Browsing by Author "Lassila, Mikko Johannes"

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Sähkökemiallinen ja biologinen vedyntuotanto
    (2010) Lassila, Mikko Johannes; Kanninen, Petri; Kemian ja materiaalitieteiden tiedekunta; Linnekoski, Juha
  • No Thumbnail Available
    Item
    Towards real-time monitoring of moisture in fluid bed granulation; implementing near-infraned spectroscopy
    (2011) Lassila, Mikko Johannes; Pohjanjoki, Pekka; Kemian laitos; Kemian tekniikan korkeakoulu; School of Chemical Engineering; Kontturi, Kyösti
    Orion Pharma has acquired a new type of fluidized bed granulator whose structure does not allow interrupting the drying operation to measure the moisture content of a wet mass. Traditionally off-line methods such as loss-on-drying (LOD) based on infrared heating or Karl Fischer titration has been utilized. In this thesis near-infrared spectroscopy (NIRS) was investigated as an alternative real time moisture analyzer. As an extremely promising technique it could replace the conventional methods also in processes where pausing the drying is not a problem. However, there are various challenges related to the implementation of NIRS. One of them is that data provided is rather complex and chemo metric techniques such as data preprocessing and multivariate calibration are required to accomplish a quantitative or qualitative analysis. The aim of this work was to find an ideal way to measure with a NIR spectrometer and to examine if NIRS is a reliable technique for following the evolution of the moisture content of a pharmaceutical wet mass and defining the end point of drying. During the work calibration models based on the combination (1 930 nm) and overtone (1 450 nm) peaks of water were built for a placebo product. As a reference method LOD was utilized. The models were tested with two other batches. The soundness was evaluated by root mean square error of calibration (RMSEC) and prediction (RMSEP) and coefficient of determination (R2). In this study a tentative at-line model was built for one product. According to the results the most accurate chemo metric calibration method was built with partial least squares (PLS) regression using a combination of standard normal variate (SNV) and de-trend as pretreatment of the spectral data. The most precise predictions were achieved by limiting the spectral region to 1 800 - 2 000 nm. When testing the model RMSEP 0,18 w/w-% and R2 0,993 were obtained using eight latent variables. The moisture content of the test samples was between 2,4 w/w-% and 9,2 w/w-%. Principal component regression (PCR) also showed promising results with RMSEP 0,34 w/w-% and R2 0,975.