Hello:

Our Mfg. Facility has thousands of measuring devices which are in several types. They are used in the manufacturing process, tool room, gage lab, and final testing. My question is what have others done in meeting the MSA requirements of bias, linearity, stability, repeatability, reproducibility? We have done GR&R's on each type of gage on initial certification for measurement systems used for in process control and final testing. Also GR&R's have been performed on the measuring systems used in our Gage Lab. Also key Control Characteristics have had GR&R's performed as per customer requirements

Repeatability (precision), Reproducibility, Bias (accuracy), Linearity, and Stability on all gages used in the system.

We have developed excel spreadsheets to document all studies except stability. For stability we do an X-R chart using 5 measurements from a known standard (master?)on a weekly basis. The resulting Cp and CpK's are used for evaluation.

We have found this method painless and acceptable to three Lead Auditors we have dealt with.

I have found that Gage R&R alone will not pass the muster these days and if an auditor accepts them I would wonder about the knowledge base of the auditor.

------------------


Email me and I will send you a copy of the excel (2000) spreadsheets for bias, linearity, and stability. They are self explanatory.

ASD...

It has to be decided by the company looking at the charactristics of the measurements and the measurement systems.

Once the charracteristics have been decided e.g. stability, bias , linearity, R&R , etc.
the study has to be conducted for the same.
The phase 1 study may be helpful in taking such decisions.

However only R&R MAY NOT be sufficient.

Thanks
e.s.deo

We have successfully taken the following approach...

1. Gage R&R required for ALL gages/test equipment.

2. All 5 MSA studies required for tests of special characteristics.

With that said, we have also recently been discussing "what is appropriate?" (Reference the first sentence in 4.11.4) There are cases for particular tests on electronics that a bias study is not appropriate and it becomes our obligation to demonstrate through sound statistics and reasoning that it is an "inappropriate" statistical study.

------------------

MEASUREMENT SYSTEM ANALYSIS (MSA)

STABILITY:

When considering the subject of stability in connection with measurement system, it becomes extremely important to differentiate between what is generally referred to as measurement system stability ----

a ) The amount of total variation in the system’s bias over time on a given part or master part : Known as "Stability over Time"

and

statistical stability, the more general term which is applied to not only stability ,but to repeatability, bias, process in general., etc.



to understand the difference between the two stability, let us consider that there can be 2-measurement systems, measuring exactly the same master part, both of which demonstrate statistical stability, yet one system may have significantly higher variation in its bias over time than the other. From a statistical standpoint, they are equally stable. From a traditional gage stability stand point, the system with greater bias variations over time is considered less “ STABLE “ than the one with lower bias variation. ----------------- (page-21,chapter-2-section-2 )

Statistical properties of measurement systems :

The measurement system must be in statistical control. This means that variation in the measurement system is due to common causes only and not due to special causes. This can be referred to as statistical stability. ---------- (page-5, chapter –1 , section-2 )

Quality of measurement data
The statistical properties most commonly used to characterize the quality of data are

Bias and Variance.

The property called bias refers to the location of the data relative to the master value and the property called variance refers to the spread of the data.

---------- ( page-3 , chapter-1,section-1)

Specifically, the procedures assess the following statistical properties;

Repeatability, reproducibility, bias, stability, and linearity.

Collectively, the procedures are sometimes referred to as “gage R&R” procedures. ------------ (page-15,chapter-15,-section-1)

DOES THE GAGE R&R ASSESS THE SAID STATISTICAL PROPERTIES OTHER THAN REPEATABILITY & REPRODUCIBILITY. YES/NO.

If yes -------------- how

Analysis of results ---graphical analysis


Stability

From range chart stability is determined by: a point or points beyond the control limit ; within operator or within part patterns, ------- ( page-46 , chapter 2- section 4 )





linearity

the averages of the multiple readings by each appraiser on each part are plotted with the reference value or overall part average as the index. This plot can assist in determining :

linearity (if the reference value is used ) ------- (page-52, chapter 2-section4 )

CONCLUSION

GAGE R&R study is an apt exercise for statistical stability and if used, as said above all statistical properties are revealed. Like stability, linearity, repeatability & reproducibility. GRR should be done first and GRR values should be brought below 10% . This can be achieved by understanding the graphical representation and taking appropriate steps. By doing so we are achieving the "Statistical stability".

Without data –based knowledge of the state of control of a measuring process, R&R figures are only descriptions of the data obtained during study. They have no meaning for future performance. Assessing the repeatability ,reproducibility, etc, of a measurement system for which the state of stability is unknown may cause more harm than good. When talking of measurement system statistical stability, the length of time a system is stable is often a major point of discussion. However by means of TIME STABILITY, the length of time a system is stable can be found by using x-bar r-bar control chart. This time stability is to be done after statistical stability in other words called gage R&R. Incase if time stability is performed prior to gage R&R the bias readings will not be exact as readings are contaminated with repeatability and, reproducibility errors.

Above all, any manufacturing process is supposed to be statistically stable if CP & CPK are controlled as they are representing spread & bias (centrality. ).The normal practice to control SPREAD first then to CPK the bias or centering OPN.

Similarly, we have to look at measurement system.

First control repeatability and reproducibility errors of SPREAD by doing gage R&R STUDY and then go to TIME STABILITY to know the extent of DRIFT OR BIAS the CPK.

How / where does uncertainty come into play here? (see Measurement System Analysis (MSA) - Bias, Linearity and Stability (http://elsmar.com/Forums/showthread.php?t=1009) )

Originally posted by Frank45:
Hello:

Our Mfg. Facility has thousands of measuring devices which are in several types. They are used in the manufacturing process, tool room, gage lab, and final testing. My question is what have others done in meeting the MSA requirements of bias, linearity, stability, repeatability, reproducibility? We have done GR&R's on each type of gage on initial certification for measurement systems used for in process control and final testing. Also GR&R's have been performed on the measuring systems used in our Gage Lab. Also key Control Characteristics have had GR&R's performed as per customer requirements

As per PPAP Third Edition, Appropriate MSA studies need to be conducted to all the measuring system available.

