Starting to Calculate Uncertainties

D

Debbie

I am starting to plan to calculate uncertainties for a calibration lab. I've seen many approaches taken, i.e., experiments to determine random variation, analyses of the history of calibration and an approach that involves taking the partial derivative with respect to the various parameters that influence the result.

My first uncertainty calculation is pressure drop which of coarse is related to air flow, diameter of the tube, temperature etc. Would the partial derivative approach alone be OK -- or do I have to run an experiment in addition to performing the calculation.
Also, am I right to assume that once I determine all the uncertainties using the partial derivative approach, that I also consider the uncertainty of the standard that I'm using?

I'm desperate and confused-- so please help.

Debbie
 
G

Graeme

Debbie,

I would like to suggest some things as an action plan. You probably have already done some of this, but just in case ...

Research and Study:

Arm yourself with Tools:
In particular, look at the FREEware measurement uncertainty calculator by Chris Grachanen. you can download it from Cal Lab Magazine, (broken link removed)

Apply the Principles, but Start Simple:
  • Get the low hanging fruit first. Start with what you already know or can easily find out, and get that documented first. (Example: the accuracy specifications of your measuring instruments.)
  • I often continue with a simple sensitivity analysis of the major paramters that I can measure or control. (For example, if a measuring instrument's uncertainty changes by 5 ppm per degree C, calculate it out at a few measurement points to see how much effect it has.)
  • Along the way, document your processes for determining various factors.
  • Make sure you have every uncertainty factor properly characterized as Type A or Type B. Type A is factors that you measure yourself and analyze statistically. (Example: hourly values from your temperature/humidity datalogger.) Type B is numbers you get from other sources. (Example: the accuracy specification of your anemometer.)
  • Before going to the work of setting up a designed experiment run, ask yourself if the factor is really a significant contributor to your measurement result, to the point where it makes economic sense to reduce it. If it is a significant portion and you think you can economically reduce the variation, then DOE may be justified. If the amount of variation is a trivial part of the result it may not be worth the effort.

Does this help some?

------------------
Graeme C. Payne
ASQ Certified Quality Engineer
[email protected]


[This message has been edited by Graeme (edited 06 June 2001).]
 
Top Bottom