| Author | Message | ||
     Gabi Levin |
Hi Bruce, Good to hear from you. The 1940 or so band of water would be nice if he does not have any starch, look at many pharma formulations, lots of starch, not all of them of course, but very common, other cellulose types also absorb in same region but starch is the most prominent there. Still, worth a try, if he can acheive the reference method desired precision. Gabi | ||
| Bruce H. Campbell (Campclan) |
Chen, There is another approach you may consider. The water band near 1900 nm is usually well separated from other functional groups. (The only ones commonly encountered are PH, CN and nearby is CO. All of these latter ones are of much less intensity.) Thus, why not do a direct absorbance measurement? The precision, if it comes close to that for UV/Vis measurements, would be about 1 to 3 percent, relative. At the 1.5 percent water level, this would give you an absolute precision of about 0.03 percent. The spectrophotometer could monitor that wavelength and one other to serve as a background. Thus there would be two wavelengths used, leading to a very inexpensive spectrophotometer. There is software that could be used to measure the absorbance of the water band on a continuous basis. Alternatively, a rapid scanning instrument could be used together with derivation of the signal to less then effect of particle scattering. You could test this by looking at spectra you already have (I assume you do) for samples that have been analyzed by your refence method. This approach doesn't obviate reference method errors but those errors will be present anyway. This approach should be relatively easy to test and is simpler than the "usual" NIR chemometric calilbration. I would think less than a day of work would point out the feasibility of doing the water determination this way. Bruce | ||
| Lois Weyer |
Hi Gabi, I've found that SNV treatment brings samples of different densities or probe coverage together very well, especially in the case of something like this with such a large water presence. I agree that the area of sampling would be less off line, but for calibration it would be OK because one would be titrating the same sample as the NIR is seeing. As for samples changing upon removal - I guess this is partly dependent upon the temperature and how hygroscopic the samples are. When I experimented with drying modified silica, the water pickup was very dramatic. And, how would you know if you are losing moisture or not when removing a sample unless you actually have a real, offline calibration to compare with? | ||
| hlmark |
It's interesting that Kathryn should bring up the idea of drying to a specification based on spectral characteristics. This is not a new idea; Plugge and van der Vlies proposed it in 1993 and called it the Conformity Index (Plugge, W. and Van der Vlies, C.; "Near-infrared spectroscopy as an alternative to assess compliance of ampicillin trihydrate with compendial specifications"; J. Pharm. Biomed. Anal.; 11(6), p.435-442 (1993), and apparently it was approved for use as an alternative method in the pharm industry. There's also another paper about Conformity Index: Ritchie, G., Mark, H., and Ciurczak, E.; "Evaluation of the Conformity Index and the Mahalanobis Distance as a tool for process analysis: A technical note"; AAPS PharmSciTech 4(2), article 24 (2003) (http://www.pharmscitech.org). I've also written some software to implement this concept; since it's commercial we can't talk about it on this forum. But if Chen wants to contact me off the forum I'd be glad to discuss it with him. Howard \o/ /_\ | ||
| Kathryn Lee |
What excellent advice everyone has given! I have a perhaps naive perspective. It seems that the fluctuations in the spectra over short times would give some limit on the accuracy and provide a range for acceptibility. Could he just dry to a certain spectral range without knowing the exact moisture value? I would imagine that the sample changes moisture content a bit while cooling or transfer or whatever step is next, so that knowing the exact moisture is not important for determining when to stop the dryer, but knowing what spectral response will give the desired level of moisture would be sufficient to perform dryer control. Is the spec for the material in the dryer or when it is out of the dryer and cool? The required moisture level spec was probably determined on a cool sample that had been exposed to the environment for a while. So if you correlate the spectra taken on the dryer with samples taken and measured per the specification test, then you will be able to use NIR spectral response to determine when to stop or adjust the drying process. Thus you are correlating a spectral response to an end moisture value, not the moisture value of the material when the spectrum was taken. This may be a bit more risky depending on the changes in the environment over time, but everyone is already suggesting very careful calibration. The effect of the environment and the next handling steps would have to be accounted for anyway in the spec of the material. 1.5 may be good for the winter when it doesn't pick up much moisture, but 1.2 may be better in the summer when it can pick up more moisture. Does the material being dried have to be a certain moisture level for the next step, or does the moisture level just have to be low and known? For example, if it is added in and a certain amount is needed, then the weight added can be adjusted for moisture level. It is hard to believe that 1.50 +/- 0.05 is the spec since everyone has said that that degree of accuracy is hard to achieve. Perhaps the specification determination needs to be examined to make sure it is right. Are there also specs for the particle size, or is the moisture level being used as a measure of quality for other physical chemical aspects as well? For example, was a moisture level of 1.5 determined to be the spec because it correlates to a specific particle size or flow property or microbial level? You might want to check to see what the specs really should be, and how they were determined. | ||
| Gabi Levin |
