response factor statistics for gas anaysis

[littlepeaks]

I work at a laboratory that does a proprietary method for gas analysis and separations using two dissimilar packed columns, each connected to its own thermal-conductivity detector. We currently perform a weekly one-point calibration on each column of each GC. A known amount of gas (from a calibrated-volume loop) is injected into each GC (once per column) and the response factor of the resulting peak is calculated. We do a control chart on each column, plotting the RFs -- plus or minus 3 standard deviations is in control, and you can't have, I think more than 7 points on one side of the mean in a row. The RFs are used to calculate amounts of that gas in gas samples for that week.

Even thought the gas samples are very clean, every once in a while we get a point out of control for one reason or another that we can't find a reason for (or more than 7 in a row on one side of the mean). I am uncomfortable doing the control chart this way, and wonder if there might be some other factor (such as RT, etc) or a better way of statistically monitoring GC performance. It seems like one of the problems is that the range of +/- 3 SDs is so small. Also, what is the effect on RFs caused by aging of the TCDs -- would they increase or decrease?

So far the problems we've experienced with the GCs are the "ubiquitous" gas leaks that occasionally occur, and the truly weird problems -- like the setscrews worked their way loose in one of the multiport valves, and the whole valve body was turning, so that the gas would never go to the second column.

Hope someone can help me with this "not so run-of-the-mill" questions. Before I started here, I have never worked with packed columns.

[chromatographer1]

You have a very thoughtful post.

With the right choice of packed columns, combined with the proper care and execution, their repeatability over thousands of injections is well known.

Counter point to that is if a rogue sample is introduced and changes the characteristics of the liquid phase or its support then changes can happen in the results, even though perhaps only in a single circumstance.

Likewise, if the column is not used consistently under the same circumstances differences in results can and will happen.

As in the real world, your situation may have special requirements and special circumstances, which of course, you deliberately avoided stating with probable good reason of propriety.

This also negates the likelihood of getting any profitable response to your questions.

Remember to avoid using the column 'cold', or not preparing the sample in the same manner every time, this includes purging water vapor entering the sample lines through leaks or diffusion through permeable seals.

Keep your detector in the same condition, 'all warmed up', before keeping any results of any injection.

Keep your columns and hardware also in the same condition, use a few blank runs to get the parts to a reproducible condition before any analytical work is performed.

Just like you would not start your car in the winter and immediately execute a quarter mile test run, preparation and warm up are essential to your analytical testing.

good measurements,

Rod

[Don_Hilton]

The range of +/- three standard deviation units may sound small, but given that it is the result of a measurement showing how tightly the results for the method should cluster, it actually a large range relative to the expected range of results. If you obtain measurements and the error is normally distributed, you have a chance of something like 2 out of 1000 that the results will fall outside the +/- 3 sigma range. Thus, if a result falls out of range, it is worth further investigation - or at least a repeat analysis because you have a strong posivility that something has changed.

On the question of using something like retention time rather than response factor: What is the measurement that is critical to the process? It appears to be the capabiltiy to provide consistant response for a level of gas. This this is a check on your paycheck. A check on retentention time may also be valuable, but if a change in retention time resutls in a change in measured response, you already have an alarm built into the system by alerting on a change in response factor.

Otherwise on establishing QC measures (And I've picked this up from someone else along the way): Don't just measure what you can easily measure; measure that which you need to measure (that which gives you useful information relative to the process).

[chemviren]

h

[walter]

Packed columns should be very robust for gas analyses. We have a system that has been in use for over 12 years and performs well. They still have a place in chromatography labs for some applications. If the factors are multiplication factors you usually observe an increase over time. The rate of change will depend on the samples you are running. Some analytes will degrade the filaments at a faster rate. This could also be dependent on the type of detector you are using, a single filament vs a dual filament detector. Factors can remain stable for long time depending on the application. An alternative to plotting the response factor would be to plot the concentration of a critical analyte in a control sample. That is essentially the same thing.