Poor linearity FID Shimadzu GC2010

[AICMM]

So, just to clarify, you ran a 500 ng/ml mix (say at 1 uL), a 50, a 5, a 0.5, a 0.05 and a 0.005 ng/ml mix to get all these data points? And these are esterified FAMES, right?

Best regards,

AICMM

[EUROFINS]

Have a look at http://www.jhtolsma.nl/, here the amount on column is given, the highest amount is 250 ng on column.
Analytes are FAME = Fatty Acid Methyl Esther.

Jan-Hendrik

[AICMM]

Jan-Hendrik,

I went and looked at your data on your website. On the web page you say undecane in your graph and here you are saying FAME's. I certainly think undecane is a better candidate to look at linearity over FAME's considering all I have read about the issues with FAME analysis posted here.

On your web page you note you forced zero. What does the graph look like without a forced zero?

Your largest amount on column is 250 ng and you seem to go through 6 orders of magnitude. So your low point looks to be around 12 pg on column. That's way down there for an FID. Have you considered what the linearity looks like if you go up from 250 ng? If it really was FAME's and not undecane then 12 pg of FAME on column may be subject to reactive site loss??

Finally, you have not answered the question about splits versus actual dilutions of the solutions. The reason that is important is that you may have a split issue rather than a detector issue. I have seen two split ratio's 10 fold apart and they were not linear at all.

Best regards,

AICMM

[EUROFINS]

I compared undecane and decane with about 20 FAME's ranging from C4:0 to C22:6 and C24:0. All linearity graphs look the same!!!
I varied several injection parameters but it dit not have any influence on linearity, the only parameters that influenced linearity were the detector gas settings and Jet diameter.
My analysis ranges from 0.05% to about 85% relative fatty acid concentration. Not forcing by zero does not make sense because it creates large deflections in the low region of calibration.

Best regards,
Jan-Hendrik

[Joerg1]

Hallo,

Does anybody knows the end of this story?
I have the same problem now with my GC2010. Non-linearity with the FID with my FAME Applications. As posted, it is getting better with N2 instead of H2 as make up gas, but still not good.
FAME application with one point calibration. Check sample in the lower concentration is 90% recovery in the upper concentration is 120%. I changed: the septum, liner, syringe, fiz nossel, column, filter of the split/splitless injector. Baked out the system, asked shimadzu for help. Changed the detector gas volume, use He instead of N2 as make up gas. Checked the standards and application on a GC6850 with He as a column gas, 100% recovery.
Has anybody an idea? Thanks,

[Peter Apps]

Welcome to the forum.

To make sensible suggestions we need all the details of what you are doing; injector, column, detector, flows and temperatures and times. How do you make up your standards ?, what are they dissolved in, how do you inject them.

Peter

[Jake]

Just to clarify our issue: It wasn't that the detector alone was non-linear. We could take a single FAME and do a serial dilution over our range of interest and the coeffient of determination would be 1. The problem arises when you compare different FAMEs in mixtures present in differing amounts. Basically the slopes of the individual FAMEs are not parallel. So depending on the relative amounts, the FAME profiles could be close, or way off. We've noticed ALL GC/FIDs do this, but to varying degrees. The Shimadzu was the most egregious.

I happen to know that other companies have reported this issue to Shimadzu, and one big company said they would not buy more Shimadzu GCs unless the issue was resolved. Shimadzu's explanation for the descrepany in the data in the comparison of GC2010 data to that of a 6890 was along the lines of a sales pitch for the higher sampling rate capabilities and data smoothing capabilities of the 2010 (I think it was this; that was awhile ago and it's a bit foggy now). They claimed dropping the sampling rate of the 2010 gave data similar to the 6890, but alluded to which instrument generated the most accurate data (i.e., which instrument was 'right') was not determined. We tried this. Didn't really help. Apparently they're unaware that we can buy authentic standard mixtures of FAMEs at different relative amounts and demonstrate the Shimadzu always gave much poorer data in the comparison.

