| Author | Message | ||
     maina kerbrat (mainak29) Junior Member Username: mainak29 Post Number: 7 Registered: 2-2011 |
Dear Gabi, Thank you, this is the method for the dosage of moisture in tuna. It is possible to analyse the moisture in the tuna in the lab. But, We have the project to analyse by NIR protein and fatty acids, and our lab couldn't analyse that. It is for that, that we analyse tuna with another lab. But for the moment, I'm waiting for the beautiful Spectralon, wait and see!!! See you Gaby and thanks so much for your help. Ma�na | ||
| Gabi Levin (gabiruth) Senior Member Username: gabiruth Post Number: 54 Registered: 5-2009 |
Dear Maina, I do not know the substance in which you measure the moisture - but unless it is a volatile material, that is - if the only thing that will evaporate upon heating to 80 Celsius is water you do not need an expensive laboratory for moisture. You need an oven where you can do the following: 1. Take sample cups made from aluminum - if you do not have ready made small cups, you can use aluminum foil - and make small cups out of it. Weigh the cups accurately to the 1 mg. I believe you have such precision balance. Place the cups in the oven for overnight at 100C. Weigh again, if the weight did not change - your are OK. make sure you number the cups and note each cup it's exact weight. 2. Place a sample (5 to 10 grams is more than needed)in the cup, weigh it accurately and place it in the oven at 80C for overnight. Weigh in the morning after it is cooled completely. 3. Place again in the oven for 3 hours and weigh again. If there is still a weight change, place in the oven again for 2 hours and check the weight again. If weight has not changed, you can now calculate the total weight loss. You can now divide the weight loss by the total initial weight to know the % moisture. If you have volatile compounds in your product such as alcohol, this method is not good. But if it is regular agricultural product it is usually no problem to use this method. This way you can actually increase your data base at a very low expense. I hope this helps you Gabi Levin www.brimrose.com | ||
| maina kerbrat (mainak29) Junior Member Username: mainak29 Post Number: 6 Registered: 2-2011 |
Sorry Gaby, My database is a little bit small! But it is my database, and I 'll try to find the best model. With your technic, I think it is possible. In mauritius, the big problem is to find a reliable laboratory for reference method analysis of moisture. And it's very expensive. It is for that, that my database is poor. How many samples sould I take with the new Spectralon? The same number? 67? I think that is enough because of the problem of reference analysis... sorry... | ||
| Gabi Levin (gabiruth) Senior Member Username: gabiruth Post Number: 53 Registered: 5-2009 |
Maina, I must admit that I am surprised that you only have 67 samples in your existing calibration - but if this is the case, then it will definitely be useful to keep all old samples and add the new ones to the model. This way it will be more robust. I wish you success, Gabi Levin Brimrose corp. of America www.brimrose.com | ||
| maina kerbrat (mainak29) New member Username: mainak29 Post Number: 5 Registered: 2-2011 |
Hi Gaby, your response is very clear! Now, I have 67 samples in my database (with old reference). I think it is good to have the same number of new samples? When I'll have my new beautiful model. And after, I'll suppose i have a good prediction. I put my validation set in my calibration set, for having a new calibration model of moisture. And I remove the Old spectra or I keep them into my new calibration set? Thank you! Ma�na | ||
| Fernando Morgado (fmorgado) Intermediate Member Username: fmorgado Post Number: 18 Registered: 12-2005 |
hello : One tip more. Before make any, if you has programed outlier detection for your model will be good check if new samples reading with new reference are outlier. Maybe you has program score , spectra residual or the minus usefull H for your model. If your new samples are realy a big spectral outlier you can say �new reference realy afect my predictions.� Sometimes H outlier samples predict fine, that depend of the Model, parameter, number of factor programed to use , etc. and of course the type of sample Fernando | ||
| Gabi Levin (gabiruth) Senior Member Username: gabiruth Post Number: 52 Registered: 5-2009 |
Maina, There is no need to do any pre-processing to correct the difference between the old and new reference sample. Use the same processing you used to create the original model. If you used first derivative for the original, use the same for the new spectra, get the new ref values from the ref method, HPLC, whatever you use, add to the whole set you already have, run the new regression, if you have a few outliers, remove them, (it is useful to check if any of the outliers happens to be a new sample, or only old samples become outliers, but it has little significance if at all) finalize your model and test it with new set of some 20 samples that is not part of the model and check the quality of the predictions on the separate set. If predictions satisfy your required SEP, you can now add the additional samples that were the validation set to your total model, so you will have a model with about 40 new spectra in the new model. It worked for me in a good number of times when a change occurred in a spectrometer, or reference sample. Gabi | ||
