Sampling Shmampling



I get a lot of questions on calibration, but today, I want to talk a little bit about sampling, since it is absolutely key to take proper samples when you’re trying to build a calibration.



Why is sampling important?  It’s important because you are trying to get sense for what an instrument is telling you when it indicates something.  The only way that you can do that is to take physical samples of your process and compare the instrument’s responses with your manual evaluations of those samples.



There are two pieces to this.  Well, three actually.  One, of course, is the method someone uses to analyze a sample.  This is usually reasonably well documented, at least for the “Official” test.  TAPPI, for example, will publish an estimate for the repeatability of every lab test it recognizes  (the repeatability for the official consistency test, by the way, is 10%, which, to my way of thinking, ain’t so great. But hey, it is what it is).  Of course, if you’re not following the official procedure exactly, then your repeatability might not be as good as that.   I’ll talk more about this in another post).



The second aspect of this is just how representative the sample that’s being analyzed is of the process it was taken from in the first place.  If the sample you’re extracting from the process isn’t representative of the process, then you are basically analyzing something that really doesn’t mean much. Put another way, if your samples aren’t representative, then you are wasting your time with your calibrations.  You won’t get very far at all.



What do I mean by representative? 



Since it’s impossible to analyze absolutely all of your stock, you have to estimate what’s in your line by analyzing just a little tiny bit at a time – this is the sample that I’ve been talking about. If a sample is representative, then it means that you could have taken any number of samples in the same way and gotten roughly the same result. Of course, keep in mind that you won’t ever get absolutely the same result because the process isn’t homogenous, and no sampling method is absolutely perfect, but you can get reasonably close if you try.  Put another way, your samples will likely be close to the average of the stock in the line, and have a narrow two sigma.



If, however, the sample isn’t representative, then that means that you could get any number of widely different results each time you captured a sample.  You wouldn’t get samples close to the average, plus they would probably be biased one way or another, and your two sigma would be wide.

   

A good sampling regime is one in which a proper sampling valve is installed in a straight length of pipe of at least seven pipe diameters.  Valves are allowed to flow for a while to ensure the sampling line is flushed of any residual stock. 



A bad sampling regime would be something like a ball valve that’s just welded on the side of a pipe somewhere.  There is no thought given to the nature of the flow in the line at that point.  Is the stock flow stable, or is it turbulent?  Has the stock dewatered?  Was there some left over stock still in the sample line from yesterday or last week before you captured it?



And it’s not enough to ensure that your samples are merely statistically representative of the process.  You also have to ensure that both you and the instrument are actually looking at the same stock.



I really think that most people simply don’t pay enough attention to this last point. 



Why do I say that?  Here’s an example. 



I was once asked by a customer to help calibrate some of their equipment because they were having all sorts of problems and disagreements with their results.  They thought the problem was with the instrumentation.  As it turned out, it wasn’t the equipment at all, but with how they were sampling their stock.   

The equipment was installed in a stock line which then dumped into a chest.  The samples, however, weren’t taken from the same stock line as the instrumentation was in.  Instead, the samples were taken from the discharge of that chest.  The chest had a residence time of about 30 minutes, so whatever came out of the discharge was stock that had been mixed for thirty minutes.  There was no way that the lab could ever analyze the same stock that the instrument was exposed to. 



This situation was set up to fail.  It was guaranteed that the lab analysis and the instrument would always disagree because they were measuring two different things at different times.  Any effort expended under these conditions is a waste of time, because as the man from New England said,  “You just can’t get there from here”. 



When you install a sampling valve, you want to take care that it is close to the instrument that you are trying to calibrate so that you can be sure that both you and instrument are analyzing the same stock.



Let me also make the point that you shouldn’t balk at the cost of the sampling valve.  Yes, it’s more expensive than a ball valve, but it makes absolutely no sense at all to save a few hundred dollars on a sampling valve when you’re trying to calibrate a $50,000 instrument that will hopefully have a multimillion dollar impact on your process.  Saving those few hundred dollars may completely invalidate the whole thing.



So, here’s what you should shoot for when sampling your process for an instrument.



1)      Select a proper sampling point

a.       Site the sampling valve close to the instrument for which it is intended.

b.      Ensure that you will be sampling the same stock that the instrument is analyzing



2)      Ensure you are getting representative samples

a.       Use a proper sampling valve

b.      Install in a section of straight line at least seven pipe diameters long.

c.       Install in the side of the line, or according to the manufacturer’s recommended method.

d.      Open the valve fully when preparing to capture a sample.

e.      Allow the valve to flow to ensure that any residual stock is cleared from the line before capturing a bucket full



3)      Bracket the instrument analysis with Samples

a.       Capture samples of stock before during and after the instrument has completed it analysis.

                                                               i.      Capture a bucket of stock

                                                             ii.      Start the instrument analysis

                                                            iii.      Capture a second bucket of stock

                                                           iv.      Let the instrument complete its analysis

                                                             v.      Capture a third bucket of stock

Yes, it’s a lot of work, but the result is worth the effort.  You'll have captured meaningful samples that you can then analyze to build you calibration with.