I can see where this could become somewhat contentious. I have heard some equally good philosophies on a few different methodologies for interval adjustment. Some of the philosophies are:
1. Use a convenient fixed interval (such as 12 months). This makes work forecasting much easier, but 'over-calibrates' some and 'under-calibrates' others.
2. Family/Class Interval Adjustment. This has good reviews in areas of electronic test equipment (such as the Fluke 77 example (and I am not trying to sell them, only use as a common model). Many electronic test equipment are either unaffected or little affected by level of use, and subsequently, use of a statistically significant sampling of numerous of the same model or model/family may be adequate to determine interval adjustment. This with the implicit exception of 'dogs' or 'gems' (mostly 'dogs'). It is not too difficult to define a dog in terms of reliability. This type of adjustment, however, may not work well at all with mechanical instruments, as they may sometimes tend to go out of tolerance based more on usage.
3. Individual instrument adjustment. This is also quite common, and most readily adapts to the uniqueness of each piece of test equipment. It is higher maintenance, though, and may be a more costly method (in our lab, we are currently using this method, but it is fully automated in our custom database package).
I have seen numerous variations of the above. The bottom line is that your well-defined method be adequate to eliminate calibration interval related risk to product (or service). That is really the bottom line of calibration. If I were a tire manufacturer, and I had temperature controllers to cure an intermediate adhesive layer between my tire tread and the steel belts (fictitious example only; no implication that any tire companies have this problem). The bottom line would be that I calibrated those temperature controllers accurately enough and at adequate intervals to assure that the temperature controller accuracy was continuously maintained to preclude that as a contributing factor to a tire defect.
I only have one more small hand grenade to toss into this discussoin. While visiting with some process engineers in Asia last year, one of them showed me a detailed lengthy report about how important it is in process control NOT to over-control the process. All quantitative process variables have variability (a given). If in controlling a process there occurs a single anomalous variation which throws the process from nominal out toward a control limit, the instinctive reaction would be to adjust the process back to nominal and correct for the error. These engineers showed me the mountain of statistical data that basically says when that happens (process is maintaining at about nominal for a long period, then abruptly shifts away from nominal for one reading), you should adjust nothing. The process statistically will come back to nominal. And the probability is that it did not truly drift away from nominal (I can't, by the way, relate any details of this report for company reasons). When single measurement drifting away from nominal occurs, leave it alone. It was an anomaly.
Now to explain my longwinded rambling. I think we should use some of this philosophy in calibration interval adjustments. If I have a family of Fluke 77's (say 30 of them). My data tells me that they should have (for example) a 2 year interval. If I then calibrate one of the units, and find a gross out of tolerance, in accordance with the above theory I proposed, that gross out of tolerance was an anomaly. And in my lengthy experience with Fluke 77's, that is what it would have been. I have seen corroded function/range switches (somebody left their meter in their car and it got rained on). This is an anomaly, and doesn't truly reflect the propensity of the entire population, and in this case, doesn't even reflect what that specific unit is likely to do next interval. The owner of the meter feels stupid when I give him the compassionate fatherly lecture about leaving such a nice meter out in the rain. Next year, all the units are in tolerance. My point is that I believe in only adjusting intervals based on statistically significant sample. I am doing some short term variability testing on an hp 3458A high accuracy meter at the moment. Short term (smaller data sets) have tendencies to show different patterns of variability than longer term (significant amount of data to be analyzed). We need to be careful about over-adjusting calibration intervals.
Mechanical instruments are a different matter. There is much more of a physical wear and tear issue applicable to those instruments. Micrometers heavily used will perhaps exhibit different wear and tear than those that sit in a tool box all the time, and so must be treated differently. I won't comment on that area, as I do not have much experience in mechanical tools. Individual instrument interval adjustment may well be more appropriate in those cases.
------------------
