My company supplies liquid raw materials to foam producers, including the automotive industry. We are TS16949 certificated, so we are required to demonstrate acceptable R&R for all parameters on our product specifications.

The problem is, these specifications were set without taking into account the inherent variation in measurement of parameters such as 'cream time' 'end of rise time', 'free rise density' and 'foam crushability'. These parameters can be affected by small variations in atmospheric conditions (temp, humdity, almospheric pressure) as well as trace contaminants (in the extreme, a few ppb can cause foam collapse).

It's a bit like trying to apply precise measurements to cooking a cheese soufflé - good/bad decisions are relatively easy, but measurement repeatability is not good.

Our auditors have suggested we switch from variables to attributes to see if the R&R improves. Problem is, customers want to see actual data (it helps them set up their process). We have also looked at widening the specification ranges, but to get below 30% R&R would need a specification which almost everything would meet, including product we know would fail at the customer.

We are now considering asking the IAOB for a 'sanctioned interpretation' which would exempt this kind of test from the strict requirements of the MSA Guide, on the basis that these requirements were intended for dimensional measurements and not abstract performance tests.

Does anyone out their have similar issues with MSA which they have solved, or would also support requesting a sanctioned interpretation for this kind of testing?

It would be great to get some views which might support our case with the external auditors.

Can anyone lend some knowledge on this?

This is a complex situation with no easy answers. Attaching the full results of your MSA study would help us help you.

One "Must" is to control the environmental factors that you stated have a strong influence on the results. If this is not done, there is nothing else that will help.

What is the major source of variation, Repeatability or Reproducibility? If it is Repeatability, taking multiple measurements and averaging will help. The variation may be reduced by the square root of the sample size. For example, the average of 4 samples will vary 1/2 as much as the individual measurement. The average of 9 will vary 1/3 as much as the individual.

Beyond that, I need to see your data and analysis.

This is a complex situation with no easy answers. Attaching the full results of your MSA study would help us help you.

One "Must" is to control the environmental factors that you stated have a strong influence on the results. If this is not done, there is nothing else that will help.


I think Miner is on the right track here (as usual). In the past, I've solved similar problems simply by changing what was being measured--measuring process parameters rather than product parameters. In order to make this work, you need to establish with a fair amount of confidence that all of the relevant process variables, when controlled within specified limits, will produce conforming product. Then measure the variables and focus your MSA on those measurements. One requirement of this strategy is customers who will accept it, but if you have good data the idea shouldn't be difficult to sell.

The problem is, these specifications were set without taking into account the inherent variation in measurement of parameters such as 'cream time' 'end of rise time', 'free rise density' and 'foam crushability'. These parameters can be affected by small variations in atmospheric conditions (temp, humdity, almospheric pressure) as well as trace contaminants (in the extreme, a few ppb can cause foam collapse).

It's a bit like trying to apply precise measurements to cooking a cheese soufflé - good/bad decisions are relatively easy, but measurement repeatability is not good.

Our auditors have suggested we switch from variables to attributes to see if the R&R improves. Problem is, customers want to see actual data (it helps them set up their process).



I am not sure that I would dismiss attribute data so quickly. While your customers may like to see “actual” data, if by your own admission those values are not worth much, then is the “actual” data really worth much?

You may can add an attribute measurement to get better analysis. Many times by using the appropriate statistical tool, you won't have that significant of a loss of efficiency.

I like that you mention your measurements can be affected by small variations. As someone in the calibration industry, I would ask “Do you have equipment sensitive/accurate enough to accurately measure the variations?”


We have also looked at widening the specification ranges, but to get below 30% R&R would need a specification which almost everything would meet, including product we know would fail at the customer.

Why is there so much disparity between your analysis and the customers? How do they assess the product? Is there any possibility in analyzing the difference in measurement systems?

Many thanks :thanx: for the helpful and interesting replies.

Taking the points in order:

Attached is a data file for the first 'geltimer' MSA we did. This measures the time taken for a polymerizing mixture to reach a specific viscosity, as measured by a 'rotational torque rheometer'. The R&R is not good!
In the table you will see three columns in bold.These are the basic info for the gage study:
operator
blend equivalent to 'part'
Geltime this is the result in seconds
There is also a column for the lab temperature that is our covariate (call it noise).
In this case blend 3 is the standard and the average value for it is 126 seconds , following our rules the spec would be +/- 6 or 7 seconds ( 12.6/2 ).Total range should be 12 or 14 seconds.

Controling all of the main environmental factors (temperature, humidity, atmospheric pressure) is not possible in all our existing labs.
Taking multiple measurements and averaging results certainly helps with the R&R, but we don't have the resource in QC to make this a viable permanent solution.

Switching to process parameters rather than product parameters would be a major step forward, if only our incoming raw materials were consistent. Unfortunately they are not, because our (internal) supplier's process is very complex and almost never in a steady state due to both deliberate and non-deliberate causes.

