MSA (Measurement System Analysis) on Multi-Phase Measurement System

K

Kelly W

We run a test that has two steps. First, the sample is exposed to a controlled damaging influence for 16 hours. Second, a numeric (distance) measurement is made of how much damage the sample sustained. There are 2 of these distance measuring devices, and they are used by 6 different operators. The product that we test tends to exhibit high response variability between pieces of the same production, so a standard test uses 6 pieces. There is no "standard material" that can be run through the measurement system.

The people who ask us to do this test also occasionally ask, "What's your Gauge R&R on that test?" I've been mumbling for a couple of years something about GRR not being appropriate, but now I need to do one of 3 things:

1. Figure out how to do a GRR on this measurement system.

or

2. Explain accurately why GRR is not an appropriate analysis of the system.

or

3. Find an alternative analysis that provides useful information about the system.

It shouldn't be hard to do a GRR on the instrument that provides the numeric observation of the damage (an LVDT with associated analog-to-digital hardware). But that will tell me nothing about the variability introduced by the controlled damaging process.

It's clearly a destructive test, but all the discussions & articles I've read on destructive testing GRR still only apply to a single instrument, which happens to destroy the sample as it is acquiring its numeric value.

An analogous measurement system that I can imagine might be a "ballistic indent" - propel a steel ball at high speed into an aluminum plate, then use a depth gauge to measure the dent. The propulsion system could have, say, angle or speed variations, and the depth gauge could have its own variations. But the customers get a single result of the test, depth of dent, and so they are essentially asking "What's the Gauge R&R on the [ball propulsion-depth gauge] measurement system?"

I can't find anything on MSA for a multi-phase, destructive test - any suggestions?
 

Miner

Forum Moderator
Leader
Admin
You do have a complex situation here.

First, let's discuss whether this is truly a non-replicable test. Obviously, the parts are destroyed, but can a destroyed part be measured multiple times? This will have implications on the design of an effective study.

Second, the measurement system does include the 16 hour degradation process, but it is more complex than the standard R&R study because you have more sources of variation. Ideally, the study should be designed in such a way that you can separate out the sources of variation, so you can take action on the largest sources if necessary.

If you can provide more detail on the test, I can help you design an appropriate study. For example, what are the variables on the 16 hour damaging portion of the test. The more detail provided, the more assistance I can provide.
 
K

Kelly W

The destroyed parts are measured just before the end of the exposure cycle. They could be measured multiple times, as long as it was within a fairly short period of time, say 15-20 minutes. Continued exposure continues the damage, although the exposure cycle is designed with the belief that the measurement takes place during a period of fairly slow additional damage.

I guess the variables in the damaging portion of the test would be:

- Errors in positioning of the parts. This would be sort of an operator error, and generally will be either right or not.

- Which side of the chamber the part is placed

- Stability of conditions (temperature and relative humidity) during the 16 hour period

- Accuracy of conditions during the 16 hour period. They are always intended to be the same.

Another bit of possible interest - the damage response varies by product. For some products it is negative, for some there's hardly any response, and for others it is positive.
 

Miner

Forum Moderator
Leader
Admin
Thank you for the additional information. Let me consider an effective way to design the appropriate study, and I will get back on this.
 

Miner

Forum Moderator
Leader
Admin
Sorry for the delay. This took some thought.

First, this cannot be analyzed as an ordinary R&R study, so I recommend that you run the study as I describe, then post your data. I can then analyze it for you.

Here is my recommendation for your study. Select a multiple of three for the number of parts. I recommend 12, but 9 would be acceptable. Have three operators set up 4 (or 3) of the parts, identifying which operator set up which part. Also note which side of the chamber was used, but do not specify this in advance. Record the temperature and %RH at intervals throughout the test.

Perform the rest of the R&R study as a crossed study. If you can get the nine measurements (3 operators x 3 replicates) done within your test window, that would be ideal. If not use 3 operators x 2 replicates.

Once I analyze the results, I will probably recommend a follow-up DOE on the test conditions (i.e., side of chamber, temperature, %RH, etc.) to determine the significant factors.
 
K

Kelly W

Thanks for this initial design. It may be a couple of weeks before I can acquire the data - I'll get it to you at that time.
 
K

Kelly W

Quick follow-up question...

> Also note which side of the chamber was used, but do not specify this in advance.

If not specified, they could conceivably put all the parts in one side. Should I do anything to try to get equal representation of both sides of the chamber?
 

Miner

Forum Moderator
Leader
Admin
The problem is that you do not want to mix an investigative study with an R&R study.

By not specifying the side, we will get random results as in real life. Recording the chamber side, allows this to be analyzed as a covariate and the magnitude of the effect determined.

If you specify the side as in an investigative DOE, you may be inducing variation into the R&R study that may not exist.

You always go back to: What is the purpose of the study?
 
K

Kelly W

I finally had an opportunity to collect this data - attached in spreadsheet form. Hope too much time hasn't passed...
 

Attachments

  • GaugeRR.xlsx
    15.9 KB · Views: 143

Miner

Forum Moderator
Leader
Admin
Not too much time unless it is on your end. I am looking at your data. Did you record the end at which the samples were located? I didn't spot it. Also, what is your tolerance for this test, if applicable?
 
Top Bottom