Biodiesel production and prediction of engine performance using SIMULINK model of trained neural network

Abstract

Due to depletion and higher prices of crude oil, biodiesel is gaining more importance day by day. Biodiesel is renewable and eco-friendly and its emission profile is much lower than fossil fuels. A large number of researches have been done on identification of new feedstocks and cheaper technologies for biodiesel production. Among many feedstocks used oils have been reported as a cheaper feedstock for biodiesel production. Transesterification of veg fried oil and non-veg fried oil was studied in a batch type reactor with NaOH and methanol. The reactions were optimized, veg fried oil and non-veg fried oil gave an maximum ester yield of 91% and 87% respectively for 0.6 wt.% of NaOH with a molar ratio of 6:1 for 3 h at 65 C. Fatty acid profile of these two methyl esters were similar and their parent oil was found to be palm olein. Fuel properties were some what closer to each other. These esters (B100) and their blends with diesel (B25, B50, B75) when utilized in DI diesel engine showed better emission profile. B75 was found to be the best. An artificial neural network (ANN) was created with brake power and biofuel blend as input and brake specific fuel consumption, brake thermal efficiency, NOX, HC, CO and smoke density as output. Back propagation algorithm was used and the data obtained from engine test was utilized for training the network. The SIMULINK model of the trained neural network was generated to predict the fuel emissions and performance. The trained neural network with a correlation coefficient of 0.9989 and 0.999 gave better predictions for B15, B30, B60 and B90, the results were found to be acceptable.