Historically our strategy has been to set tight specifications, apply 100% inspection and adjust nonconforming product into spec by adding more of the deficient active ingredient (back to cooking). The cost of doing this is that we make more adjustments (rework) than would be necessary if we had confidence in the test and could therefore relax the spec.

Our customers test the product by actually using it in their process (no receipt inspection - they rely on our certificate of analysis). This means that a nonconforming batch can stop the line, with high consequential costs.

You have serious problems. Both Repeatability and Reproducibility are issues. Operators Mario, Michael and Peter are distinctly different from the rest in both the average and range. Training is probably required here. Removing their effect highlights other operators.

Standardization of method and training/testing to the standard method WILL help you some, but still does not get you close to where you need to be.

From all indications, this gage simply is not adequate for your needs at that tight a tolerance. Averaging will help, but as you indicated, this is a resource issue.

Viscosity is normally a destructive test. I assume that you are relying on the homogeneity of the batch to say that all samples from one batch are identical. Is this a valid assumption? If you are doing this, collect your samples in as close proximity to each other as possible to minimize within batch variation.

Nice reply posts, by the way. I like the detail you provide, and that you provide responses to each post. I really wish I could give you more assistance.

I’ll chalk this up to the limitations of written correspondence, but I’m missing something here. Here are my questions/assumptions:


If the customer does not do any type of testing, how do they know whether the product is good/bad other than how it reacts?
You are attempting to perform R&R on a process that I understand that you have little/no control. If this is correct, then something may have to give.
You display a specification for time. However, if you are really interested in viscosity, then given all the other variables the time will have to fluctuate.Given Miner’s input regarding viscosity, do you think you might could focus on that and develop some consistency? Maybe assure that the measurement equipment can provide accurate readings for your purpose and the customer’s confidence.

You could keep your process the way it is (making the best of it, though), and focus your refinement on a true objective criteria-the viscosity.

:thanx: for the responses - this should all help vs. the auditor and with a Six Sigma project we have running.

In response to BradM's points:

* Our customers offload our products and feed them directly into their process. They have some possibility to adjust mixing ratio between the 2 components, but that doesn't allow for significantly out-of-spec deliveries, just batch-to-batch variations within the spec limits. To avoid the need for machine adjustments, our customers need consistent quality, and leave it up to us to ensure they get it.

* We control our process - the product variability is due to the raw material variability, over which we have little control.

* The key parameter is the gelation time when the 2 components are mixed - a measure of reactivity. The 2 liquids react to form a polymer, which at a certain molecular weight changes from liquid to gel. The customer needs the liquid mix to fill all corners of the mould before setting.

We are indeed looking at an alternative test device - a vibrating needle curemeter. Hopefully this will show improved R&R.

Back to my original point though - is it realistic for this kind of application test to ever meet the R&R criteria in the MSA Guide? If not, then we should be able to get an exemption (sanctioned interpretation) from the IAOB.

I don't believe this kind of test will meet the R&R criteria. We used to do a somewhat similar test in the adhesive industry and had trouble with that also - but we did not have to meet TS criteria.

What are the raw material variabilities? Percent active ingredient, percent byproducts, percent volatiles? Are there tests you or your supplier can do on those?

The most variable raw materials are both forms of isocyanate MDI - polymeric and mixed isomer. As the names imply, these are mixtures of different isomers and polymers of MDI, and the variability is therefore mainly in molecular weight/visosity and reactivity (NCO value). The upstream plants producing MDI are mainly concerned with volume and efficiency, on the basis that variations in quality can be dealt with downstream. Most of the MDI is sold into less demanding applications (eg. rigid foam insulation and woodbinders).

i had a similar problem when i was working for a moulded foam manufacturer....

typically I used to get GRR of

30% or more when we used a 2" by 2" by 1" core piece of foam

45% or more when we used actual part (skin effect?)

5% or less when we used a spring as a standard

for 50% compression test

For PPAP requirement of 10% or less, we submitted the capability of test equipment (INSTRON?) using the spring and our customers (gm, ford, chryster and others) accepted our results, recognizing the inadequacy of existing method of measuring foam firmness. Please note, we provided other information related to the physical properties of the foam (eg, tensile, density, etc).

Bulk Supplier...ask yourself what customers need?....if i am not mistaken, all they want is a piece of paper stating GRR of 10% or less....if that is the case, you should devise a testing method which will get GRR of 10% or less (like i have done)...you don't have to tell me there is no value added to creating a testing method which means nothing...but sometimes, you have to do what you have to do.

BTW, you are from BASF or Union Carbide? :cool:

The raw material is a mix of isomers and polymers with impurities, produced by a process which is run in campaigns to make a range of products to match market demand.

We plan to look at a different instrument to test the product - a 'vibrating needle curemeter'. Does anyone have experience with this?