Avoiding Calibration Overkill
Ratio requirements need to be re-examined
by Philip Stein
Early in the 20th century, establishment of laboratories devoted to measurement science, such as the U.S. National Bureau of Standards, now the National Institute of Standards and Technology (NIST), focused formal attention on measurements as a subject worthy of study in itself. Engineering applications of measurements began during this period with the understanding and codification of the need to stabilize commercial measurement applications over long periods, leading to our modern practices of calibration and traceability.
Anyone who has learned anything at all about calibration and traceability in the last 100 years has been taught the 10:1 principle. This ratio, 10 to one, is usually referred to as the test accuracy ratio (TAR) or test uncertainty ratio (TUR), and conventionally describes the ratio of the accuracy of the unit being calibrated to the accuracy of the standard used for the calibration.
When calibrating an instrument, we're told to always use standards, masters or reference materials whose accuracy is 10 times as great as that of the unit under test (TAR of 10:1 or greater). The reason, presumably, is to ensure that any errors in the calibration are negligible compared to the resolution of the calibrated instrument. There are several flaws in this logic, and several problems with the 10:1 principle.
Foremost, at least in my mind, is that the accuracy of the standard used is only one of a large host of issues that can affect the quality of the calibration, and it's often one of the smallest of those problems. For example, one of my consulting customers measures the color of toner powders used in the printing industry. Its customers were complaining about lot-to-lot variability of product, and the metrology system was suspected as the culprit.
The company had seven integrating-sphere colorimeters whose precision extended to 0.1 unit (in the international color system known as CIELAB). Its reference standard system was a more precise colorimeter that read to 0.01 unit. After traceably calibrating the reference system, it would make up standard samples of its own toners for distribution to working instruments around the world as references.
What I found was variation in the preparation of samples presented to the colorimeter, both during the calibration process and during routine measurement of product. A series of measurement capability studies (the analog to process capability studies), showed that errors due to sample preparation were more than 100 times as large as those due to calibration inaccuracies in the instrumentation. The 10:1 principle was being faithfully followed, but the measurements were no good. The capability studies also measured the measurement error contributed by several other sources of variation.
The principle demonstrated here is that calibration, traceability, sample preparation and dozens of other factors that affect accuracy are all part of a measurement system. Accuracy ratios, by focusing on one small aspect of the system, lead everyone to ignore (or at least pay too little attention to) other parts of that system, often with disastrous results.
The 10:1 rule can also wind up imposing absurd and costly requirements. Another client makes plastic molded parts, and the dimensional tolerances called for are plus or minus 0.020 inches. Given the intended application for the parts, this is a reasonable specification. The calipers used to measure the parts are, therefore, calibrated to 0.002 of an inch, well within their capability (although the measurement system, which includes distortion of the parts while being measured, may not meet that capability). The calipers are, in turn, calibrated against a set of gage blocks, which can easily meet the required accuracy of 0.0002 inches. When we send those blocks out for calibration, however, they will have to be measured to within 0.00002, and the service that does our lab work will require 0.000002 (two millionths) of an inch from NIST.
Well, we're not outside the realm of possible accuracies for any of these processes, but surely something is wrong. A reasonable accuracy for calibrating the calipers is one or two thousandths, period. The rest of this process, including having a requirement for traceability at all, is vast overkill.
I do realize that there are many applications where accuracy and traceability of dimensional measurements are crucial, both for the immediate customer and to support a more global interchangeability. What's happened here, though, is that blind application of the rule has resulted in unnecessary costs and trouble--and I believe that a large majority of calibrations done in the United States today fall into this overkill category.
Softer ratio requirements
We can reduce, but not eliminate the problem by softening the requirements for a ratio. ISO 10012-1, for example, supporting the ISO 9000 series, recommends at least 3:1 and prefers 10:1 (but does not require any specific value). ANSI/NCSL Z-540-1, another important measurement standard, calls for 4:1.
A lower ratio reduces the chance of requiring absurd calibrations, but doesn't address the measurement system issue--that the ratio of the standard to that of the instrument being calibrated is often only a small part of the problem. The real problem is that the 10:1 rule is used instead of thinking, and instead of applying good measurement science and engineering.
In 1999, the metrology community is struggling to reduce the use of the TAR and replace it with the uncetainty budget. Uncertainty budgeting does respect the measurement system issue. In fact it's based solidly in the understanding that there are many sources of variation in a measurement, and a specification of the quality of a measurement must include them all. Making an uncertainty budget can be a highly technical and time-consuming effort, but it doesn't have to be. Subsequent columns will address this important topic in detail.
PHILIP STEIN is a metrology and quality consultant in private practice in Pennington, NJ. He holds a master's degree in measurement science from George Washington University, in Washington, and he is a Fellow of ASQ. For more information, go to www.measurement.com.
Column To Focus on Measurement
"MEASURE FOR MEASURE" is a new department that will appear every other month in Quality Progress. It is being written to provoke discussion of some controversial topics in measurement.
The highly technical details of precision measurements are often boring if you're not a specialist, yet everyday measurements are crucially important to us as quality professionals, as consumers and as citizens. Consider the area of standards. The ISO 9000 and 14000 series of standards and their supporting documents include IS0 10012--specialty standards that deal with measurements in the 9000/14000 context. The QS-9000 series of automotive standards has a separate measurement systems analysis volume, and the new TL-9000 telecommunication standard refers to hardware, software and service metrics that are being developed, although they have not yet been published.
These and other examples indicate a strong, new worldwide interest in measurements and measurement science that demands our attention and understanding. Almost everyone in this country--and around the world--is fanatically interested in measurements and statistics--as long as we don't get too technical about them. We follow every possible measure about our favorite sports teams and the performance of their players. We watch our weight and the gas mileage of our cars and trucks, and we measure, track and work hard to improve every aspect of our business life--especially our own income and the bottom line of our employer or enterprise.
Somewhere behind the scenes of all these measures is metrology--measurement science. Measurement has been with us, in the form of weights and measures for commerce, since before the Rosetta stone. The science of measurement began with Galileo and other Renaissance philosophers who realized the need to study and understand the measurements they were making.
Wherever there is a measurement, there is measurement science to be done. Even financial measures, such as profit and productivity can benefit from some application of metrology, and subsequent columns will address this topic. In addition, recent developments and changes in national and international quality system standards have made it important, even essential, for every quality professional to understand and be able to work with the basics of measurement.
For those who'd like to know more about metrology, an excellent resource is ASQ's Measurement Quality Division. It publishes a newsletter, conducts a conference every year and has a wonderful Web site at www.metrology.org. Philip Stein