| Author | Message | ||
     Donald J Dahm (djdahm) Senior Member Username: djdahm Post Number: 34 Registered: 2-2007 |
Waleed: It is admirable that you want to improve your �art� by practicing on sample data. At the risk of sounding like a broken record, don�t neglect learning the science behind spectroscopy and the statistics behind Chemometrics. Raphael: Pretreatments are done to enhance the differences in spectral features in which you are interested and/or minimize those in which you are not interested. Applied indiscriminately, they can have deleterious effects. While I don�t know the details of your situation, I assume that you are blending two powders. You probably say to yourself, I want the chemicals uniformly blended in the proper proportions, so I don�t care about things like particle size, so I will try to �remove some physical effects�. I will give a cautionary example, which may or may not be applicable to your case, but it illustrates the need for care. Suppose you have a �perfectly uniform� blend of which you have a standard spectrum. You have just the right proportions of two chemicals, each of which is in separate ingredients that are being blended. You now think you know what the relative peak heights in a sample of the proper chemical proportions looks like. Now suppose the next batch is also �perfectly� blended in the proper chemical proportions, but one of the ingredients had a larger particle size than it did in the standard batch, while the other stayed the same. The peaks from the material in the larger particles will be enhanced, and those in the other type particles will be diminished. You will conclude by comparing the spectrum from any place in the new mixture to the standard spectrum that it is not mixed well. Scary stuff. | ||
| Waleed Makarem (pharmad) New member Username: pharmad Post Number: 2 Registered: 11-2009 |
Dear Raphael , I am trying to use NIR in content uniformity. I need to learn the same point you are studying . I would appreciate if you supply me with sample data and the technique you used (if not confidential) . Thank you in advance for your support | ||
| Peter Tillmann (tillmann) Member Username: tillmann Post Number: 14 Registered: 11-2001 |
Raphael, do I understand this correctly? You use e.g. derivatives of the spectra and try to calculate SD/Ave. This can't work out well because derivatives will be ZERO at some wavelengths/wavenumbers. If you can't use oridary absorbance or reflection data use MSC as spectra pretreatment instead. Since you are using powers this is apropriate. Yours Peter | ||
| raphael beauget (raph) Member Username: raph Post Number: 12 Registered: 4-2007 |
Good morning, Firstly, thanks for your prompt replies. Spectra are displayed as absorbance spectra (I used a highly reflective ceramic as reference), then I applied some preprocessings like derivative 1st and smooth 17 pts (in order to remove some physical effects). My problem is to find how can i display RSD spectrum...I have two spectra : SD and Average but i can't display the ratio SD/Av cause some points in average spectrum are so close to zero. Is there somebody that have already met a similar issue ? Raph | ||
| W. Fred McClure (mcclure) Junior Member Username: mcclure Post Number: 44 Registered: 2-2001 |
Rapahel - A good way to make a reflectance standard is to take an ordinary bathroom tile to a milling machine and lathe, smooth the back to give you the desired size & thickness. I find that it works as well as the "store bought ceramics." Good luck, Fred | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 207 Registered: 9-2001 |
Rapahel - The reason spectra become negative depends on how the spectrum is being measured and displayed: 1) If the spectrum is displayed as a transmission or reflection spectrum, negative values mean that there is too little energy coming through the sample (if the measurement is by transmission) or coming back from the sample (if the measurement is by reflection). In this case you need to increase the energy being measured. This may require changes to the sample, changes to the way it's measured, or changes to the instrument, to illuminate it more brightly or collect the energy more efficiently. 2) If the spectrum is displayed as an absorbance spectrum (regardless whether it's measured by reflection or transmission), it means there is less reference energy than sample energy. You need to increase the reference energy. How you do this will depend on your instrument and the reference material in use, and again whether you're measuring in transmission or reflection. If you're measuring by reflection, you should NOT use an empty sample cup as the reference (as I once came across a case of). You should use a piece of highly-reflecting ceramic (or equivalent) that are usually provided with NIR instruments for that purpose. Howard \o/ /_\ | ||
| raphael beauget (raph) Member Username: raph Post Number: 11 Registered: 4-2007 |
Tony, Howard - I'm working on Bruker MPA and OPUS software v6.5 (lastest version). For answer to your question, the zeros in my average spectrum are the reason of this discontinuity (my average spectrum cut abscissa at several points). I thought apply some preprocessings in order to have only a positive average spectrum. Did you have already display a RSD spectrum ? Many thanks and sorry for my poor english Raph | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 204 Registered: 9-2001 |
Raphael - I'm not sure what you mean by "(discontinu spectrum)". Could you please clarify that term? \o/ /_\ | ||
| Tony Davies (td) Moderator Username: td Post Number: 182 Registered: 1-2001 |
Hello Raph! Are you working with a vendor data system or something more general Unscrambler or Excel? One thought; have you got some zeros in your average spectrum? Best wishes, Tony | ||
| raphael beauget (raph) Junior Member Username: raph Post Number: 10 Registered: 4-2007 |
Hello everybody, I'm trying to work directly on NIR spectra in order to check the Blend Uniformity. In this way, I'm calculating average spectrum and standard deviation spectrum from spectra collected at different locations of the blender. I'd have the RSD spectrum. Unfornately, SD spectrum divided by average spectrum couldn't give a good result (discontinu spectrum). I'd know if someone have already met the same issues and if you could have an idea for solving this problem. Thanks you Kind regards, Raphael |