Hi Lois, Well, good points about other possible solvents, if that is the case, it needs to be taken into consideration - although, working in FBD in pharma for some years, it has been my experience that well before we get down to 8% moisture these have long been gone for practical purposes. In some cases we could definitely see the negative influence this has on the calibration curve at the higher end, 12-14% and higher. About KF - in at least two instances in FBD application with major pharma - the KF gave us freaky results as well as did the LOD - 10 to 20 minutes pass and the KF value is higher than it was before. It is a question of weighing accuracy, KF procedure accuracy, sample weight, etc. About doing any off line FBD - it is very risky for the simple reason that it is not at all like the fluidized bed - when it is off line the density of the powder is different, and it is not something I would recommend. Besides - when the powder is fluidized if the measurement lasts 10 seconds you average a larger volume of sample than you would off line. Temperature - when you collect on line - you have all these effects built in - in particular if you run over several batches (15 or more)to include various variabilities. About the loss of moisture in a nylon bag - once you are down to 8% it is really insignificant for the short duration and handling. The room is usually not higher than 70F, so it is not significant issue. Thermal degradation products - that's another story -if this is a true situation than they have many other probelms with their product and they would probably not be able to process it in a FBD anyway. Many thanks to all who contribute, it is an interesting issue. Moisture, simple, isn't it. Gabi Levin | ||
| David Russell (Russell) |
If your granules are small (<2mm) and uniform, NIR will meet your need. As mentioned previously, sample presentation and reference method accuracy are key success factors | ||
| David Russell (Russell) |
If your granules are small (<2mm) and uniform, NIR will meet your need. As mentioned previously, sample presentation and reference method accuracy are key success factors | ||
| Lois Weyer |
I'd like to add some comments/questions to this discussion. First, in my experience, LOD does not always give an accurate moisture content, and KF or perhaps GC of extracts need to be the ultimate referee. When I was measuring moisture in things like methyl or hydroxypropyl cellulose, there was some residual methanol or isopropanol that would have been measured along with the water. Also, there are sometimes degradation products that are created during any of the heating methods and these are not moisture. So, there are more sources of variation being added to the "moisture" measurement. Second, where is temperature considered? Presumably, the dryer is at a higher temperature, and this will affect the shape and position of the water peak. So, any offline NIR calibration would have to be done either at the dryer temperature or at several different temperatures. Third, I haven't seen any way of removing sample from the dryer quickly enough or in a container sealed enough that it doesn't lose some moisture. I think the only way to know how much moisture is actually in the substance in the dryer is to calibrate offline with NIR spectra taken at the dryer temperature or a variety of temperatures, in a sealed container and then subsequently analyze by KF. After that, you can follow up with online sampling to look for trends or correlations. For one such calibration, we actually worked in a programmable constant-humidity room so that our samples (paper) didn't change. | ||
| Gabi Levin |
Hi David, Your suggestion as to the use of an oven drying is a very good one, and from my experience in other moisture applications is the way to go, and even there it has proven useful to have duplicates and triplicates to get good reference. Only issue with it - it is slow like hell. With fluid bed you can't really "look" at the same sample you collect for the reference, your best situation is when the bed is fluidized and if you have reasonable uniformity of the bed, to collect spectra from as close as possible to the thieving point. The common practice in the industry is to collect the sample into a plastic bag and then put it into the LOD (whatever heating method it uses) and run it for ~10 minutes. Today these devices have a small microprocessor that calculates based on the loss of weight rate the anticipated weight at complete drying condition and gives you a value. Performing calibration based on off line samples will not work in their case, and will not be accepatble. I don't think that monitoring the NIR by quick LOD after calibration and validation will be acceptable because what will happen is that they will start finding large deviations between the reference and the NIR - total tragic situation in the pharma industry. If you use a certain reference method during calibrations, that's what you will use to monitor. KF will not help that much either, because if they have to wait for the lab to run KF, they might as well wait for the long oven method that will be more accurate and I would recommend to use more than 1 gram sample to reduce the error due to weighing accuracy. Hope this helps a little. Gabi Levin | ||
| David W. Hopkins (Dhopkins) |