| maina kerbrat (mainak29) New member Username: mainak29 Post Number: 4 Registered: 2-2011 |
gaby, great! If I well understand, I blend the old spectra and the new spectra in my new database of calibration. I try to find a good preprocessing to correct the bias generated by the old reference. Like OSC algorithm with 3 components for exemple? But I 'll try with simple preprocessing. Because, I have another problem, The software for the spectrometer doesn't allow to have the spectrum file in *.dx or other filetype! The software is IndicoPro, do you know it? When I take a spectrum, the spectrum is automatically in Digital number Unit. And after, I clic on the button to transform it in Absorbance unit. I don't know what is it the algorithm to to switch from one to another... I hate IndicoPro software!!! When I will have the new spectralon, I try your advices... Thanks! | ||
| Fernando Morgado (fmorgado) Intermediate Member Username: fmorgado Post Number: 17 Registered: 12-2005 |
Hello : Interest comment from gabi. When the teory go to swim died for not have practice. In the pass I have a similar problem with a customer. We add samples to the database read with the new reference and after make the Model we compare result using old model and new model ( including spectra taked with new reference) and the result was diferent. Maybe using derivate in the Model you can correct the desviation in the Y axis produced for use diferent reference en the spectra. About use stored reference this is valid if the system is stable in the time, for example some FT-NIR can use stored reference for years. I don�t know if the Maina Instrument can make that. I know a normal procedure is read reference sample to sample. About the procedure indicate for gabi, is not bad, you can try. Remember not use lab v/s NIR for check the model created, you will need a external validation set, not outliers, inside the range model and across the range. Good luck | ||
| Gabi Levin (gabiruth) Senior Member Username: gabiruth Post Number: 51 Registered: 5-2009 |
Hi guys, I will throw in my nickel 1. You can not use your old stored reference spectra for doing the predictions - if you could you would only need it once in a life time - you need it frequently so you can correct for energy at each wavelength at time of measuring, not some historical value 2. You can not correct by using simple difference between old predictions and new predictions and use Excel for that - would be nice if we could. 3. One possible way, maybe not any better than others suggested is to clean your spectralon, as recommended by Art - I have no better idea how to do it, or get a new one - and then collect spectra from some twenty or so samples that span most of the range of values, do the reference analysis - and add them to your file and run a new regression using Unscrambler - or whatever you prefer, and create a new model using all old spectra and the new 20 spectra with the new ref data. Then validate the new model with some new 20 samples across the range - check your predictions - if satisfactory, add these 20 also into the model and you are good to go. The chemometric should be able to adjust you loadings to accommodate the difference between the dirty and clean spectralon. gabi | ||
| maina kerbrat (mainak29) New member Username: mainak29 Post Number: 3 Registered: 2-2011 |
Thank you for all your Answers! (and Ok for cleaning my spectralon, I unsderstood ;)) Howard, unfortunately, I haven't the old reference spectrum. Yes, with the time It became yellow. And Now it is more yellow than 6 months ago (the date of my database for my calibration set!). I use Unscrambler Software for modelling and I love So much matlab!!! I understand what do you mean Art, I will try tomorrow morning. Yes, I'm in Mauritius, and it is the night! Thank you sooo much guys! Ma�na | ||
| Fernando Morgado (fmorgado) Intermediate Member Username: fmorgado Post Number: 16 Registered: 12-2005 |
Dear Maina : I dont know your software capacities. But I think you program can not make some standarization of the old taked spectra to the new spectra. When I talk about Bias I say a Bias not in the spectra. The idea is you can use your actual Model using the new reference. The prediction result for the same sample using diferent reference will be diferent. The idea is you comparate the result using excell . Only calculate result with old reference minus result with new reference. Probaly you will find a constant desviation in Moisture ( more high or more low) but constant. For example if you find the result is 1.3 % more high, go to the Model menu and probably you will find a slope and Bias adjustemn option. There ( if the desviation a similar) using the example input in Bias -1.3 ( minus 1.3). With that , using the new reference you will obtain the same result when use the yelow reference. Maybe your program can use stored reference, I don�t know, if you have this option you can measure actual samples using that reference. For complete new models use new reference. fernando | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 395 Registered: 9-2001 |
