MSA requirements for Systems that Measure Process Parameters

Hi all,

At this thread I want to start more general discussion on MSA for systems that measure process parameters. Let me explain my point of view:

1) According to ISO/TS 16949 requirements organization should define its own special characteristic that can include process parameters (7.3.2.3)
2) According to 7.5.1.1 Control Plan shall "include methods for monitoring of control exercised over special characteristics (see 7.3.2.3) defined by both the customer and the organization"
3) According to 7.6.1 "Statistical studies shall be conducted to analyse the variation present in the results of each type of measuring and test equipment system. This requirement shall apply to measurement systems referenced in the control plan."

From the above statements we can conclude (and auditors do it very often) that we should conduct MSA studies also for measurement systems that control process parameters and not only product characteristics.

And here comes my problem-question. Techniques described in MSA manual are dedicated for product characteristics. The key assumptions for them is that every part/sample can be measured several times. Even if we consider non-replicable studies and nested methods, we make statistical assumption that we can prepare some parts that "pretend" to be the same part. The same assumption is used for bias studies ? we can measure the reference multiple times.

But in my opinion non of MSA models (nested or crossed) is applicable for process parameters where we have only temporary process parameters (like pressure, temperature or force). The same problem is for bias, stability and linearity studies.

On the other hand, when we discuss MSA topics we usually think about accuracy (bias) and precision (variation) of a measurement system. So we can say that calibration gives the answer to both of these questions. But it is true only if we calibrate the whole measurement system in its work environment. For example, we can calibrate thermometer in any lab but it gives us no answer about the bias in the whole measurement system in which this thermometer is used. The same is true to the precision.

So here are my final questions:
1) Can we use techniques described in MSA manual for assessing measurement systems used for process parameters? In my opinion, no.
2) Is calibration sufficient for assessing such systems? In my opinion, can be, but only if we calibrate the whole measurement systems, not the measurement device alone.

I think that these questions were hoovering around the Forum for some time but they were not formulated directly. Thank you for your comments and opinions. I think it will help for many of us.

A quick answer is: yes MSA is applicable and necessary for input porcess parameters. Calibration of the gage is NOT sufficient. it is even more important for 'special processes' as they are your only real control for output quality. You do need to use the nondestruct or destruct approaches for some of the parameters you describe, but people do this all of time and get good answers.

I could describe numerous times when product failures and excess variation were caused by a crappy measurement sytem on an input parameter...and we discovered that by doing the MSA...

It is obvious for me that MSA is necessary for input process parameters and it was not my question. I am asking not IF but HOW to analyse such measurement systems.

Let me give you an example. I have a crimping process in which I have discovered strong linear corelation between the height of the crimping (which is for me a cricital characteristic) and the pressure applied to the press. So according to ISO/TS requirements I have decided to put in Control Plan a process characteristic of the pressure on the crimping machine. Continuing in this way I'm obliged to conduct MSA study for the manometer that measures the pressure too.

And here comes my question again. How to conduct GRR study for the manometer if I can not repeat measurements for the same "part" (which is pressure in this case)? In my opinion even nested GRR will not work here because it is hard to assume that I can prepare homogenous "samples" of pressure.

OK is pressure a setting - or is it a settign with an independent measure of the presssure?

If it is only a setting, then you would need to add independent measures of the pressure as an assessment of the 'setting' to the actual applied pressure.

You may not be able to go with the standard AIAG approach as pressure may not vary too much and still create an in-spec output. So the whole 10 'parts'X3ReadingsX3 operators <XX% of the toelrance is out.

In similar situations (where some input parameter like temperature or pressure are 'set') I've built two sequential parts manufactured using materials as equivelent as possible using the exact same setting. I measure the actual pressure ro temperature independently. Then created two more sequential parts at a different setting. I repeat this until I have roughly 15 to 30 pairs (depending on the cost) at settings that are at the input min, target and max. This should confirm that I have the correct input range to guarantee the output tolerance range.

Then I plot my results on a Youden plot against the input min/max. I do three Youden plots, one using the 'settings' for the pairs, the other using the actuals for the pairs and the third using the actual vs setting for each part. this last part is essential to ensure that the setting is close to reality and that the actuals are close to the setting when there are multiple cavities or locations controlled by a single setting. I don't really calculate any statistics - typically the graphical result is sufficient to see if there is sufficient discrimination...