Dear Chen, Moisture is the easiest and the hardest consitituent to measure in with NIR measurements. It is easy because the moisture band is so large and relatively isolated from the other constituents. It is difficult because of sampling problems (which have been alluded to) and because of reference method considerations, which have also been alluded to. To answer your question, I think you should be able to achieve 0.05% SEP, if you pay close attention to details. That is the level we were able to achieve with moisture in flour, an application I can talk about from the old filter unit days. I would suggest that you follow a strict procedure that the sample that the NIR sees is the same sample that is analyzed by your reference method, and that care must be taken to minimize the time and handling between the NIR scan and the reference method. A closed cup for NIR measurements preserves the moisture content up to and after the scan. Then use that sample for the reference test. Then you need to use a method that will give you at most 0.03% SD of repeats. That will eliminate the IR balance and you will have to work to get that kind of reproducibility from KF. An oven method such as 1 gram dried at 130 C for 1 hour or 105 C for 3 hours was standard for wheat flour, and should work for most pharmaceuticals too, I would expect. With careful attention to detail, I think you should be able to approach the accuracy you desire. Then you can relax the requirement for the standard method, and use the IR balance to monitor the performance of the calibrations, with higher limits on the agreement as you determine for the moisture balance. But at least you will have a method you know has an accuracy of at least 0.05% (or whatever you find you achieve) when a highly reproducible reference method is used to check the calibration. I hope this helps. Sincerely, Dave Hopkins | ||
| hlmark |
Chen - Gabi's earlier message (the one at 11:49) raises the suspicion that even with all the agitation in the fluid-bed dryer, the moisture content may not be homogeneous, that there are regions, or even "clumps" of material with different moisture content. That would explain his observation of the measured moisture increasing at a later time: the measurement is local and a region of higher-moisture material moved into the sensor's range. If the material really is non-homogeneous this way, then no amount of improvement of the measurement is going to change your accuracy, because your measurements are reflecting real differences in the material. If that's the case, then what you need to do is improve your FBD behavior, before you worry about a better measurement. Howard \o/ /_\ | ||
| Gabi Levin |
Hi Chen, Thanks, this confirms the suspicion I had, but did not want to jump to conclusion. I have tons of experience in FBD. First drying to the bones and then wetting is something that is definitely an undesired process, because if you have done any granulation inside the FBD, now you will destroy the value in it, because if you spray inside the FBD to wet it, you will get renewed agglomeration and the granulate will change. The problem with trying to achieve +/-0.1% moisture is not in the NIR - it is in the IR balance which does not give you this level of quality reference measurements. The IR balance is a method to heat the powder inside the balance and do the LOD method. Thinking about Karl Fisher? Can be problematic as well, depending on the sample size, etc. So where is the advantage of NIR? If you work on line, like we do in so many places, you can get frequent readings, much more frequent than with any other method, and you base your end point on a significant number of results, which represent, because of the motion in the bed, when it is fluidized, measurements from different "places" within the bed. This provides a much more reproducible end point. However, with the reference method not being better than 0.25 to 0.3% (or even worse according to data from a large pharma company) achieving +/-0.1 requires some serious work on the reference method. If you write to me, I will be able to help more. gabi Levin [email protected] | ||
| Chen |
Tony, Gabi, Yes this is a drying operation in a Fluid Bed Dryer where we want to stop the drying to a predetermine value of moisture, i.e 1.5 +/- 0.1 instead of drying completely and then after adding water to achieve the target. The reference method is infrared balance. Thanks, Chen | ||
| Tony Davies (Td) |
Dear Chen, Are we correct to think that you have a drying operation? What are you using for your reference method? Do you want to monitor the drying operation so that you can stop it at a predetermined given level? Are you drying much below the material's stable water content in your working conditions? Or do you have many materials to be dried to different water contents? Best wishes, Tony | ||
| Gabi Levin |
Hi Chen, To your question - can we lower the SEP to about 0.04 if I understand correctly - Well, first and foremost we need to know the moisture SEP of the reference method, remember - rule of thumb is that you can't be better than the reference method. Not knowing what is the application and what is the reference method, it is difficult to answer your question, but let us provide one small advice: If you can afford a shorter range for your calibration, say 0 to 4%, you will obtain a smaller SEP, however, remember that now the reference method shall be such that the SEP of the reference will be less than 0.15% if you want the NIR to give you a SEP of ~0.17%, and it must be below 0.04 if you want the NIR to give you ~0.04 for SEP. If for example you have a 4% range with reference method that yields only 0.25 or 0.3 for SEP (such as most common LOD methods) you will find it difficult to have a good calibration for narrow range of moisture. I have experience in many fluid bed operations, and one of the very common phenomenon that occur there is (and I have plots showing that) that the LOD can act strangely - i.e, you pull a sample at time X and then at time X+10 minutes, but the LOD shows "increase" in moisture, not logical, is it. But this happens, and it does affect the quality of calibration. When something like that happens and you have NIR readings every 15 seconds, and you plot the LOD (every 10 minutes or more)alongside - the LOD pops up above the curve of the NIR. If you want some more specific information you can write to me at the following e-mail [email protected] and I will forward to you more information. I hope this helps you to better realize what you can achieve. Thanks, Gabi Levin | ||
| Chen |
Dear All, I would like to know from your practical experience what could reasonably be the lowest SEP achievable for moisture detection in powder by NIR, in an online process ? I' ve seen some data where mositure is measured from 15% (in mass) down to 0 with SEP = 0.4 but I wonder if it can be lowered by a factor of ten. Thanks for your input ... |