| Author | Message | |||
     Lez Dix (lez_dix) Member Username: lez_dix Post Number: 12 Registered: 10-2006 |
Usual welcomes James I have a colleague who works a lot with Gramms, FT-NIR and animal feed. I must stress as an instrument supplier we have a vested interest so can I ask what FT-NIR you are using for this work? My colleague may be able to help, feel free to email me if that is easier. Lez | |||
| Ana Garrido-Varo (garrido) Junior Member Username: garrido Post Number: 10 Registered: 1-2006 |
Hello James and all, This is a message from Cape Town one day after the end of NIR2011. I am quite happy to see that while some members of the NIRS community were learning from others in Sudafrica, that discussion forum was as usual quite live. Thanks again to all the contributors. James, we have experience with the prediction of the percentage of ingredients in compound feedingstuffs. One of these ingredients was salt. We got quite nice results (non with PLS but with LOCAL). We used quite large data sets for cal and val and the reference data were the percentages provided by the formulation department of a feed mill plant. We published that info in APPLIED SPECTROSCOPY 69, Volume 59, Number 1, pp. 69-77. 2005. The paper will help you to understand how and why we used LOCAL, but please do not believe the SEP values for salt. I just noted that they are wrong. We changed units (% by g/kg) along the text and we made some mistakes that were not noted either by the referees nor by us until today!!!!. For that reason I prepared for you a doc file with the right results for salt. I am attaching the file to this post. If I do something wrong (which is not rare to me...) and you do not get the file, please let me know. Furthermore, I would like to know the mean and range of your salt content values for cal and val sets. I did not understand well your sentence "The calibration does not hold true when a sample is outside the range. ". If I uderstood well you were trying to predict a sample with a salt content that was outside the range of the calibration set. Thus is normal that you can�t predict well the sample with your calibration. Please could you e-mail to me with additional information on your data set?. Regards Ana
| |||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 425 Registered: 9-2001 |
James - regardless of the theory, or the statistics, it seems that in practice, your calibration models are not sensitive enough to the salt cnncentrations in your sample to show a difference between samples with different amounts of salt. The results you describe, where the calculated value is always around 0.3%, indicates that the model is not sensitive to the salt concentration, but is simply predicting a value around the B0 value of the model, which is probably also close to the mean salt value in your calibration set. If you set all the calibration coefficients, except B0, to zero you would get that same result. As for books to read, IM Publications (the sponsor of this discussion group) offers a selection of books about NIR and calibration information. At a more advanced level is the Handbook of Near Infrared Analysis, which is published by CRC Press but I think you can also obtain from IM Publications. \o/ /_\ | |||
| James Mitchell (bones29) New member Username: bones29 Post Number: 2 Registered: 4-2011 |
Thanks for the welcomes and the replies. We have always thought it would be difficult and with conflicting views from different people have confused things. The moisture in the meal is <14% and is a 'dry' meal. After reading the recommended paper I think having a low moisture content and a very low salt content <0.5% has compounded the difficulty. As stated in my orginal post I'd be happy just to know if the salt is outside the ideal range but my models keep putting them in the ideal range. Being new at all this would it be possible to be advised if my method is correct on developing a calibration on Grams? I current do the following: Run data set without any preprocessing. Then start with the smoothing ie SG and GAP up to about 13 points. Then try different pathlenght corrections - but have found these seem to degrade the models. I then pick the model with the highest R^2 and then start to remove the outliers (which identifying I find to be difficult sometimes) until Ive removed no more then about 10% of the samples. Ive been told that the lower the factors the better but Im confused when it comes to which is better a high R^2 with a high factor or a lower R^2 with a lower factor - ie 0.69R^2 with a factor of 18 against 0.54R^2 with a factor of 8. Again many thanks for the help | |||
| Tony Davies (td) Moderator Username: td Post Number: 256 Registered: 1-2001 |
