Published Paper Details:

- In this work, a cascade-forward neural network model has been developed to represent a reactive distillation process used for the production of biodiesel from an esterification reaction between palmitic acid and methanol. In order to obtain data for the network, the parametric utility of an Aspen HYSYS prototype plant of the process developed using Distillation Column Sub- Flow sheet and Wilson model as the fluid package was utilized. The neural network model developed had six input parameters (palmitic acid feed temperature, palmitic acid feed pressure, methanol feed temperature, methanol feed pressure, reboiler heat duty and reflux ratio), and the output parameter was the molefractionofthebiodieselobtainedfromthebottomsectionofthereactive distillation column. For the training of the neural network model, six different random number generators (Messene twister, multiplicative congruential generator, multiplicative lagged Fibonacci generator, combined multiple recursive generator, shift-register generator summed with linear congruential generator, and modified subtract with borrow generator) were tried by varying their seed numbers from 0 to 70, and the one with best performance, together with the corresponding seed number, was selected for the development of the cascade-forward neural network. The results obtained from the training and simulation carried out for the developed model showed the good representation of the process by the developed model because the estimated sum of absolute error, mean of absolute error, sum of squared error and mean of squared error of the model, which were the performance criteria used, were found to be favourable and had values of 1.16E-02, 1.93E-05, 5.46E-07, and 9.10E-10, respectively. Also, the performance of the developed model in predicting the mole fractions of the produced biodiesel was found to be very good as the sum of absolute error, the mean of absolute error, the sum of squared error and the mean of squared error, in this case, were estimated to be 8.39E-03, 1.40E-05, 2.17E-07, and 3.62E- 10, respectively. In conclusion, cascade-forward neural network has been demonstrated to be very good in modelling this complex reactive distillation process for the production of biodiesel.