We eventually gave up and retired our Shimadzu 2010. We now have a lab 6890s and 7890A/Bs.

[Joerg1]

Hi,

Thanks’ for the answer, it do not sound very encouraging. Just to save some money, we decided to buy a GC210 from shimadzu and now we seem to have a big problem with the FAME analytics. I asked our shimadzu technician for help, but unfortunately, he also sees no mistake in the application or GC.
We tried a different nozzle with 0.8 and different gas flows for the detector with the same result of poor linearity. Therefore, it is not the FID. I will inform about our further progress.

[Peter Apps]

An FID detector is about the most linear and stable part of a GC. In order of lieliehood, problems arise on the inlet and on the column. That's assuming they are not operator-related.

Since different FAMES have different empirical formulas they have slightly different response factors on an FID, and so their calibration lines will not be exactly parallel, or parallel with the calibration for an alkane. Whether that matters or not depends on how accurate and precise you need the results to be and the contribution of other sources of uncertainty.

Peter

[Jake]

An FID detector is about the most linear and stable part of a GC. In order of lieliehood, problems arise on the inlet and on the column. That's assuming they are not operator-related.

Since different FAMES have different empirical formulas they have slightly different response factors on an FID, and so their calibration lines will not be exactly parallel, or parallel with the calibration for an alkane. Whether that matters or not depends on how accurate and precise you need the results to be and the contribution of other sources of uncertainty.

Peter

Peter - yes, that's true. And I'm assuming anyone doing FAME analysis understands how to use theoretical and how to determine empirical relative response factors for FAMEs. For whatever reason, the issue I've described is greatly exagerated on the Shimadzu 2010 compared to the various Agilent and Thermo GCs we've used for FAMEs, and as mentioned by others previously in this thread, even comparing to the previous Shimadzu model. Even switching to splitless injection didn't improve the problem for us. We did try the different jet also mentioned in this thread, but it didn't really help. As I've mentioned previously, switching to He make-up greatly improved the issue, but it all became too much of a chore. For single component analysis, or analysis where standard curves are being used for each analyte, the Shimadzu would be fine (we still have ours, in case that need arises), but for typical FAME profiling, it's a bust.

[Peter Apps]

Hi Jake

If you stick around on here for a while you will soon no longer assume that people understand things !

Considering how widely it is used, and how important it is to quantitative analysis in GC, the FID is remarkably poorly understood. I would hazard a guess that there is something about the geometry of the Shimadzu FID that differentiates it from the other makes - or perhaps it is the voltages they use. It is intriguing that helium improved matters - the main function of nitrogen as make up gas is to slow down diffusion in the bottom of the flame so that eluting molecules get comprehensively hydogenolysed (is that a word ?!) by hydrogen free radicals diffusing down from the combustion zone.

It sounds to me as if Joerg1 has a different problem to you anyway.

[Jason M]

Hi Everyone.

I've followed this topic for a while now with interest and I thought I'd add my experience to the mix.

I've known about this issue for close to 10 years now and it has been extremely frustrating. Firstly, it's weird and hard to diagnose. Secondly, because of the fact that it's weird, some other analysts may not believe it's a real issue and it can take some convincing to prove that it's the machine that's the problem and not something wrong with the analytical technique.

To summarise my experiences:

I analyse canola, sunflower and sometimes, camelina seed for fatty acid composition using area normalisation. These three types of seed have a number of fatty acids that range from extremely small up to huge for certain ones like C18:1. I use SGE BPX-70 columns.

