Stability & Variability Studies - Tensile Testing

We have conducted capability studies for tensile testing and results are from 1.85 to 5.2cpk.
Our auditor for TS stated that results look fine but we should conduct stability s3/s4 and variability v3/v4 as per pages 143 onwards of msa manual. For destructive test samples.
The more we look at it the more confused we get.

Can anyone give us examples of what charts we should use and how to go bout the studies, I have asked all quality people I know and they have never done them. Please help..........:mg:

Re: stability & variability studies

I can help explain. What you want to show is whether the testing methodology is consistent over time (stability). I recommend making specimens from one homogeneous lot to use throughout the testing program.

First they want you to perform a capability study of more than 30 parts to evaluate the short term variation and develop the control limits. the control limits are to be calculated Then, they give you the option of Xbar-R or I-MR, because if there is significant difference in test 5 specimens (for example), then you need to pull 5 samples. If there is no real difference between samples, pulling one is fine. You need to pull whichever sample size over time (1 test every week or so, for example). 7 tests minimum per calibration period should be adequate. If it "stays in control" (using Western Electric Rules) you are good.

The variable portion is a lot tougher, just by trying to meet the assumptions - particularly the the parts cover the expected extended range of the process variation. That's a fun one to attempt to predict! But essentially (if I can read their minds..you are right, they did not write it very clearly) you perform the test once....then perform the test again on a similar group of parts, then run the Gage R&R calcs on them, as if the two time frames were two separate operators.

Re: stability & variability studies

bobdoering said:

I can help explain. What you want to show is whether the testing methodology is consistent over time (stability). I recommend making specimens from one homogeneous lot to use throughout the testing program.

First they want you to perform a capability study of more than 30 parts to evaluate the short term variation and develop the control limits. the control limits are to be calculated Then, they give you the option of Xbar-R or I-MR, because if there is significant difference in test 5 specimens (for example), then you need to pull 5 samples. If there is no real difference between samples, pulling one is fine. You need to pull whichever sample size over time (1 test every week or so, for example). 7 tests minimum per calibration period should be adequate. If it "stays in control" (using Western Electric Rules) you are good.

The variable portion is a lot tougher, just by trying to meet the assumptions - particularly the the parts cover the expected extended range of the process variation. That's a fun one to attempt to predict! But essentially (if I can read their minds..you are right, they did not write it very clearly) you perform the test once....then perform the test again on a similar group of parts, then run the Gage R&R calcs on them, as if the two time frames were two separate operators.

Click to expand...

Qualtech11: Thanks,
We have conducted the cap studies using results gathered from the last 50 samples, which incidently covers the last calibration period. The control limit for example is 49.1Megapascals as a minimum(customer spec, no max limit) and we usually see around double the minimum.
So should we take say 5 samples from each month period and put into Gage r&r sheet (which we do have).

Re: stability & variability studies

QUALTECH11 said:

Qualtech11: Thanks,
We have conducted the cap studies using results gathered from the last 50 samples, which incidentally covers the last calibration period. The control limit for example is 49.1 Megapascals as a minimum (customer spec, no max limit) and we usually see around double the minimum.
So should we take say 5 samples from each month period and put into Gage r&r sheet (which we do have).

Click to expand...

For stability, put 5 samples per month need to go into a control chart for stability. Which chart depends on your data. Can you post your data in an excel chart? That would help me fab up an example.

The Variability is essentially your gage R&R study. Now this gets ugly, and I am going to explain the "expected" way to do it, even though it is not likely possible to do. First you need to have an idea of what historically your process variation will be - that is over many days, many lots, many operators, etc. Then you need samples that represent that variation. Forget the spec for a moment. Lets say you expect the process variation to be 50.0 to 70.0 MPa. Then you need 10 categories of specimens - some at 50 MPa, some at 52 MPa, some at 54 MPa. and so forth. Then, you need to pull some today....then tomorrow....then the next day. Use the data from each day as if each day was a separate operator, then run the gage R&R calcs.

Now here is where the practitioners and the academics may part ways. How do you get such a variety? The academics think you can just sort them out, or make bunches of different lots...which, of course is ridiculous for studying a new gage and for a destructive test. So, the practitioners often will fabricate parts to represent the variation. For example, if you were using turned dumbbells, you might turn them at different diameters to represent the spread of tensile strengths. It might take some study to get the relationship between the diameter and the expected tensile strength, but it is just a suggestion for trying to dial in the variation so you have consistent parts for the categories.

