Choose Your Words Carefully
We need to be more precise when talking about specifications
by Philip Stein
Specification is a highly specialized language--a means of communication.
When a manufacturer or supplier needs to communicate some facts, specifications are usually the language of choice. These facts usually relate to the properties or performance of equipment or the desired outcome of a process.
Most of the time, specifications are stated with descriptive words and quantities described by numbers. We could say "bigger than a breadbox," but we could also say 40 x 60 x 80 centimeters (approximately 16 x 24 x 32 inches). Quantitative statements (numbers) often result in a more precise description, even when the numbers are only approximate values.
Specifications are so widely used, and so useful, we often don't even notice we are using them. They are needed in virtually every circumstance in which people need to communicate details about objects, equipment or processes.
Specifications are particularly important in metrology. We measure things to determine their specifications or confirm specifications we have read or been told. In addition, our measuring processes and equipment also have their own specifications. We use these to determine how well we can measure things.
We also use specifications to communicate with equipment suppliers. That's how we can confirm equipment we purchase or borrow will be able to perform the tasks for which we are acquiring it.
More than meets the eye
Specification is a process that can extend far beyond simple descriptions of the obvious properties of a system. Consider the specifications for a complex electronic system, such as a calibrator. We certainly want to specify the ranges of voltage, current and resistance the calibrator can generate, and we will need to specify the uncertainty with which it can perform those tasks.
A proper specification for such an instrument will also state its ability to maintain its generated output and uncertainty under different conditions of power line voltage, load impedance or output current, and under changing environmental variables, such as temperature and humidity.
Also specified will be the period during which the instrument is expected to maintain these properties before it requires service, maintenance or calibration. Further specifications will include size, weight and cost.
For example, the Fluke 5520A calibrator has 15 pages of specifications in densely packed tables. In addition to many expected items, there is a nomograph showing how long the unit is capable of operating at its maximum rated output current of 11 amperes. Specifications can be complicated indeed.
Tolerances and uncertainty
No measurement is perfect. There's always some small amount of error or uncertainty when we measure something. When we measure, the most important thing we communicate is the result, but part of the communication that takes place about any measurement should be a description of the uncertainty of that result.
The specification of a measurement process or system or a measuring instrument should always discuss the uncertainty of the measurements made.
The new standard approach for expression of uncertainty in measurement is much younger than the use of specifications. Most of the time the notion of a variation or range of a specified value is not shown according to this new standard, but rather is described in terms of tolerances. A typical tolerance statement for a length would say "length = 1.00 � 0.01 inch." This format is so familiar and so widely used we rarely think much about it.
If we think critically about a statement of tolerance, it becomes clear it's not statistically appropriate. What are we saying when we say "length = 1.00 � 0.01 inch"? The precise meaning of this tolerance is that the measured result has a length value that lies somewhere between 0.99 and 1.01 inches, but we have no information at all about where the value is within that interval. This description can also be expressed as a uniform or rectangular distribution, which communicates exactly the same thing in statistical language.
Now think about what the real-life situation must be. A manufacturing process for an object of a certain length, if it is anywhere close to being in control, will produce parts distributed according to a Gaussian or normal (bell) curve, not a rectangular distribution. In a normal distribution, we do have some information about the location of the result. It's more likely to be near the mean than far away. The rectangular distribution, however, has no such information.
In addition, the normal distribution insists some results will be far from the mean. A tolerance states absolutely no values fall outside the limits, but a normal distribution says some values must fall outside the limits. Even though the chances of that happening are small, they're not zero.
Inspection of these parts would seem able to guarantee none fall outside the limits, but even that isn't true. The existence of measurement error in the inspection process leads to the conclusion that we might not reject something that fails our criteria. Thinking of process uncertainty in terms of tolerances is therefore certain to lead to some mistakes, but tolerances are so deeply rooted in our culture and our engineering training and experience, they are unlikely to be replaced.
Specifications and tolerances are often used interchangeably, and this can lead to other mistakes and confusion.
When a laboratory calibrates an instrument for us, a certificate or report of the results is generated and sent back to us with the item. More often than not, this certificate will say the item was "calibrated to manufacturer's specifications." This is very unlikely to be true.
While the calibration might have demonstrated the instrument met the manufacturer's tolerances for the parameters measured, there are many more specifications that probably weren't checked at all. Weight, temperature sensitivity and surface finish, for example, are specified by the manufacturer but ignored, for good reason, by the calibration lab.
We need to use more precise language when we talk about specifications and tolerances. The results are often important or even crucial, yet the words we use are approximate and sometimes just incorrect.
More frequent use of the International Guide to the Expression of Uncertainty on Measurement (GUM) instead of tolerances would certainly help improve this communication. Some standards are now considering this approach. For example, ISO 14253-1 specifies a GUM-like approach toward the description of tolerances for dimensional measurements.
It will take a major cultural change in our profession, though, to completely accept this shift.
PHILIP STEIN is a metrology and quality consultant in private practice in Pennington, NJ. He holds a master's degree in measurement science from the George Washington University in Washington, DC, and is an ASQ Fellow.
If you would like to comment on this article, please post your remarks on the Quality Progress Discussion Board on www.asqnet.org, or e-mail them to firstname.lastname@example.org.