I previously worked in a government lab where they had stopped using the GC-2010 and were sticking with the older GC-17A. I was told that the GC-2010 didn't provide accurate results and in AOF interlaboratory proficiency trials, the results from the GC-2010 would have made our lab look bad. I found this to be the case in my own tests. Basically, large peaks were too large, small peaks were too small and some mid-range peaks barely changed relative to the other fatty acids. So for example, from memory, if C18:1 was 65% on the GC-17A, it came out at around 67.5% on the GC-2010. C18:0 may have been 2% on the GC-17A but 1.6% on the GC-2010. The AOF interlaboratory results agreed with the GC-17A but the GC-2010 results would have been outliers in many cases. Also, I ran AOCS Standard Mix 6A on both systems. This AOCS standard agreed with the GC-17A but the GC-2010 was way out (>2 standard deviations from the mean for various fatty acids).

I contacted Shimadzu about it back then and for many months, while discussing this with one of their engineers, I tried all kinds of possibilities including gas flows etc as some of you have tried. In the end, nothing worked. So my manager arranged for the GC-2010 to be swapped with a GC-17A. After that, we had no more issues.

Swapping for a GC-17A is no longer feasible as this GC is no longer supported by Shimadzu and when parts run out, the GC will be history.

Fast forward to my current private lab where we have a heap of GC-2010s. When I started here, I originally wanted an Agilent machine because I didn't want the same GC-2010 issue to pop up and cause me grief. I was overruled and therefore, had to hope for the best. Luckily, the GC-2010 plus that I got gave accurate results. AOCS standard Mix 6A results were accurate. We also got a Shimadzu GC-2010plus GC-MS/FID system and it too provided good results.

However......

About a year or two ago, the FID jet got blocked on the GC-MS/FID system. We removed it, unblocked it and reinstalled it. After that, the results were bad like they were in my previous job- large peaks gave results that were too large, small peaks gave results that were too small mid range peaks were almost unchanged.

Then last year, as part of a new preventative maintenance program, the FID jet was replaced in the other GC-2010 that was working fine. After that, the results were bad.

But now it was getting clear that it was directly related to the FID jet. I asked for Shimadzu to get involved again to try to sort things out. They were able to replicate my issue in their lab using my column and samples. While they tried to sort it out, I tried a few other things. Based on comments in this forum, I tried using a 0.8mm jet rather than the 0.3mm jet. The results improved somewhat, but the sensitivity dropped too much. Plus the results were still a bit skewed.

So I asked one of the Shimadzu engineers to come in work on the issue. I asked him to bring a heap of FID jets as well. After a number of failed attempts at modifying some parameters, the FID jets were swapped in and out one by one and voila! We found one that happened to bring the results back in spec. Since then, it has worked as brilliantly as before and I've had no more issue. Noone is allowed to touch the FID jet in that GC anymore! The good new results were confirmed by AOCS Std Mix 6A and also by good fatty acid results that our lab got from a Canadian interlaboratory certification program.

I had no such luck with another GC 2010 that I had recently installed. It worked fine at first. But then the large peaks became larger, the small smaller. It wasn't as large a shift as for the other GC. But it was concerning enough as there would now be less agreement between this GC and the one that was fixed. After trying about 5 FID jets, none could be found that brought the results back. So this GC isn't being used for fatty acids at the moment.

While all this was going on, a Shimadzu analyst gave me some gas flows that brought the results back into spec. I tested them on a couple of GCs and it worked. The only down side is that the sensitivity dropped off. But for a GC that was out of spec, it brought it in spec. For a GC that was already in spec, it made virtually no difference.

The gas flows for my system are:

Makeup gas (nitrogen): 5mL/min
Air: 270mL/min
Hydrogen: 30mL/min

To improve sensitivity, I was given another set of flows:

Makeup gas (nitrogen): 20mL/min
Air: 175mL/min
Hydrogen: 40mL/min or 30mL/min

The new flows did improve sensitivity. But I've had more trouble with noise. I had to flip the FID lid closed to stop air movement disrupting the flame. These flows are still a work in progress. There may be better flow rates that improve accuracy, sensitivity and flame stability.

