Author |
Message |
Tony Davies (td)
Moderator Username: td
Post Number: 242 Registered: 1-2001
| Posted on Wednesday, August 04, 2010 - 9:12 am: | |
Hello Venkynir, The correlation coefficient is influenced by the range. The most important information is given to us by the SEP which answers the question "How variable are the answers from this model"? If the SEP is less than that required for the analysis then you have a method. R and r^2 are important but not the most important or useful statistics. Best wishes, Tony |
venkatarman (venkynir)
Senior Member Username: venkynir
Post Number: 112 Registered: 3-2004
| Posted on Saturday, July 31, 2010 - 8:26 am: | |
Hi Tony I do agree with you we have to do different pretreatment and SE but your statement " Correlation is not the most important statistic in regression analysis" when your are linking two methods(here HPLC & NIRS) we have to think about it?. Is it |
Tony Davies (td)
Moderator Username: td
Post Number: 240 Registered: 1-2001
| Posted on Saturday, July 31, 2010 - 3:19 am: | |
Hello Deepak, The main prblem for this analysis is the very small range of your analytical values. A 2.5% range is very small for regression analysis. How good is your HPLC method? What is the standard error of repeatablity (SER) for "bind" duplicates? [Blind means that the analyst does not know that they are analysing the same sample]. I can suggest an alternative approach. I assume that you have a specification for the product. You need to collect sets of samples which (by your HPLC method) are either above or below the specification allowing for the possible errors in your HPLC method ( i.e they are 3 x SER above or below the limit). Then you test new samples for identity to either group. You may need to test different pretreatments of the spectral data. One other point. Correlation is not the most important statistic in regression analysis. The standard error of prediction (SEP) which is the error in predicting the independent validation samples is the first test of any NIR method. You can compare this to the SER for your reference analysis. Hope this helps, Best wishes, Tony |
Deepak Sharma (deepak)
Junior Member Username: deepak
Post Number: 7 Registered: 11-2007
| Posted on Saturday, July 31, 2010 - 12:08 am: | |
Dear Sir Let me elaborate. 1 It is a fine free flowing powder. 2 I want to measure both (physical and chemical)properties but the basic is to determine purity of the sample by quantitative analysis against a referee method i.e a HPLC. 3 Material is white powder. 5 As discussed , inferential method is a HPLC quantitative method (Assay method) 6 we have samples in training set and validation set point 4 and 7 are not clear to me, can you pl explain for me . We are using FOSS NIR System for analysis. we are getting a correlation of 0.2 and residual difference between NIR value and inferential method value is about 0.8 during prediction. Our working range for samples is 97 % to 99.5 % i.e samples in our calibration model is having assay from 97 to 99.5 %. Hope this will help you to understand my problem. Pl suggest, can we go ahead with this data. thanks / Regards Deepak |
venkatarman (venkynir)
Senior Member Username: venkynir
Post Number: 111 Registered: 3-2004
| Posted on Friday, July 30, 2010 - 1:23 am: | |
Deepak &Aniruddhapisal Deepak your enquirers is not clear . 1. First powder materials homogeneity is important. 2. What do you want measure either physical or chemical / both . 3. Colored material( NIRS not prone for color ) but it helps training set clustering 4. Source and tolerance permitted 5. Training set samples and inferential methods accuracy and its standardization( approved by FDA/AOCS etc.. ); 6. Number of training set ,validation set & prediction set 7 If the case of DAS resolution required if above said all determined first than we can think about your questions . I have seen calibraiton become easy now-days with helps of chemometrics. |
aniruddha pisal (aniruddhapisal)
New member Username: aniruddhapisal
Post Number: 2 Registered: 7-2010
| Posted on Thursday, July 29, 2010 - 11:17 pm: | |
Dear Deepak, plz create a new question as you have used somebody else's platform. (and me too) plz. execuse. |
aniruddha pisal (aniruddhapisal)
New member Username: aniruddhapisal
Post Number: 1 Registered: 7-2010
| Posted on Thursday, July 29, 2010 - 11:11 pm: | |
