Experimental and Modeling Study of Esterification Reaction for Synthesis of Butyl Butyrate: Desirability Function Approach for Optimization and Prediction Comparative Study of RSM and ANN

Butyl butyrate was synthesized by esterification of butyric acid with n-butanol using homogeneous catalyst methanesulfonic acid (MSA). The esterification process was optimized by the application of response surface methodology (RSM) and artificial neural network (ANN). 3 level-4 variables central composite design (CCD) of RSM and MLP 4-9-1 network of ANN was chosen for the experimental design and analysis. The quadratic response model of RSM was optimized using desirability function approach. Effects of independent variables on the yield of butyl butyrate were investigated. Various training algorithm such as IBP, QP, GA, LM, BFGS, and CG was used for training experimental response data for the ANN study. By sensitivity analysis, the relative significance of 36.98 % confirmed that the molar ratio was the main affecting parameter on the yield of butyl butyrate. In prediction comparative study, ANN model was found better than the RSM model with high values of R2 (0.9998) and lower values of RMSE (0.2435), SEP (0.324 %), and AAD (0.0086 %) compared to RSM (R2=0.9862, RMSE=2.3095, SEP=3.076 %, AAD=0.6459 %). The accuracy of the RSM and ANN models were judged by validation test by performing unseen data experiments.