The ISO17025 standard says that when you make a determination of in-tolerance or out-of-tolerance, you must include the uncertainty of the measurement. I know Troemner will call a weight out of tolerance if an in-tolerance measurement plus the uncertainty would result in a value outside the limits. For this reason, we have been directed by some of our clients to send their weights to a different vendor.
I was wondering how many labs out there include the uncertainty when determining in/out.
I wouldn't trust any lab that didn't include measurement uncertainty when determining in or out of tolerance. A measurement result doesn't mean anything without a measurement uncertainty.
There are ways to "guardband" to change the test limits based on the uncertainty and an allowable false-accept percentage. This is legitimate and the percentage is chosen based on the amount of risk you want to take that the item may actually be out of tolerance even if the reported measurement number is within the test limits.
Yes, if you are within the limits of your uncertainty there is a possibility that the instrument under test is out of tolerance. But it is just as likely that it good. When a big part of your client base is Pharma, an out of tolerance investigation can run into thousands of dollars and potentially millions if batches of drugs are determined to have been affected. Are you so sure of your uncertainty that you're willing cost the industry so much money based on a probability? I feel it should not be the calibration lab making that determination. We should report that based on or measurements, with these uncertainties, we have determined that the instrument meets spec. It is up to the client to determine if the uncertainty should be included. Think about measurements like humidity. Rotronic is making probes that are rated at ±0.8% RH. One of the best standards available to most labs is the Thunder Scientific and it is rated at ±0.5%RH. So any Rotronic probe that exceeds 0.3%RH from nominal should be called bad. That doesn't seem to be right to me.
It should absolutely be the calibration lab making this claim. You are removing liability from the customer. You can try returning the item to the customer marked "indeterminate", but that does nobody any good and I would be looking for another cal lab if it were my money.
A measurement result is always a statistical distribution and the reported value and its uncertainty describe the shape, the parameters of the distribution. You give this this data to the customer. If you are a good lab who does good work, you would guardband the test limits so that only points marked "pass" do so with 95% or 99% certainty. If the customer will accept some risk, you can guardband the limits to a specified false-accept or false-reject ratio. This is done to save money on repairs or more lengthy calibrations that require adjustments.
Without challenging anyone's knowledge of 17025, I'd like to observe that since it's obvious that unless we are able to match the manufacturer's uncertainty we can't certify an instrument is in tolerance without accounting for the difference.
Here is a link to a paper discussing FAR and the old Z540 4:1 ratio us old timers (see section 4.6) are used to.
http://www.isgmax.com/Articles_Papers/Risk%20Analysis%20Methods%20for%20Z540.pdf
N79, I think you have a better understanding of this subject than I do. Do I understand correctly that we might adjust the gaurdband limits to account for differences in uncertainty?
If you are calibrating the device to OEM specs, then so long as your measurement falls within the listed specs minus your uncertainty, the device is within spec.
Now, if you are using an OEM procedure that lists a test tolerance, then yes, you would change the acceptable tolerances based upon your measurement uncertainty.
However, I don't think I have seen any OEM procedures that include uncertainties in their performance verifications. They are typically a hard pass/fail.
Quote from: briansalomon on 11-19-2017 -- 20:51:59
Without challenging anyone's knowledge of 17025, I'd like to observe that since it's obvious that unless we are able to match the manufacturer's uncertainty we can't certify an instrument is in tolerance without accounting for the difference.
Here is a link to a paper discussing FAR and the old Z540 4:1 ratio us old timers (see section 4.6) are used to.
http://www.isgmax.com/Articles_Papers/Risk%20Analysis%20Methods%20for%20Z540.pdf
N79, I think you have a better understanding of this subject than I do. Do I understand correctly that we might adjust the gaurdband limits to account for differences in uncertainty?
Yes, although you don't really adjust the guardbanded limits as much as it just falls out of the equations. The simplest method, as CalibratorJ pointed out, is just to subtract your uncertainty from the test limits. But in a lot of measurements, this results in a low TUR. For whatever reason we seem to still like the 4:1 rule so it causes some heartburn to see a 2:1 or a 1.2:1 ratio. Guardbanding essentially changes how much your measurement + uncertainty can overlap the test limits and still be considered a pass. This is based on how much risk you are willing to accept as a false-accept ratio.