I've been informed that Shimadzu is bringing out a new GC model with a redesigned FID. Hopefully, this new FID works far more accurately and reliably than the current GC-2010 one. If so, it's been suggested that it might be able to be swapped directly with the one in the GC-2010 to save having to buy the new model GC. To test it properly, it needs to first of all work in the new GC. Then it needs to work in the GC-2010. Then it needs to work no matter how many times the FID jet is swapped with another. This last one is crucial because I've already experienced GCs that work fine from installation. But once the FID jet is replaced for the first time or removed and reinserted, the problem suddenly appears.

So time will tell if this issue will be resolved once and for all.

Jason

[Peter Apps]

Hi Jason

Thank you for the very detailed story - if nothing else it confirms that the Shimadzu FID on the so1o had some "idiosyncracies" shall we say.

In all of this long saga, did anyone think to take a 0.3 mm jet and ream it out to 0.5 mm ?

Peter

[Jake]

Hi Jake

If you stick around on here for a while you will soon no longer assume that people understand things !

Whoops - silly me. Good point there. From what I've seen, I should no better. What I've noticed is that the folks with the most knowledge and experience with GC/FID analyses are the ones who know just how difficult it is to do correctly.

@Jason - as with Peter, thanks for the detailed explanation of your situation. When we first noticed this, I couldn't believe we were the only ones doing FAME analyses that noticed the unacceptable GC2010 behavior. Youre 'big peaks too big, small peaks too small' statements bring back bad memories! The real troubling thing to me is that you can pass AOCS system performance requirements with a 2010 since in the standard mixture used to establish empirical relative response factors, all the FAMEs are present in equal mass amounts. But once you start measuring real samples, your profiles are going to be way off. This can be readily demonstrated with authentic standard mixtures from NuChek Prep. What I've learned at my current gig, where we've now processed nearly 500,000 FAME samples, is to just have NuChek Prep make authentic standard mixtures of our target FAME profiles. Very enlightening to how bad some contract labs are for FAME profiles. My guess is they just ensure their instruments meet AOCS spec performance but never test authentic standard mixtures.

I've always enjoyed my relationship with Shimadzu - sales people, field engineers, regional managers, etc. have all been great to work with. I've even visit Shimadzu in Kyoto on their dime. Sounds like Shimadzu in Australia has been very helpful too. I was having lunch recently with a group of Shimadzu folks, and one of the field engineers mentioned he hadn't visited us for awhile. I told him our 2010 had been retired, and the reason why. He said he hadn't heard anything about our issue and was surprised. So it sounds like communication could be a bit better at Shimadzu.

And back @ Peter - no, we never tried modifying the jet. By the time it got to that point, we were up and rolling in high-throughput mode in the lab and didn't really have the luxury of instrument time to trouble-shoot any longer. We were processing close to 5000 samples/week, so everything was running 24/6.

[Jason M]

@Peter. It would have been good to test a 0.5mm jet. But as the 0.8mm jet still produced results that were a bit out, I think the 0.5mm jet would probably have been a bit more out than the 0.8mm jet. And yes, there are definitely some idiosyncrasies with the GC-2010 FID. The GC-17A/FID systems that I've used have always been good with area normalised fatty acid results.

@Jake. Yes I agree. If you test relative response factors with peaks that are the same size, you'll certainly be able to do relative response calculations. But the real issue comes in when the peaks are a variety of sizes and relative response differences are swamped by the much larger non-linear responses of varying peak sizes. I'm keen to get the NuCheck Prep standards that you mentioned as these have certified mass percentages and there are mixes that are similar to canola. So these standards are more definitive than a standard like AOCS Std Mix 6A which has the averages of 78 labs as the reported results on the certificate.

And yes, Shimadzu Australia have been very helpful and have made a great effort in investigating it in their Sydney lab and trying to find a solution. I believe that this issue has been communicated back to Japan and hopefully the new FID design addresses it. Overall, Shimadzu have great products. So I hope this issue is addressed once and for all.