Modeling of cyclic variability using artificial neural network in a diesel engine fueled with alternative fuel blends

In this study, cyclic variability of a diesel engine using neat diesel fuel, five different butanol-diesel blends (%3, 6, 9, 12 and 15) and two different ethanol-diesel blends (%5 and 10) were modeled by using Artificial Neural Network (ANN). Also two different models were developed for engine performances and exhaust emissions. A naturally aspirated, single-cylinder, four-stroke and direct-injection diesel engine was used to evaluate their cyclic variation, engine performance and exhaust emission at different speeds and full load condition. The coefficient of variation of indicated mean pressure was used to evaluate cycle-to-cycle variation. The results have indicated that cyclic variability exhibits a increasing trend according to an increase in the alcohol blending ratio. In addition, specific fuel consumption and effective efficiency increased as carbon monoxide (CO) and nitrogen oxides (NOx) emissions decreased with increasing amount of alcohol (ethanol or n-butanol) in the fuel mixture. The back-propagation learning algorithm with two different variants (Scaled Conjugate Gradient (SCG) and Levenberg–Marquardt (LM) algorithms), single layer, and logistic sigmoid transfer function were used in the developed network. Mean square error, mean absolute percentage error and coefficient of determination values were used for performance of the networks. The all developed ANN models have showed a good agreement with the experimental results.