to estimate how precise the data is to the defined tolerances or specification limits set by the user. This
method is used to analyze the variation coming from the gage due to its limitations in measurements and therefore requires
only one part to be measured by one operator, multiple times. P/T ratio is commonly applied to qualify a machine or instrument
used in production or quality control (QC) environments.
With Instron® systems, if a user is interested in learning the GRR capabilities within 10% tolerance, then a non-destructive test
can be introduced such as the spring example above, but considering only one spring and one operator to run the test and
measuring stiffness multiple times. Again, a specimen that closely represents the intended application requirement is
recommended for P/T analysis. P/T ratio follows the same guidelines defined by AIAG in Table 1.
[Link] | Page 4 of 5
Type 1 Gage Study - Sample Report
Reported by: Rajiv Iyer
Gage name: Sample GRR Tolerance: Sample
Date of study: Sample Misc: Sample
Run Chart - Sample
0.4276 Ref + 0.05 × Tol
0.4272
Response
0.4268 Ref
0.4264
0.4260
Ref - 0.05 × Tol
1 4 7 10 13 16 19 22 25 28
Observation
Basic Statistics Bias Capability
Reference 0.4268 Bias 0.00006167 Cg 1.40
Mean 0.42686167 T 1.667376123 Cgk 1.30
StDev 0.000202571 PValue 0.106
6 × StDev (SV) 0.001215427 (Test Bias = 0)
Tolerance (Tol) 0.017071 %Var(Repeatability) 7.12%
%Var(Repeatability and Bias) 7.67%
Figure 3: P/T results summary for diameter measurement of one part [3]
Figure 3 shows detailed results from a Type I gage study performed when measuring the diameter of a part. The P/T value
indicated by the parameter “%Var (Repeatability)” is 7.12%. Based on guidelines from AIAG, the P/T ratio is below 10% and
therefore it can be concluded that the gage is capable of measurement.
This concludes the detailed overview of the GRR studies that are important when understanding an Instron® system’s
measurement capabilities.
For more information and discussions on this please contact us at [Link].
References
1. Montgomery, C. D., 2009, Introduction to Statistical Quality Control (6th ed.), John Wiley & Sons, USA.
2. Chrysler, Ford, and GM (2010), Measurement Systems Analysis Reference Manual (4th ed.), AIAG, Detroit, MI.
3. MINITAB 17, Statistical Software Package.
[Link] | Page 5 of 5
