I am thinking that the requirements on precision of every energy or power measurement device (section 2.1) is too stringent. There could be many thousands of devices whose measurements are added to make a level 3 measurement. Why do we demand that each and every of of them be 1% devices? 100 10% devices yield a 1% measurement, assuming random variations. And I do not see where we are defining what we mean by a 1% error. This is 1 sigma? 90% confidence? sum of statistical + systematic? If I follow this document, if I have single meter, measuring a tiny part of the total power but its off by more than 5%, I cannot even get a level 1 measurement. That seems like it would rule about most systems, if people are really being honest. Or, I have to discard most meter measurement built into power supplies, or hand-built by vendors into their equipment, because they do not pass this requirement. Depending on what you mean by % error (full width of the distribution, etc), I am thinking that most vendors don't calibrate that tightly, and I am not sure what documentation you require. Documented by who? How often? Every one, or just statistical sampling? On the other hand, with many, many measurements, statistical errors will be greatly reduced.
What you could do instead is just ask that people assign an error to their measurement, and explain how they did it. (How they propagated the errors) Then it will not matter what the accuracy of the equipment is, you really want to know how accurate is the reported answer.
-Paul
Paul Coteus
IBM Fellow
Chief Engineer Data Centric Systems
email: coteus@us.ibm.com
external phone: 914-945-2667, internal phone: 862-2667, cell 914-646-4205, fax: 914-945-4219
I am thinking that the requirements on precision of every energy or power measurement device (section 2.1) is too stringent. There could be many thousands of devices whose measurements are added to make a level 3 measurement. Why do we demand that each and every of of them be 1% devices? 100 10% devices yield a 1% measurement, assuming random variations. And I do not see where we are defining what we mean by a 1% error. This is 1 sigma? 90% confidence? sum of statistical + systematic? If I follow this document, if I have single meter, measuring a tiny part of the total power but its off by more than 5%, I cannot even get a level 1 measurement. That seems like it would rule about most systems, if people are really being honest. Or, I have to discard most meter measurement built into power supplies, or hand-built by vendors into their equipment, because they do not pass this requirement. Depending on what you mean by % error (full width of the distribution, etc), I am thinking that most vendors don't calibrate that tightly, and I am not sure what documentation you require. Documented by who? How often? Every one, or just statistical sampling? On the other hand, with many, many measurements, statistical errors will be greatly reduced.
What you could do instead is just ask that people assign an error to their measurement, and explain how they did it. (How they propagated the errors) Then it will not matter what the accuracy of the equipment is, you really want to know how accurate is the reported answer.
-Paul
Paul Coteus IBM Fellow Chief Engineer Data Centric Systems email: coteus@us.ibm.com external phone: 914-945-2667, internal phone: 862-2667, cell 914-646-4205, fax: 914-945-4219