Organisation will be get benifited if we understand the intent of MSA Requirement mentioned in QS 9000 Standard.

MSA has Phase I and Phase II.

In Phase I, Bias and Linearity need to conducted before accepting the new gauge.

For old gauges, Bias and Linearity study is conducted by measuring the master value for 12 times and compute the bias as per MSA manual.

As a Phase II, GR&R study is conducted to quality the gauge to use in appropriate measuring system (system compraises part, appraiser, environment).
As ongoing stability of gauge is done to confirm the gauge variation statistically stable.

Hence GR&R alone is not sufficient.

As you are having plenty of measuring gauges, you can group the type of gauge and part tolerance for which the gauge is used and find out the Close tolerance for which the gague is used. Do GR&R study for that type of instrument alone.


------------------

Hello:

Our Mfg. Facility has thousands of measuring devices which are in several types. They are used in the manufacturing process, tool room, gage lab, and final testing. My question is what have others done in meeting the MSA requirements of bias, linearity, stability, repeatability, reproducibility? We have done GR&R's on each type of gage on initial certification for measurement systems used for in process control and final testing. Also GR&R's have been performed on the measuring systems used in our Gage Lab. Also key Control Characteristics have had GR&R's performed as per customer requirement.

Could you tell what method did you use for Attribute Gauges R&R
Thank you,
Andrew

All :

I have one measurement gage that 95% of the time used as product control ( rejecting and accepting part) and the less of time used as statistical analysis ( regression analysis , statistical analysis for process improvement).
In ANOVA minitab,I obtained result as below :
% contribution : 15%
% study variation : 35.50%
% tolerance : 23.67%
ndc : 2

Based on the AIAG requirement,all are rejected except % tolerance.A few modification being made to further improve the gage,but no significant difference of the result.

What is the impact if i release it to production ?Can i continue to release it to production used for product control only ?The ndc is 2,the gage still can discriminate upper and lower part.
Is it safe to be used ?
Please advise

The only metric that applies to a gage used for product control is the % Tolerance = 23.67. If the characteristic to be inspected is not critical, the gage should be acceptable for this use.

The other metrics apply to a gage to be used for process control or statistical analysis, so the gage would not be suitable for that use.

Note: This assumes that the parts measured represent the typical process variation. If the parts are less variable, these numbers will look worse than they actually are. The converse is also true.

Miner :

Thanks for your explanation.

In MSA third edition,page 45,the book has explained that gage with ndc=1 can be used for control if :
1)The process variation is small when compared to the specification
2)The loss function is flat over the expected process variation
3)The main source of variation causes a mean drift

Questions for each statement are :
1)The process variation is refereed to total variation or total gaeg GR&R in minitan anova ?How to quantify that the variation is small compared to specification?
2)what does it mean for " The loss function is flat over the expected process variation" ?
3)What does it mean for The main source of variation causes a mean drift" ?

Thanks.
Kien kit

In MSA third edition,page 45,the book has explained that gage with ndc=1 can be used for control if :
1)The process variation is small when compared to the specification
2)The loss function is flat over the expected process variation
3)The main source of variation causes a mean drift

Questions for each statement are :
1)The process variation is refereed to total variation or total gaeg GR&R in minitan anova ?How to quantify that the variation is small compared to specification?
If you have selected parts that represent the actual process variation, the ANOVA table will show a result for Part Variation as a % Tolerance. Unfortunately there is no criteria for what is acceptable. See this post (http://elsmar.com/Forums/showpost.php?p=214891&postcount=27) for a file "MSA 3rd Ed." There are two options that may help you. The first is the Probable Measurement Error (PME) proposed by Donald Wheeler. The second is a Gage Performance Curve described in the MSA 3rd Edition manual.

2)what does it mean for " The loss function is flat over the expected process variation" ?This refers to a concept proposed by Genichi Taguchi. It refers to how quickly the cost increases as the variation deviates from the nominal dimension. A flat loss function means that there is little cost increase as you progress further from nominal. This might be the case if the specifications are unnecessarily tight and the dimension is not critical.

3)What does it mean for The main source of variation causes a mean drift" ?An example of this would be tool wear in a precision machining process. You typically will see little variation beyond the wear of the cutting tool. I believe that the idea is that you would have plenty of warning before you actually went out of spec that even a marginal gage could see the shift.

Miner :

Regarding to Gage performance curve,
The probability of accepting a part of some reference value is given by the relationship :

Pa = Φ [UL - (Xt + b)/σ] - Φ [ LL - (Xt + b) /σ ]

I will like to know how we calculate for Xt , b and σ.

How do I interpret them from ANOVA Minitab?

What is the meaning for Nm ?

Regards
Kien Kit

See this earlier post (http://elsmar.com/Forums/showthread.php?p=214891#post214891). The attached file in this post will create a gage performance curve. If you would like to see the math behind it, just unhide the worksheet containing the calculations.

An example of this would be tool wear in a precision machining process. You typically will see little variation beyond the wear of the cutting tool. I believe that the idea is that you would have plenty of warning before you actually went out of spec that even a marginal gage could see the shift.

ndc=1 is basically attribute gaging. Sure, you can use it, but that is about as much information as you will get. I would recommend ndc calculated using control limits - not tolerance - to be at least 10. As you might imagine, this is a point that I disagree with AIAG on. They seem to be permitting some sloppy gaging. :cool: