Comparison of response surface methodology and feedforward neural network modeling for polycaprolactone synthesis using enzymatic polymerization. (March 2019)

• Record Type: Journal Article
• Authors: Pakalapati, Harshini; Arumugasamy, Senthil Kumar; Khalid, Mohammad
• Is Part Of: Biocatalysis and agricultural biotechnology. Number 18(2019)
• Publication Date: 2019-03
• Subjects: Polycaprolactone (PCL) -- Bio-polymerization -- Ring opening polymerization -- Response surface methodology -- Artificial neural network modeling; Agricultural biotechnology -- Periodicals; Enzymes -- Biotechnology -- Periodicals; 660.6
• DOI: 10.1016/j.bcab.2019.101046
• Languages: English
• ISSNs: 1878-8181

Abstract

Abstract: This study highlights the optimisation of process parameters for the synthesis of a biodegradable polymer – polycaprolactone using response surface methodology and artificial neural networks. Temperature, time, mixing speed and solvent volume are the parameters considered for optimisation and polymer yield was chosen as response. The results obtained from RSM displays a good agreement with 3.4% percent deviation between predicted values and actual values. Further, feedforward neural network (FFNN) modeling is developed with five different training algorithms. Out of all, Levenberg-Marquardt training algorithm proved to be best with lowest MSE, MAE, MAPE values of 0.10, 0.18 and 0.02 respectively. Both the techniques have been successful in predicting the biopolymer yield. Coefficient of determination (R 2 ) and Absolute average deviation (AAD) value for RSM are obtained better proving RSM superior to ANN in this study.