| Author | Message | ||
     Tony Davies (td) Moderator Username: td Post Number: 131 Registered: 1-2001 |
Hi Sol, 1) Have you asked BOMEM? It sounds like an instrument problem. They will have a check list to help isolate it. 2) If you ask a new question; please start a new thread. Hope you get it resolved. Best wishes, Tony | ||
| Solomon Abebe (sol) Junior Member Username: sol Post Number: 6 Registered: 3-2006 |
Hi Bruce, I have a question about FTNIR instrument. (ABB BOMEM with GRAMS software) Recently, the spectrum i am getting from the spectrometer is different than usual for the same solution i had before. The difference is the recent spectrum not as smooth as use to be (noisy) and the absorbance spectrum drops to -1 when it suppose to start from 0 line. I took a background spectrum before taking 15 absorbance spectra the first 9 was okay except not smooth as they use to be but the rest of the spectra base line shifted down. When i check the single beam spectra they are different as well. How come the spectra shifted down? Is there something wrong with spectrometer parts such as the detector or the source? Thanks for taking time to help me. Solomon | ||
| Solomon Abebe (Sol) New member Username: Sol Post Number: 3 Registered: 3-2006 |
Bruce, you have understood it correctly. However, to do a calibration model on Kinetic spectra (spectra collected during cooling) is difficult because there will not be a reference concentration value for each spectrum. Unless an automatic means of measuring the concentration value employed while the spectra are taken. I am trying to use the existing calibration model to predict different cooling rate crystallization, I know there will be error of prediction but I am hopping it will be a small one. Regards | ||
| Bruce H. Campbell (Campclan) Moderator Username: Campclan Post Number: 86 Registered: 4-2001 |
Sol, If I understand you correctly, you did a calibration when the samples were in equilibrium and are now trying to use that calibration for samples that are not in equilibrium. If this is correct, you are introducing a new variable, the cooling rate and its effect on crystal formation. I would suggest instead doing calibration on a selected cooling rate. Then test it on validation samples to discover the variation, if any, from the reference values. If there is a sigificant difference in the validations using cooling vs. equilibrium you would have to choose which one to use. As I see it, the choice would be influenced by the ability to follow the cooling during actual sampling within acceptable variation (which I gather you really want to do) vs. a longer analysis time for using the equillibrium calibration. I am assuming here the equilibrium calibration would be more exact than the cooling calibration. Bruce | ||
| Solomon Abebe (Sol) New member Username: Sol Post Number: 2 Registered: 3-2006 |
Thanks for quick reply, The answer for your question is yes. You reason for my question sounds reasonable. Because, the calibration model is built on static measurement (taking each calibration spectrum during equilibrium). The error of prediction for 0.1deg.C/min cooling rate is smaller than the 0.5deg.C/min cooling rate. I bet, if i slow down the cooling rate i may get smaller error of prediction because, the spectra may resemble the calibration spectra. Am I right? | ||
| Bruce H. Campbell (Campclan) Moderator Username: Campclan Post Number: 85 Registered: 4-2001 |
Sol, I think your error of prediction is related to the crystal formation. With a slow cooling, the crystal formation will be closer to an equilibrium situation while with fast cooling there probably is a hysterisis effect. This all can affect the diffuse reflection in different ways. There is a question though, and that is what is the error of your validation set for the two situations? Does the validation data show the same characteristics as the error of prediction? | ||
| Solomon Abebe (Sol) New member Username: Sol Post Number: 1 Registered: 3-2006 |
Dear Bruce, before i ask the question i have described the situation. I built PLS calibration model on NIR raw and 2nd order derivatives transflectance spectra to predict solution concentration on cooling crystallisation of X organic compound from water solution. The calibration model built up on data collected between 15 to 80�C. That means make up a saturated solution of X organic compound/distilled water (a compound concentration to be predicted) at 15�C and take triplicate spectra and use the average spectrum as a member of calibration data and continue doing tha same thing for 20, 30, 40,to 80�C to build the entire calibration matrix. The cooling crystallisation experiment conducted on three cooling rate; 0.1, 0.5 and 1�C/ min. the process carried out as follows. Make up a solution of 23 g of X organic compound/500 ml of distilled water (saturated soultion temp. 70�C) in 500 ml reactor and heat the solution to 80�C and hold at that temperature (10�C above the saturation temperature) for an hour and half to completely dissolve the compound and cool at 0.1�C/ min. The kinetic spectra of FTIR and FTNIR taken during cooling period. The same procedure used for 0.5 and 1�C/ min cooling rate. When I use the calibration model to predict the solution concentration, which is 23 g of X compound in water solution before nucleation, I get low error of prediction on slow cooling (0.1&0.5�C/ min) but high error of prediction on fast cooling (1 �C/ min). My question is, why I do get high error of prediction on fast cooling crystallisation? |