Modeling And Scale Up Prediction Of High Shear Wet Granulation Based On Artificial Neural Network Model

Granulation is a kind of common unit operation which is widely used in multiple industrial processes.The raw powders were hold together by some combination force to form the granule product during the granulation process.There are many variables that can affect granulation process,such as formula properties,mixer geometries,operating conditions and so on.Most of these variables have a complex high nonlinear relationship with granule product properties.This makes research on the mechanism of granulation process becomes a very complicated task.Until now,the study of granulation can just give a qualitative description of which mechanism region will a particular granulation process fall in,and what is the possible process phenomenon,and cannot give a qualitative prediction of granulation product properties.Artificial neural networks model has obtained a wide range of applications in many different fields such as pattern recognition,predictive control and so on for its excellent ability of information processing.The characteristic of good at dealing with high nonlinear relationship makes it especially suitable for granulation process modeling.In this thesis,a semi-empirical model which is suitable for industrial application has been established for high shear wet granulation process by artificial neural networks method.This model can make a quantitative prediction of particle size distribution parameters of granule product.The main content of this thesis includes the following aspects:The main variables that affecting the granulation process were determined by literature review and experiment observation.Four key parameter groups such as total relative swept volume(RSVtotal),relative theoretically liquid availability(RTLA),liquid injection factor(LIF)and stokes viscosity number(Stv)were defined to cover these main effect variables and give a standard description for granulation process.This standard description method can be used in different formula and different equipment.The key parameter definition method can decrease the dimensionality of the problem;reduce workload of experiment and modeling process.In addition,the dimensionless key parameters are independent of equipment scale which is suitable for scale up of industrial application.The granulation experimental work was carried out from pilot plant scale to industrial scale.All the formula properties,operating conditions and particle size distribution parameters were recorded.The physical properties of raw materials which is necessary for the calculation of key parameter groups' value were measured.Then the value of key parameter groups of each batch experiments were calculated and used as input to make prediction of granulation product particle size distribution parameters.The establishment of artificial neural networks model.Using the MATLAB software,selects key parameters as inputs to establish the artificial neural networks model of granulation process,make prediction of particle size distribution parameters of granule product.The effect of network topology on prediction performance was investigated by using of a network which has variable hidden nodes.The optimal network topology was selected finally.The performance of different combination of transfer function and training algorithm was compared,and the optimal combination was selected.The importance of different key parameters on particle size distribution of granule product was analyzed by mean impact value(MIV)method.For a bit of poor performance of the prediction of span,two commonly used approaches in artificial neural network establishment were applied to make optimization,and the possible reason were discussed.Finally,the prediction performance of the artificial neural networks model was validated for different formula,different equipment scale and different operating conditions.