TAF PUBLISHING

In this study, a feed-forward multilayer back propagation with Levenberg- Marquardt training algorithm Artificial Neural Network (ANN) was developed to predict the particle size from the extraction of ginger rhizome using super-critical carbon dioxide in Rapid Expansion Super-critical Solution (RESS). Solid oil particle formation analysis is carried out using Scanning Electron Microscopy (SEM) and Image processing and analysis software, ImageJ. The ANN model accounts for extraction temperature (40, 45, 50, 55, 60, 65, and 70¡ÆC) and pressure (3000, 4000, 5000, 6000, and 7000psi) the size of the particle. A two-layer ANN with two input variables (extraction temperature and pressure) and one output (particle size) with 35 experimental data was used for the modeling purpose. Different networks were trained and tested by changing the number of neurons in the hidden layer. Using validation data set, the network having the highest (nearest to the value of one) regression coefficient (R) of 0.99721 and the lowest (nearest to the value of zero) Mean Square Error (MSE) of 0.00031 was selected as an optimum ANN model. The suitable ANN model is found to be one hidden layer with seven hidden neurons.