Darronqui, Elaine. Pharmaceutical process technology : from new materials to new technologies. 2010, PhD Thesis, University of Basel, Faculty of Science.
Official URL: http://edoc.unibas.ch/diss/DissB_9213
On the second chapter an investigation of the influence of drug load, liquid addition rate and batch size on the power consumption profile recorded during a high shear mixer granulation is presented. The materials used were microcrystalline cellulose and paracetamol. HPMC was added as a dry binder into the mixtures, and distilled water as the granulation liquid. Three ratios of drug:excipients were used. Percolation theory could be applied to describe the growth behaviour and granules properties. The liquid saturation was calculated and used to compare different moments of the granulation and also results among the formulations. The powder physicochemical properties strongly influenced the amount of liquid penetration and the free surface liquid which is necessary for the coalescence of primary particles and/or agglomerates, and consequently affected the power consumption profile. Thus the drug load was the most influential factor. The power profile was practically independent of the batch size. The liquid requirement was linearly dependent of the mass loaded. The liquid addition rate made a slight impact on the total amount of water used and in the granulate growth kinetics. The “in process” measurement of power consumption showed to be a reliable analytical tool for monitoring the moisture content and particles agglomeration growth.
In the third chapter, in a collaboration project with University of Belgrade, the use of an Artificial Neural Network to predict characteristics of a wet granulation was studied. The initial aim was directly predict the profile of power consumed during the granulation process; using as inputs properties of the starting material as well as process variables. Data preparation is a fundamental and very important step once that the software will learn by observation of the data presented to it. As inputs formulation and process parameters were chosen, and the output was the power value (watts). As the experimental values were a time series/sequence, the order of the data is important to avoid that the software when taking a value doesn’t break the sequence, leading to a misunderstanding of the data. Different snippets (fragments of networks) were tested and the final topology that showed reduced error and better predictions was a so called Gamma-Recurrent Hybrid network. The results showed that the network was able to satisfactory predict the power consumption curve of formulations containing higher amounts of excipients. For the formulation with higher drug load (90%-wt. of paracetamol) the predictions were unsatisfactory. The experimental granulation process executed for that formulation generated more irregular power consumption curves and that wide variability among the inputted samples could result in a difficult learning from the system, and can be the reason for the lack of network precision in predicting the behaviour of this formulation. The absolute error for the prediction of the power consumption value was 76,35; relatively small compared with other systems tested. Submitting different inputs it was also tested the ability of an adaptive system to predict the relevant S2, S3, S4 and S5 points of a typical Leuenberger PC profile, and for that basic static feedforward backpropagation networks resulted in good predictions. Predicting the future output of very complex systems is a difficult task. Adaptive systems however have shown themselves, trained on the right data, quite capable of producing good predictions.
|Committee Members:||Betz, Gabriele and Hoogevest, Peter van|
|Faculties and Departments:||05 Faculty of Science > Departement Pharmazeutische Wissenschaften > Ehemalige Einheiten Pharmazie > Industrial Pharmacy Lab (Betz)|
|Bibsysno:||Link to catalogue|
|Number of Pages:||151 S.|
|Last Modified:||30 Jun 2016 10:41|
|Deposited On:||07 Jan 2011 09:53|
Repository Staff Only: item control page