Monday, October 14, 2013

Sampling v2.0

I want to take another look at proper sampling because it so key to a good calibration.   While there are statistical tricks to get the most out of anything you produce calibration-wise, if you don't have good sampling, you are, in the best case, creating big problems for yourself.

We want to collect samples such that they are representative of the process.  Samples that are representative have an average that is very close to the average of the whole process at that moment in time. Samples that are not representative will have averages that are not at all similar to the process. 

Collecting representative samples isn't difficult, but you do have to follow certain rules. 

1) Collect samples from lines where the flow characteristic is known to be stable, i.e., in plug flow.  Stable flow means that you will likely not have any turbulence in the line that might de-water your stock or otherwise introduce non-representative sampling.   The easiest way to ensure this is to find a straight length of pipe that is at least seven pipe diameters long, and without any bends or obstructions in it.  

2) Make sure the pipe is full.  No, really, make sure the pipe is full.  Choose lines that are horizontal, or vertical lines with flow going up.  Choosing a vertical lines with flow going down is asking for trouble.  Do not take samples from chests if you can avoid it.

3)  If you are planning to use your data to build a calibration for an instrument, you should make sure that the sample port is close to the instrument in question.  There is no point in running analyses if the instrument is in another line or on the other side of the mill.

4) The sample port should have an internal extension that protrudes roughly to the center of the stock line.Use proper sampling valves, if you can, and avoid ball valves that have been installed on the side of a pipe.  The image below illustrates how variable things can get as they move through your stock line.  As you can see, it can sometimes be a challenge to get that "representative" sample.  That said, your best chance will be to take samples from the center of the pipe as opposed to the sides.
Variability in a stock line


5) Open the valve and let the stock run freely for a few seconds to ensure that all the stock from the last sample is fully discharged from the sampling line.

6) Collect a large quantity of stock (i.e, a gallon or two at minimum - five gallons is better).

7)  When back in the lab, agitate your large volume of stock and take at least two small samples.  Analyze each according to your favorite method and average the results.  This will yield you one data point.

6) If you haven't done so before, run a Total Error Variance (TEV) to estimate the quality of your sampling and analytical technique.  TEV's are sort of a poor man's six sigma.  They will provide you an estimate of how much variability in your analyses is attributable to your sampling and how much is due to your technique.

If you don't have a TEV in hand, send me an email and I'll send you a copy of our spreadsheet that you can use.

No comments:

Post a Comment