| Author | Message | ||
     venkatarman (venkynir) Senior Member Username: venkynir Post Number: 72 Registered: 3-2004 |
Hello CH foo Can you spell out SG factors (data,windowidth,poly,dif/smooth) Have you take care of baseline correction ? Have tried with SNV /Maxi normlization method? More the samples with leave one out method surely will give good model to you . | ||
| C. H. Foo (chfoo) New member Username: chfoo Post Number: 5 Registered: 8-2008 |
As earlier posted I have done variations in each of the following steps: 1)At Quantitative/Regression/sample selection-select mahalanois distance, S-G then 2nd derivative and apply Maths and Save 2)Then Quantitative/Regression- use MLR and S-G +2nd Derivative. 3) PLS also tried -Calibrations are better with PLS better using factor 5 onwards. -Same sample in two different vials give different results. Validation on the vials have earlier been done by another colleague. -How much sample should we place in the vials and how compact? -Is it necessary to grind samples? Seem to get different result for grind and without grind Still trying to get my head around. | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 71 Registered: 3-2004 |
Dear C. H. Foo What type of pre-processing you have tried ? Have able to identfiy the outlier?. Have you tried for"leave-out" methods ? What criteria you are taking for fixing reproducibilty and repeatable results. | ||
| C. H. Foo (chfoo) New member Username: chfoo Post Number: 4 Registered: 8-2008 |
Thanks for all your help. I have been trying many combinations of pre-processing without success. I could achieve pretty good calibration curves (PLS method being better)but still could not get reproducibilty and repeatable results. I have tried grinding the samples but with no improvement. I cannot really find any particular wavelength that stands out. Please give suggestions. | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 70 Registered: 3-2004 |
Dear C. H. Foo In the case of Quantitative analysis your method is correct .However you can try different pre-porcessing steps to have a refine value and select that method of pre-porcessing as final one. Pre -processing plays vital role . More over the paper by ChaoTan et.al Analytical Science Feb.2007 vol.23 page 201 gives more information about Calibartion transfer using Wavelet. This is intersting one .It may helpful to you . | ||
| Tony Davies (td) Moderator Username: td Post Number: 178 Registered: 1-2001 |
Hello C.H. Foo, Welcome to the fun and complexities of NIR analysis. Tom Fearn and I have written a series of articles in my column in Spectroscopy Europe under the heading of "Back to Basics" (or B2B). You might find them helpful in "getting your head around NIR analysis". You are correct it is not like HPLC! You can download the columns at: http://www.spectroscopyeurope.com/td_col.html I suggest you start at April/May 2006. You have to do a lot of work to establish a calibration but when it is done you go on using it (until something changes [instrument, formulations etc]). Best wishes, Tony | ||
| iyas (iyas) Member Username: iyas Post Number: 13 Registered: 7-2007 |
did you entered , real samples form real batches analysed by hplc or else because hand made samples is not enough i think can you explain to me 50% to 150% Active 4g to 12g Lactose 93g to 85g Excipients 3g in all. ????? and what is the active material regards iyas | ||
| C. H. Foo (chfoo) New member Username: chfoo Post Number: 3 Registered: 8-2008 |
This is what I did ( I am not too familiar yet) 1)At Quantitative/Regression/sample selection-select mahalanois distance, S-G then 2nd derivative and apply Maths and Save 2)Then Quantitative/Regression- use MLR and S-G +2nd Derivative. 3) PLS also tried Others tried -the above with S-G and 3rd derivation PLS method appear to work better but when applied to analysing samples, results not reproducible. Anther question that I would like to ask is once a method is developed and validated, are routine samples done using the equation placed in the library or is it necessary to run calibration curve again for each day's run. | ||
| raphael beauget (raph) Junior Member Username: raph Post Number: 9 Registered: 4-2007 |
Hello CH foo, Which pretreatments did you use ? Cause, issue of reproductibility could be fixed by applying the right pretreatments (derivative, smooth and normalization for example). Cheers | ||
| C. H. Foo (chfoo) New member Username: chfoo Post Number: 2 Registered: 8-2008 |
Accuracy samples prepared at 50% to 150%of the theoretical active amount(4-12%w/w)as follows: 50% to 150% Active 4g to 12g Lactose 93g to 85g Excipients 3g in all. Total weight: 100g Is this the correct way of preparing accuracy samples for NIR analysis? I am getting my head around in this which is different from other validations eg HPLC/GC which I am familiar with. The instrument is Foss XDS rapid Content analyser using Vision software package. Thanks for responding. | ||
| iyas (iyas) Member Username: iyas Post Number: 12 Registered: 7-2007 |
can you tell us the material you are trying to measure and the device you are using and the formulation you are working on , and if raw material , what kind of tests you are using other than id , like content or , potency , or moisture | ||
| iyas (iyas) Member Username: iyas Post Number: 11 Registered: 7-2007 |
did you consider of making different concentration during your work , the number of the samples (,ore that 20) the concentration of the component your measuring (more 1% w/w i think ) | ||
| C. H. Foo (chfoo) New member Username: chfoo Post Number: 1 Registered: 8-2008 |
I am a new NIR user and have been trying to develop a quantitative method for in process blend powder with one active. So far I have been unable to achieve reproducibility. Calibration graph was not good when MLR was used but was better using PLS. Would this be due to difference particle size in the sample? I am about to try grinding the sample to see if there is any improvement. Please advise. |