Dear Deepak, Good question. This is really important for a chemist/NIR user who are interested in qualifying the materials. However, as the question is more general (there is no information regarding the calibration model, validation, sample prediction values etc.) I feel you have to follow the systematic approach before the sample analysis. It includes, Instrument setup, creation of spectral library, method development and validation, Understand the validation results, set the acceptance thresholds and then go for the samples of interest. i know the suggestion is very general but as a first step moving towards the NIR testing you need to understand all these segments. I would be happy to explore this discussion. To All/NIR experts, I request you to explore this topic which would give a good answer to this querry and will enhance the popularity of this technique. |
Deepak Sharma (deepak)
Junior Member Username: deepak
Post Number: 6 Registered: 11-2007
| Posted on Thursday, July 29, 2010 - 12:56 pm: | |
Hello Everybody I need some help over calibration model of powder material using reflectance probe. How important is a correlation for a calibration equation developed for a powder material. If i am not getting a good correlation between NIR predicted values and referee method values, however samples predicted against this equation give me a good result.Is there any acceptance criteria for correlation or high correlation value is necessary. Pl help. thanks/regards |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 340 Registered: 9-2001
| Posted on Friday, July 23, 2010 - 4:30 am: | |
Venky - yes, you're correct: I should have qualified my statement and said "I don't know of any NIR instruments that still use them currently ..." In fact, photomultiplers may also still be used in UV and possibly even in instruments for the Visible spectral region, in situations where their extreme sensitivity is needed - photon counting is obviously the limit for that! \o/ /_\ |
venkatarman (venkynir)
Senior Member Username: venkynir
Post Number: 110 Registered: 3-2004
| Posted on Thursday, July 22, 2010 - 11:47 pm: | |
Hi Howard Mark , Photomultiplier used in beta counter for radio active measurement . You meant to say that Scintillation counter /photo multiplier could have used for cost effective way low end NIRS. |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 339 Registered: 9-2001
| Posted on Thursday, July 22, 2010 - 8:47 pm: | |
Tony - you said "If you read in a paper "SW-NIR" you need to know what range it means.' I say 'If you read in a paper "SW-NIR" you need to know what range it means TO THAT PERSON.' We can't get away from it: we need to have the numbers specified. In an ideal world, there would be general agreement about these things within the scientific community, and they, in turn would make the journals create standards for publication, that all the journals would agree to, as well. Then someone could say "SW-NIR" and everybody would know what it meant. In the real world, I'm afraid my statement still stands (unfortunately). And the spectral range of an instrument is still determined mainly by it's detector. I just realized I left out one historically important class of detectors. I don't know of any instruments that use them currently, but for many years photomultiplers were the detectors used in "NIR" instruments. Those were extremely sensitive (photon counting was possible) but they limited the upper end of the wavelength range to about 900 nm. When I was working at Technicon, one of the physicians called the wavelength range that the (then-new) NIR group worked in "Far-IR", because he was used to dealing with photomultiplier-based instruments! \o/ /_\ |
Tony Davies (td)
Moderator Username: td
Post Number: 239 Registered: 1-2001
| Posted on Thursday, July 22, 2010 - 2:54 pm: | |
Hi Chia, Pleased to help. Hi Venkatarman, The energy range for the complete NIR range is given as 255 x 1O^-21 J (at 780 nm) to 79.5 x 10^-21 J (at 2500 nm) so about 180 x 10^-21 J at 1100 nm. The penetrating power is very sample dependent but as David said it is much larger for the Herschel range. { I'm not sure that Penetration Power is the best way to think about it. The process is governed by the probability of an absorption occuring) In agricultural samples much of the "sample" is air and most samples have high scatter so the process is diffuse transmission rather than "transmission". I'm not sure what you mean by "combination" but if you are refering to the vibrations that occur then they are mainly 2nd overtones and 2nd and third combinations. As David suggested the use of the Herschel range is considerable. It is very import for agricultural and medical applications. Hi David! In general I agree with you but the sample always comes first. Then you can decide what ought to be the best wavelength range. Hi Howard, I think the question was quite valid and required an answer. If you read in a paper "SW-NIR" you need to know what range it means. It would be good if we could have an international agreement but then we DO have an international agreement but people still say "the NIR region about 800 to 2,400 nm"!!! Best wishes to all, Tony |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 338 Registered: 9-2001