Hello James, Welcome to the world of NIR spectroscopy and the Discussion Forum! Back to the original question; Can NIR determine salt content of animal feeds? (and if so why?). The quick answer is �Yes�. There are some quite old references but a good one (which you can get from this website) is: Determination of the sodium chloride content of sausages by near infrared spectroscopy M.R. Ellekj�r, K.I. Hildrum, T. N�s and T. Isaksson MATFORSK - Norwegian Food Research Institute, Osloveien 1, N-1430 �s, Norway. JNIRS 1, 65, (1993). My group at IFR Norwich published a paper in 1989 in which we reported the simultaneous determination of sodium hydroxide, chloride and carbonate in solution. A. Grant, A.M.C. Davies and T Bilverstone, Analyst, 114,819 (1989). The reason it works is because when these anions are in solution they hydrogen bond with water molecules and because they have different sizes (and other physical properties) the hydrogen bonding is subtly different and so the different species can be distinguished. Water has a very complex spectra. It may appear as broad bands but this is due to a large number of different absorptions. I�m not sure about animal feeds. I expect that the water content is quite low and exists in your samples as bound water. There may well be some literature telling you if you can successfully analyse NaCl in the dry sample. If not it might be worth preparing water extracts of your samples and looking at their NIR spectra. Best wishes, Tony | |||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 421 Registered: 9-2001 |
Gustavo has it almost right. Many different combinations of (metallic) cations with various halogens and other anions are used for window materials in the mid-IR, and they all have different properties. Gustavo missed one of the most important properties, however: the range of wavelengths that a window made of the different materials can transmit. A window made of NaCl, for example, will transmit IR to about 25 micrometers, while CsI (cesium iodide) will transmit to over 50 micrometers. Even some of our common "NIR"-type windows can be used for mid-IR work over limited ranges; sapphire, for example, can transmit to roughly 5.5 micormeters. Other important properties, which can also vary widely, are hardness, strength, resistance to attack by moisture, stability, cost, ease of working and other factors. As Gustafo mentioned, for example, ZnSe is useful for resistance to water, but there are other salts that are also resistant to attack, including silver salts (AgCl, AgBr and AgI) and various proprietary formulations, e.g., KRS-5. Manufacturers of windows for mid-IR usage can provide charts listing the various different properties for a wide range of window materials. \o/ /_\ | |||
| Gustavo Figueira de Paula (gustavo) Advanced Member Username: gustavo Post Number: 24 Registered: 6-2008 |
Francesco, all halides of alkaline metals are suitable for MIR or NIR windows because are very poor absorbers. ZnSe is also transparent, but not hygroscopic - this is why ZnSe is always prefered as ATR windows. KBr is very common as vehicle for powdered samples for transmittance measurements, but NaCl is suited as well. Alkaline earth metals salts are also suited, and being less soluble than alkaline salts, are pretty useful for liquid cells. CaF2 is a common one. | |||
| Francesco Davini (franz) Member Username: franz Post Number: 15 Registered: 2-2009 |
uhm, isn't it normally KBr? Chemically not much different but... I didn't know about the use of NaCl. I thought it was too hygroscopic for this purpose.... but I'm not a IR guy! | |||
| Fernando Morgado (fmorgado) Senior Member Username: fmorgado Post Number: 26 Registered: 12-2005 |
hello : If you remember in IR ( not NIR) using ATR the sample is put over a NaCl windows as support. The reason, Nacl Don�t show absorbance in IR , the same in NIR. The same for any salt added to the sample. Gustavo have the reason, NIR is not the better aproach for your analitical problem. Fernando | |||
| Gustavo Figueira de Paula (gustavo) Advanced Member Username: gustavo Post Number: 23 Registered: 6-2008 |
Welcome James! Never say never. From theoretical point of view, NaCl is undetectable with NIR. It can be considered as waste of time. But as you said, you can model the interaction of NaCl with the matrix. Animal feed is a complex matrix with several salts. NaCl may interact with water and raise a detectable signal, but KCl may also do the same. And CuCl, NaBr, FeClO4, etc... So, how to assign a change in water bands as being an effect of a given amount of NaCl and not from other salt (or even other organic molecule)? That's the critical point of your approach. I will not say that it is impossible, but a believe that NIR is not the better approach to solve your problem. Keep asking! Gustavo. | |||
| James Mitchell (bones29) New member Username: bones29 Post Number: 1 Registered: 4-2011 |
Hi, I would like to request some help as Im new to the world of NIR and chemometrics and currently trying to develop a FT-NIR PLS calibration to detect NaCl in animal feed in a small concentration range. Its more important to tell if the salt is outside this range than what the exact figure. I have a calibration R^2: 0.52 and SEC: 0.04 but have ran into a problem. The calibration does not hold true when a sample is outside the range. For a sample with 0.15% NaCl the model predects it contains 0.3% where as a concentration of 0.3% is reported between 0.28 and 0.32. I have tried using spiked samples to extend the range of salt but still the model does not hold true. I have been told that it is not possible to detect NaCl via NIR while others have said it can be detected by the interaction with moisture peaks. Has anyone any experience with salts or any advice? Due to being new to this field can anyone recommend any futher reading on developing calibrations. I do have a lot of questions however I dont want to take up anyone's time due to the basic nature of the questions Many thanks James |
summary salt prediction