Uncertainty Analysis for C Series Modules

[KyleL12]

I am interested in to GUM style error analysis (Type A, Type B) for measurements for voltages (NI 9219) and temperature (NI 9214). I have consulted NI Applications Engineering concerning Absolute Accuracy (AA) for a different NI DAQ device.  The NI 9214 and NI 9219 also specify AAs.  Yet looking for as much information as possible before asking for the same question, I came across the following documents:

http://digital.ni.com/public.nsf/allkb/8BA2242D4BCC41B286256D1D00815B90

http://www.ni.com/white-paper/4517/en/#toc3

I also found a GUM uncertainty budget for the NI USB-568x:

http://www.ni.com/tutorial/6785/en/

In the USB-568x uncertainty tutorial it states: “Absolute uncertainty percentages are not included in the specifications documents of NI USB 568x RF power sensors”.  Thus, the need for an uncertainty analysis. 

In the GUM analysis all errors are assumed to follow a distribution, so the idea is to determine the appropriate distribution and the confidence level/coverage factor to determine the standard deviation or standard uncertainty. With my previous contact with NI,  I was thinking it would be reasonable to assume the AA could be approximated as normally distributed uncertainty with a coverage factor of 3 because the noise error is multiplied by a coverage factor of three before adding it to the AA. 

With this additional info, I am seems my intial assumption maybe wrong and the AA could be a correction I need to apply to the readings from these DAQs.  Either the error in this correction is unknown, negligible, or perhaps not applicable. If I want a correction with an uncertainty, then I would probably what NI calls a system calibration.  The GUM analysis is suppose to be able to determine and propergate errors in measurements of any complexity, so AA should come how be included.  Even other methodologies require the errors to be of the same type before being combined, usually by Root Sum of the Squares.

Thank You

[will.i.am10]

Hi KyleL12,

The absolute accuracy is the worse-case error for measurement of a DAQ device, in other words how close the measurement is to the actual true value of the signal. This is explained pretty well at the very bottom of the second link you attached. That is true that as long as the device is within calibration these absolute accuracy numbers are valid. Does this answer your question? Or did you have any other specific questions you were looking to have answered?

-will.i.am10

[KyleL12]

I am trying to figure out how Absolute Accuracy (AA) works into a GUM uncertainty budget. Since National Instruments does a GUM analysis for a certain product and not for other devices, I am wanting to make sure I define the AA properly for the GUM analysis. They are a few ways I can do this depending on what AA exactly is.

If AA is really the difference between a true value and a measured value.  I would think this could be done say AA = 0.003 V, then V = 5.000 and Vcrr = 5.003.  Or maybe I guess AA is the uncertainty of in the difference of Vtrue - Vdaq.  Based on the language, it is an extended uncertainty and I would have to determine the standard uncertainty from it for the GUM.  It would be nice to make sure I have a reasonable assumption on the distribution and "confidence" of this error.  Even if the information isn't available, it might be known of errors that occur have a central tendency in the range of AA, etc.  The handbook I am using discourages simply defaulting to a rectangular (or uniform) distribution as they are certain criteria to be met for this distribution.

I have actually tried to search the internet for published works that used NI hardware with AA spec and reported uncertainty budgets.  The works I came across have treated the AA treated them as an extended uncertainty, yet quite a few of the works don't provide the details behind their budgets.  Interestingly, they would combine the gain and the offset components of the AA in a Root Sum of the Squares manner, which I think would be interesting to clarify too.  Based on NI's notes and what I know of accuracy specifications in the form of %*Reading + Least Significant Digit for say a Digital Multi-meter, then the components of the AA should be combined by the equations provided NI. It then can be used as an uncertainty.  

[will.i.am10]

Hi KyleL12,

I can’t say for certain how these absolute accuracy values apply to the GUM analysis. If you have any questions about the AA of the devices, I will be happy to help. It sounds like you have a pretty good idea on the meaning of AA and understand what the values mean, but are just looking for how to apply them to your analysis. If I am misunderstanding this and you do have questions about the AA feel free to let me know!

-will.i.am10

[KyleL12]

You are correct.  I think I was starting to over think it. After taking a closer look at the links as you responded to my question, the absolute accuracy is an uncertainty as in the example at the end of the first link.  I should just make my assumptions and use in my calculations. Unfortunately, this is not the toughest part of my analysis.