| Author | Message | ||
     venkatarman (venkynir) Senior Member Username: venkynir Post Number: 49 Registered: 3-2004 |
Dear Suresh ; Kindly read the paper , I have mentioned . I want to know the parameter used by you for end mark in blending. See, There are a lot of pre-procssing methods are there .Savgolay one of the step in the method of pre-processing that is smoothing the signal. What about derivative of the spectra ? Have you taken care of MSC and OSC ?. Please think and putforward the object and parameter clearly. So than we can suggest further. | ||
| SKV (suresh_kumar) New member Username: suresh_kumar Post Number: 3 Registered: 8-2006 |
Dear Venkataraman, Thank you verymuch for your response. In you message you mention that, for proper blending few parameters are required, Can you tell what are(parameters) you looking for. Regarding spectral pretreatment iam using SGoley 2-derivative (7 points smoothing). Hi Gabi Levin, Thanks for your response. I have restriction on using Zeiss instrument, now i can not change to brimrose can you suggest some method/solution for the existing system of what i have. Thanks in advance. | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 47 Registered: 3-2004 |
Dear Suresh, First there are good amount of publicaitons avilable in blending .Please read.Please avoid SIMCA. SD not lone give full picture . To ensure proper blending few paramters are required.You have not mentioned it in your post. First ensure pre-processing that makes the data be fit. Second use Supervised method at first to ensure end point. If you have further doubt please post your data . Regrading levin ; This is a very long topic of course ,but alot of work done different groups and publication avilable . I do agree this could done through post. He has clerly told he is usin Zesis diode array , may Gelvin may have AOTF in mind. yes, wide range and good samples gives better results in NIRS. Please Read the paper "International Journal of Pharmaceutics 340 (2007) 97�103" you will get some idea . | ||
| Gavriel Levin (levin) Senior Member Username: levin Post Number: 52 Registered: 1-2006 |
Suresh, This is a very long topic, and I will not be able to help you from this post. I have extensive experience - and it is much more complicated than it seems. It appears that you are using some diode array that goes to 1680 and you are probably missing very important C-H peaks above 1680, (actually I don't see any peaks that are typical to C-H bonds in the typical region among the two you mention) which is having very serious influence on what you do. Please write to me [email protected] and I will see how can we help you. Gabi Levin Brimrose | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 46 Registered: 3-2004 |
Dear SKV , First there are good amount of publicaitons avilable in blending .Please read.Please avoid SIMCA. SD lone give full picture .Your not mentioned ensure of blend point. First ensure pre-processing that makes the data be fit. Second use Supervised method at first to ensure end point. If you have futher doubt please post your data . | ||
| SKV (suresh_kumar) New member Username: suresh_kumar Post Number: 2 Registered: 8-2006 |
Hi Everybody, Can anyone guide me on, how to identify the endpoint for pharmaceutical Blending operation using NIR? I am currently trying the following method. I have scanned the sample complete region ie., 960 nm � 1680 nm . Found 1486 and 1120 nm are of my product interest, which I have confirmed with pure sample. I have applied moving block 8 spectra method and found standard deviation over time. My blending time operation is 40 min but by standard deviation method ( X-axis :Time and Y-axis is Response at 1486nm) it is showing endpoint too early, This i am saying with respect to standard deviation(less) is constant for about ten readings (say 0.1 over ten rotations). With this i couldnot able to find endpoint clearly. I am getting low SD at 10 min and even at 20-30 min etc., Now I would like to approach this problem by following First i would like to make SIMCA model by using uniform blending samples (one class) then using SD or mean plot method to identify endpoint. Is this approach is good? Advance big thanks for those, who share about any methods to identify endpoint determination. I would like to use Unscrambler to build SIMCA Model and Zeiss NIR for online measurements. |