| Posted on Thursday, July 22, 2010 - 12:25 pm: | |
Chia - I think the various comments can be summed up by saying that the terminology isn't what matters; different people use different names for the same things, and the same names for different things. What matters is what the physical and chemical effects are, that can be detected in different wavelength ranges, and what instrumentation you need to make the measurements that do the type of analysis you need. The main single determinant of that, is the detector that an instrument contains, most of the rest of the instrument is designed around (or at least in conjunction with) that. Silicon (Si) can go to approx. 1100 nm. Indium-Gallium Arsenide (InGaAs) can go to approx. 1700 nm Extended Indium-Gallium Arsenide (InGaAs) can go as far as 2200 nm, depending on the exact formulation used. Lead Sulfide (PbS) can go to 2500 nm Other types of detectors are rarely used in conventional NIR. After that, the most important factor is the type of sample that an instrument is designed to measure. A spectrometer for clear liquids is very different than one intended to measure powdered solids, for example. So don't worry about the name of the wavelength region, it's much more important that you get the right numbers. Howard \o/ /_\ |
David von Boisman (david_von_boisman)
Intermediate Member Username: david_von_boisman
Post Number: 20 Registered: 2-2006
| Posted on Thursday, July 22, 2010 - 9:24 am: | |
In most cases for agricultural products at least, the range 700-1100 is used for transmission measurements, while 1100 - 2500 (or parts of these ranges) are used for reflectance measurements. In the lover range, light may penetrate for pathlengths up to 30 mm depending on the sample type, while at 2000 nm it only penetrates for fractions of millimeters. The only exception is for liquids, where in some cases you can work in transmission up to about 1800 nm. |
venkatarman (venkynir)
Senior Member Username: venkynir
Post Number: 109 Registered: 3-2004
| Posted on Thursday, July 22, 2010 - 8:50 am: | |
Hello Tony can You brief more about energy ,penetrating power and combination at "Herschel range" to the new NIRS user . I understand few food analysis and gasoline have good application .Is it correct ? |
Chia (xaivier)
Junior Member Username: xaivier
Post Number: 8 Registered: 4-2010
| Posted on Thursday, July 22, 2010 - 8:43 am: | |
Thanks, Tony. It is very comprehensive and useful for me. |
Gustavo Figueira de Paula (gustavo)
Member Username: gustavo
Post Number: 11 Registered: 6-2008
| Posted on Thursday, July 22, 2010 - 7:07 am: | |
Hello guys, There's another practical point regarding 1100 nm as limit for SWNIR - silicon detectors works up to ~1100 nm - beyond this you will need InGaAs or PbS detectors. Gustavo. |
Tony Davies (td)
Moderator Username: td
Post Number: 238 Registered: 1-2001
| Posted on Thursday, July 22, 2010 - 6:44 am: | |
Hello Chia! The range of near infrared is specified by IUPAC as 780 - 2,500 nm. There is no official definition for SW-NIR but it is commonaly thought to be 780 - 1100 nm. In 1988 I suggested that it should be called the "Herschel range" as it was William Herschel who made the discovery that there were invisible radiations beyond the red end of the visible spectrum. Some people have adopted it! 1100 or 1200 is quite arbitary but the early NIR instruments used filters to select first and second order reflections from gratings so this gave them 1100 - 2500 and 400 - 1100nm. There are very few absorptions around 1100 which is another reason for it being the best break between the two ranges (I also suggested that the 1100 - 2500 nm range should be called the "Norris region"). I hope you know why! Best wishes Tony |
Chia (xaivier)
Junior Member Username: xaivier
Post Number: 7 Registered: 4-2010
| Posted on Wednesday, July 21, 2010 - 9:34 pm: | |
Hi, I am confusing about the wavelength range of Shortwave NIR and NIR. Some study said SW-NIR is from 800-1050 nm. But some study said NIR is from 800-1200 nm and SW-NIR is from 1200-2700nm. So, which one should I follow? Thank you. |