https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html https://www.euro-online.org/enog/inoc2007/Papers/m https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html

Recovering from a loss of a calibration, and software transfers of

ianm's picture
Forums: 

4. Recovering from a loss of a calibration, and software transfers of calibrations

Time for another question on the problem of transfers between instruments. What is the most important part of the system that needs addressing? The spectrometer? The software? The transducer (be it a transflectance probe or diffuse reflectance one)? All of the above?

An associated question is: One thought that almost brings me to a cold sweat is having something happen that would lead to loss of a calibration. What is the weakest part of any system? How can one plan to replace that part if needed? I don't see the computer part being very critical, as long as the hard drive is backed up frequently enough. That leaves the spectrometer, the fiber optics, or the transducer.

Any comments?

First response

Bruce:

I haven't done much of that sort of thing, but some observations.

Mike Zapf just told me that he was informed that if the light source goes on an instrument working in the 400-1100 nm region, you're talking about starting from scratch. We had the detector go on a 6250 and I can tell you that the energy curve for the new one was completely different. Had more energy at one end of the NIR spectrum and less at the other. Wavelength precision or reproducibility from one instrument to the other seems to be a big item. It's one of the things the Fourier people push for their instruments, claim that transfer is much easier. At best, its a lot of work to avoid the problems your talking about. Shenk's software requires you to run a set of standards every week with the instrument at the same temperature in order to track the changes in the instrument. You then make the new instrument or a fixed one produce the same spectra by altering the spectra to look alike. I asked some others once what they do and the answer was nothing, just hope nothing happens. Most of my studies have been short with a few hundred samples at best so I don't worry about it, but recently I had a student run 1800 samples and was quite concerned. Overall, I think anything that changes in the spectrometer, or even things like new grinders, can cause bug problems with calibrations. I suspect that a grating change would be a lot of fun, but almost anything is bad news.

Jim Reeves

Second response

Validation!

In general, it is not easy to control all of the hardware components closely enough for direct calibration transfer. It is rare that a calibration will perform adequately without being remapped to the altered hardware.

The issue becomes how to best remap. Fortunately, in almost all cases the answer is quite simple: (cover you ears, some people will shout their disagreement with this assertion) post-process the output of the calibration with an offset and a slope as necessary to bring the results back into line. Then, validate the remapped calibration with an appropriate number and variety of validation samples.

An associated question is: One thought that almost brings me to a cold sweat is having something happen that would lead to loss of a calibration. What is the weakest part of any system? How can one plan to replace that part if needed? I don't see the computer part being very critical, as long as the hard drive is backed up frequently enough. That leaves the spectrometer, the fiber optics, or the transducer.

Any comments?

Cold sweats are healthy. In any case they are unavoidable when deploying chemometrics based analyzers. Again, the three most important things in this context are validation, validation, and validation.

Regards,

Richard Kramer

Third response

I basically agree with Richard Kramer's response. I have presented same ideas in the following two papers:

1. "Practical approaches to transfer NIR calibrations", Paper # 432, Eastern Analytical Symposium, November 1996.

2. "Calibratin transfer: Are multivariate methods needed?", Paper # I-020, IFPAC '97, February 1997.

If any of you interested in presentation materials, e-mail your request to [email protected].

Please also refer to: E. Bouveresse, et. al., Anal. Chem. 1996, 68, 982.

Regards

Alagappan Annamalai

From: Howard Mark

Bruce - here's what I got from manufacturer's literature (manufacturers of sapphire, that is, which, by the way, is the best place to get info about optical materials. NIR manufacturers do not generally provide this sort of info about components):

1) Sapphire is unaffected by common acids or bases below 1,000 deg C

2) It is unaffected by HF below 300 deg C

3) The only material harder than sapphire is diamond. This makes it resistant to ordinary abrasion. However, since it is a brittle material, sharp blows could conceivably make small pieces flake off, which could be equivalent to abrasion. On the other hand, a sharp enough blow would be more likely to break the entire piece.

Howard