SPC for calibration check data? Daily gram scales check - What is acceptable?

[Craig H.]

Hi, everyone. Maybe you can help with a problem. I did a search and don't see this discussed. The scenario is this:

We have several gram scales. Some read to .01, others to .0001. We have purchased some NIST tracable weights for each scale that are within the weight range that each scale usually measures. We are planning on having the test weight weighed every morning before the scale is used. This number will then be recorded. The whole idea of this exercise is to have the technician alert someone if the scale is "off". The question is, "off" by how much?

Through experience with a similar scheme, and a reference in a text I am trying to remember but can't, the usual XMR chart will often, sometimes almost always, be out of control. What, then, is the best approach to setting the alarm limits? I have had it suggested that the manufacturer's published accuracy should be used, but it seems to me that age, environment, etc. might make that not the best answer.

Any ideas?

 

[Mike S.]

I'd start by just seeing what the scales are capable of. Maybe a full GR&R or even just measuring the "standard" 10 times in a row right after the scale is "calibrated". Then repeat that every day for a few days and see what kind of numbers I get. Maybe you can then set some value X +/- y for the "standard" which you measure once every day or once before measuring unknowns. One thing to be careful of -- a .0001 gram resolution scale will bounce all around just from air moving in the room that you can't even feel from doors opening, A/C or heat, etc. Also be careful to isolate it from any vibrations.

 

[Darius]

Through experience with a similar scheme, and a reference in a text I am trying to remember but can't, the usual XMR chart will often, sometimes almost always, be out of control.

I't look like you have autocorrelation on the data, if the data is too much autocorrelated, the control limits get too small (the within sample variation is too small compared to the between sample variation).

Try this estimate's formula

UCL = X_average + 3 * R_Average/1.128/(1-r^2)^0.5
LCL = X_average - 3 * R_Average/1.128/(1-r^2)^0.5

being r the correlation coefficient (making pairs of each 2 consecutive points).

 

[Hershal]

Craig,

The path you are taking to run daily CHECKS on your balances is prudent. Truthfully, any SPC method will work for what you intend. I personally think however, that the simplest possible should be used......so your operator can tell where the balance is.

One point to consider is how both the balance and the mass standards are handled. NEVER, EVER touch them with fingers. Use only gloves and tweezers supplied with the standard.

If you discover the balance to be out of tolerance (OOT), make sure you call your accredited calibration provider. Now, OOT is really determined by you. Before you start down that path though, look at your calibration certificate to get the points they used, the measurement uncertainty, and any similar factors. Since you are not doing MU (a guess on that) then you should not have to be too concerned about bouyency or local acceleration of gravity, or any of those issues.

And make sure you keep the balance clean. Do not move the balance. That can easily mess up either the mechanical or electronic zero, though the mechanical is more likely. Messing up the mechanical and electronic zeros will mean the balance will not read correctly.

Hope this helps.

Hershal

 

[Govind]

Craig.H,
Most of the points were covered by other contributors; These are some more I can think of.

While using, the standard weights choose weights that are closer to your operating range. Weighing machine may have "linearity Errors". Calibrating at initial, or end points of the range may not guarantee your operating range.

Check the levelling of the scale. The calibration setting may get disturbed if the levelling has changed. I have seen operators hand lift the Precision weighing scales and move to another table to clear the table for components! They may not realize that they just disturbed the calibration.

Creating X-bar S Control Limits and Plotting a daily calibration results is a good idea. (Since more worried about variation). We did that for Special Purpose Optical Measuring equipment. When we saw out of control situation, we were able to correct through the equipment operating Software.

Regards,
Govind.

 

[Craig H.]

Thanks!!!

I knew this was the place to come with my question!