| Author | Message | ||
     Gabi Levin |
Hi Jack, Howard made an excellent point. Possibly I can add the following: The peaks in NIR are not resolved as in Mid-IR. Thus, small absorptions due to the chemical you want to measure become obscured by other absorptions from other constituents. In powders you also have scattering effects that increase the "noise" and make it worse. Exception - if you have high absorbing compound such as water in low absorbing organic compound. I have successfully measured moisture in dichloro benzene to 20 ppm because of the hign absorbance of water in the 1930nm and the near zero absorption of dichloro benzene there. Most important to do in those events when someone finds a correlation that seems good: 1. Obtain a pure sample of the chemical you want to measure. 2. Collect spectrum of the pure chemical under same conditions as you will run in real life. 3. Analyze the "loading weights" ( a very useful feature I use in the Unscrambler) for the first several PC's of your regression. If you can not verify that you have high coefficients in the same wavelengths of the peaks of the chemical, most likely your regression is based on phenomenon as described by Howard. If this is the case, chances the calibration will not hold water for long. Soon enough you will run into samples that will not fall nicely into the same relation between the chemical you quantify and the chemicals on which the regression is truly founded. When this happnes your predictions will bear no relation to the true situation. I hope I was able to add a little. Gabi levin Brimrose Corp. | ||
| hlmark |
Jack - mainly because of the weakness of the absorbances in the NIR region. By the time you get concentrations of down to, say 0.1%, the absorbance bands are approaching the noise level of the instrument and, equally important, the drift level, so that you cannot completely rely on the meaningfulness of what the instrument is reporting. Some people have reported analyses of down to part-per-million sensitivities. Because of the above considerations, there is always a suspicion that what is actually being measured is a constituent of higher concentration that is correlated with the constituent of interest. As a practical matter, that might be OK, as long as you can rely on that correlation continuting to occur. If anything should happen that breaks the correlation, then your results will be greatly in error. Howard \o/ /_\ | ||
| Jack Lee (Jack) |
I know it is not a good trace analysis method, but I don't know why? I can't find it. If you know why, please reply. Thank you very much |