03 August 2011

Understanding Uncertainty in Measurement

{Photo from Lumq.com}
One of my university professors went to considerable lengths to communicate the concept that there is no perfect test, driving home the idea that you will always have to live with uncertainty. While this didn’t prohibit him from having numerous exams, it did help me understand that we can never build the perfect instrument. Some researchers believe that acquiring new equipment will give them better results and make them a better scientist. This thinking can be dangerous. Science is not automatically advanced by purchasing new instrumentation. Science is most rapidly advanced by knowing the limitations of your equipment, being aware of the uncertainties and where they arise, understanding the model you are using, and anticipating sources of error and recognizing them when they appear.

For a company, uncertainty is part of business. As we work to improve our products, massive amounts of data are accumulated. How do we communicate the accuracy of our products to our customers? When an improvement is made, does it justify a revision of the published specifications? Several of our board meetings have included discussions on terminology used in writing specifications. The word “typical”  appears in specifications from numerous companies. Is “typically” a synonym for “usually”? What percentage of sensors is within the typical accuracy range?

At Apogee, our solar radiation sensors have a specification for accuracy (5%), uniformity (3%) and repeatability (1%). The accuracy compares each sensor’s output to an absolute reference standard. Uniformity is how consistent our sensors are compared to each other. Repeatability refers to how a sensor performs against itself. Does the same sensor perform consistently under the same conditions?

The numbers in our specifications are based on statistical analysis. Large populations of our sensors are compared to a reference, ISO or NIST traceable where available, and the error is measured. We then calculate the mean and standard deviation of the sample from the reference standard. The specifications listed for accuracy, uniformity and repeatability represent plus or minus two standard deviations from the mean (95% of the population).

So, as my professor taught me, there may be no perfect test, but the more we learn and understand the abilities and limitations of the equipment we use, the better our results will be.

Devin Overly
General Manager

No comments:

Post a Comment