An artificial neural network approach to predict the performance and exhaust emissions of a gasoline engine using ethanol-gasoline blended fuels

摘要

The sole intention of this research work is to examine the performance and pollutant emissions of a four-stroke SI engine, allowed to operate under specific conditions with varying ethanol-gasoline blends ranging from pure gasoline to pure ethanol. The blends so chosen are E0, E10, E20, E40, E60, E80 and E100. An artificial neural network (ANN) is employed as the measuring tool to analyze and validate the experimental results with the ANN-predicted results. The power and torque output of the engine used for the experimental work showed a substantial rise using all blends. Brake specific fuel consumption showed a sharp decline when ethanol blends were used in the engine. The experimental investigation also showed an escalation in brake thermal efficiency and volumetric efficiency. Careful examination using an exhaust gas analyzer found that the concentration of CO and HC emissions were on the low side compared with emissions when the engine was solely operated on gasoline. Lower emissions resulted because ethanol contains a high percentage of oxygen. CO2 and NOx emissions, considered to be quite lethal, showed a sizeable increase when ethanol was introduced with gasoline. ANN was developed to envisage a correlation between all performance parameters and emission components using different gasoline-ethanol blends and with varying engine loads from no load, 25, 50, 75% and full load as input data, keeping the speed of the engine at a constant value of 2500 rpm. Almost 70% of the total experimental data was selected at random and was used for training purpose, while another 15% was used for validation. In order to improve the results for network generalization the last 15% of the data was utilized. The ANN model so developed generated the best correlation coefficient (R) ranging from 0.999923 to 0.999977 for all performance parameters and exhaust emissions. Mean relative error values were in the domain of 0.12-5.56%, while root mean square errors were very low. R values did not increase when neurons in the hidden layer were more than 20 such as 21, 22, 23 and 24. Therefore, a network with one hidden layer and 20 neurons was selected as the most favorable ANN. Research study and subsequent findings helped us to reach to the conclusion that the ANN approach could be considered as the best feasible way to predict SI engine performance and engine emissions in a very accurate manner.