A practical example of long term charting - I managed a mechanical testing lab for AK Steel. Before I arrived there, they had selected a scrap coil (scrapped for reasons other than properties) and sheared it into sheets. The sheets were collected on a pallet and used to produce stndard tensile specimens for SPC checking of our tensile testers. Before the start of each turn, we pulled 2 tensile specimens and plotted the average and range values for 0.2 % offset yield strength, ultimate tensile strength, and % elongation. If samples didn't meet limits, the operators could pull 1 more set. If those met limits they could go ahead and test product. If not, they had to get me or my senior lab tech involved to determine why they had not and if they could go ahead and use the machine. (The usual reason for failing was poorly machine tensile specimens, with the reduced section not parrallel across the entire gauge length.) We showed long term stability of the testers with the SPC charts.

Of course, any time we made a significant change, we had to recalculate the limits. Things that we considered to be significant included changing the grips from manual to hydraulic, updating the computer controls, etc.

Kevin H said:

A practical example of long term charting - I managed a mechanical testing lab for AK Steel. Before I arrived there, they had selected a scrap coil (scrapped for reasons other than properties) and sheared it into sheets.

Click to expand...

Good idea. It is good to have one lot for stability, because you only want machine variation for stability, not material variation and machine variation. You can't hold the machine responsible for that!

Re: stability & variability studies

bobdoering said:

For stability, put 5 samples per month need to go into a control chart for stability. Which chart depends on your data. Can you post your data in an excel chart? That would help me fab up an example.

The Variability is essentially your gage R&R study. Now this gets ugly, and I am going to explain the "expected" way to do it, even though it is not likely possible to do. First you need to have an idea of what historically your process variation will be - that is over many days, many lots, many operators, etc. Then you need samples that represent that variation. Forget the spec for a moment. Lets say you expect the process variation to be 50.0 to 70.0 MPa. Then you need 10 categories of specimens - some at 50 MPa, some at 52 MPa, some at 54 MPa. and so forth. Then, you need to pull some today....then tomorrow....then the next day. Use the data from each day as if each day was a separate operator, then run the gage R&R calcs.

Now here is where the practitioners and the academics may part ways. How do you get such a variety? The academics think you can just sort them out, or make bunches of different lots...which, of course is ridiculous for studying a new gage and for a destructive test. So, the practitioners often will fabricate parts to represent the variation. For example, if you were using turned dumbbells, you might turn them at different diameters to represent the spread of tensile strengths. It might take some study to get the relationship between the diameter and the expected tensile strength, but it is just a suggestion for trying to dial in the variation so you have consistent parts for the categories.

Click to expand...

Hi Guys, earlier today I posted some examples of charts - if I have been successful in uploading them can someone drop me a reply to let me know...........I will add again to this note just in case.....(new results)

Re: stability & variability studies

bobdoering said:

For stability, put 5 samples per month need to go into a control chart for stability. Which chart depends on your data. Can you post your data in an excel chart? That would help me fab up an example.

The Variability is essentially your gage R&R study. Now this gets ugly, and I am going to explain the "expected" way to do it, even though it is not likely possible to do. First you need to have an idea of what historically your process variation will be - that is over many days, many lots, many operators, etc. Then you need samples that represent that variation. Forget the spec for a moment. Lets say you expect the process variation to be 50.0 to 70.0 MPa. Then you need 10 categories of specimens - some at 50 MPa, some at 52 MPa, some at 54 MPa. and so forth. Then, you need to pull some today....then tomorrow....then the next day. Use the data from each day as if each day was a separate operator, then run the gage R&R calcs.

Now here is where the practitioners and the academics may part ways. How do you get such a variety? The academics think you can just sort them out, or make bunches of different lots...which, of course is ridiculous for studying a new gage and for a destructive test. So, the practitioners often will fabricate parts to represent the variation. For example, if you were using turned dumbbells, you might turn them at different diameters to represent the spread of tensile strengths. It might take some study to get the relationship between the diameter and the expected tensile strength, but it is just a suggestion for trying to dial in the variation so you have consistent parts for the categories.

Click to expand...

Hi I have attached two sets of results hopefully you can help us out now, an example of the chart in excel format would be great.

It appears you are looking at the capability issue to determine what long term variation can be expected. I took the data from 74045718 and applied the best fit distribution (loglogistic family, p=.9997) and for kicks, the normal curve (p=.4612), using Distributional Analyzer. I also took the data from 74045716 and applied the best fit distribution (Johnson family, p=.9792) and for kicks, the normal curve (p=.5698).

I would use the worst case distribution as a decent estimate of the ongoing variation.

That being the case, you would want to fabricate parts with values that range from 270 to 360 in increments of 9 kgf for 7405718 and 265 to 345 in increments of 8 kgf for 74045716. Use these for the variation study.

Keep a set of similar parts (all at about 315 for 7405718 and all at about 300 for 74045716) to run over time for stability.

Does that help?