Marina - Cleaning in the way Art describes is certainly less expensive than getting a new piece of Spectralon, but is as likely to change the value of the reflectance as much as a new piece will, requiring a similar correction. Therefore, I would recommend that before you do that cleaning, or change to a new piece of Spectralon, you collect spectra of both the yellow Spectralon and a set of samples as Fernando described, so you will be able to compare both, before and after making the change (and hopefully be able to use one or the other to make corrections). Which brings to mind another question: did you measure the Spectralon spectrum in the past, at a time before it became yellow? Comparing that with the current spectrum can tell you how much it changed. It's entirely possible, BTW, that since aborbances are often stronger in the visible than in the NIR, there might be no measureable change in the NIR, and therefore (again, hopefully) the yellowing might not affect your results. But the only way to be sure is to compare results with yellow and white Spectralon, in any case. Howard \o/ /_\ | ||
| Art Springsteen (artspring) Senior Member Username: artspring Post Number: 34 Registered: 2-2003 |
Dear Maina, Never,never,never clean Spectralon with alcohol solution! NEVER!!! (get the idea?) Art S | ||
| maina kerbrat (mainak29) New member Username: mainak29 Post Number: 2 Registered: 2-2011 |
Dear All, I explain more : The Spectralon is dirty because users of spectrometer before me don't clean after using. And, they taken a lot of spectra. When I arrived, I see the problem. I try to have good calibration moisture model. But it is impossible when I take a spectrum and predict now this the dirty spectralon. Fernando, I like your idea. So when I will have spectrum difference between the new spectralon and the old spectralon, how I evaluate the bias? I make the difference (with matlab) between the two spectra? And after, for to correct my database? I don't know why I make this after. And, Anaother question? Is It bad to clean the spectralon with alcohol solution? And sorry for my english, I'm french girl  | ||
| Art Springsteen (artspring) Senior Member Username: artspring Post Number: 33 Registered: 2-2003 |
Fernando mentions cleaning the Spectralon. The proper way is with 3M 220 grit Tri-M-Ite paper (silicon carbide paper, wet-dry), under a stream of running water until the Spectralon (or Fluorilon FW99, if you use our material) until the surface it totally hydrophobic- the water immediately beads and runs off. If you cannot get the material to this point, it is so badly contaminated that it should be discarded and a new standard purchased. And he is 100% correct- never use ANY solvents other than water around Spectralon or Fluorilon. It is a recipe for disaster! | ||
| Fernando Morgado (fmorgado) Member Username: fmorgado Post Number: 15 Registered: 12-2005 |
Hello : is a big problem. Art have the reason, but use a complete new spectralon will change the prediction of yours models, since the baseline change. The only posinility you have for continue using your Models is the follow 1.- select maybe 10 samples. 2.- Read the samples using the yellow spectralom and after read again using the new. 3.- Evaluate Bias, the more probably the bias will ba a constant. If you can define some constant Bias, apply this bias to the Model in your software. If you want expand some old database will be other problem, since will be not good combine spectra taked with yellow and new. If the spectralom is a disk is a good idea in the time use a soft sand paper for clean it , don�t use solvents or other. Good luck Fernando Morgado | ||
| Art Springsteen (artspring) Senior Member Username: artspring Post Number: 32 Registered: 2-2003 |
One other thought. You should get a new, uncontaminated reference, run the baseline with it, then run the old Spectralon. The difference between the two will show you a) how contaminated it is and b) with any luck, the likely contaminant. | ||
| Art Springsteen (artspring) Senior Member Username: artspring Post Number: 31 Registered: 2-2003 |
Dear Maina, I'm the developer of Spectralon so may be able to help but I do not understand what you mean by 'usury time', Also, I am not sure why your Spectralon is yellow, unless it has become contaminated over time. In that case, you should replace it, as the spectra you have been collecting will give you an improper baseline. Best regards, Art Springsteen | ||
| maina kerbrat (mainak29) New member Username: mainak29 Post Number: 1 Registered: 2-2011 |
Hello, I have a big problem with my spectralon. It is yellow with the usury time. I want to create a calibration model of NIR based on NIR spectra, but spectra are created with this Yellow spectralon. So the baseline, and the spectral bands of CO2 are visible (verified via a regression with Matlab software, I compared a spectrum taken there 6 months and taken now). I buy another Spectralon (white!!!!). But I don't want to lose data taken with the old spectralon. Is it possible to correct spactra, the bias due to the usury Problem? Just for your information, my spectrometer is a ASD LabSpecPro. Thank you so much! |