Application of Design of Experiments and Multilayer Perceptron Neural Network in Optimization of the Spray-Drying Process
Authorized Users Only
Article (Published version)
MetadataShow full item record
The aim of this study was to investigate the effect of changing spray-drying parameters on the production of a naratriptan/maltodextrin/lactose composite, as well as to evaluate the application of design of experiments and multilayer perceptron (MLP) neural network in studying the spray-drying process. The system was spray dried as an aqueous solution using a Buchi 290-Mini Spray Dryer (Buchi Laboratoriums-Technik AG, Switzerland). A 2(4-1) factorial design study was undertaken to select the spray-drying processing variables that significantly affect the production yield, outlet air temperature, and residual moisture content. The process parameters studied were inlet air temperature, pump speed, aspirator setting (drying air flow rate), and feed concentration. After performing this screening study, three process parameters, pump speed, inlet air temperature, and feed concentration, were selected for further analysis, applying a central composite design and artificial neural network. Ov...erall, the parameter that had the greatest influence on each investigated response was pump speed. It significantly influenced yield, moisture content, and particle size. Interaction between inlet air temperature and feed concentration was the only statistically significant interaction that influenced the moisture content. Particle size was mostly influenced by feed concentration as well as by a pump speed. A multilayer perceptron was the type of artificial neural network applied. The selected MLP structure had three layers: the first layer had three input units, the second layer had three hidden units, and the third layer had four output units. The selected MLP was trained through 10,000 epochs. Design of experiments using response surface methodology allows insight into interactions between variables, and an artificial neural network provides better prediction potential, enabling simultaneously determination of several outputs. Design of experiments and artificial neural network proved to be useful tool for optimization of the spray-drying process, where a design space for achieving the best process yields and optimum particle characteristics was established. There is a possibility that this conclusion can be drawn for other maltodextrin/lactose systems with other drug substances or for other similar spray-dried systems (the feedstock type is a carbohydrate-based aqueous solution). In future work, this will be tested with other materials.
Keywords:Artificial neural network / Design of experiments / Lactose / Maltodextrin / Spray drying
Source:Drying Technology, 2011, 29, 14, 1638-1647
- Taylor & Francis Inc, Philadelphia