Biodiesel usage is practically restricted as a blended supplement with fossil diesel. In the current study, the authors have attempted to arrive at the optimal biodiesel blend concentrations for an automotive engine. Here, the artificial neural network and genetic algorithm are coupled with phenomenological combustion modelling for the efficient operation of biodiesel blends. The engine experiments are conducted with diesel and diesel-biodiesel blends namely jatropha, and karanja consisting of 120 data points each. This set of data are used for the ANN development and validation. A multi-layer perceptron network is trained by the experimental data for predicting the engine parameters. The Nash Sutcliffe coefficient of efficiency values for the ANN predicted parameters are close to 1, indicating a close prediction. The ANN model predicted the engine output parameters with low values of mean square error, mean square relative error, mean absolute percentage error and standard error of prediction. Optimum values of biodiesel blend fraction, engine speed, brake mean effective pressure, injection pressure and timing are obtained using a multi-objective genetic algorithm. The optimised blend concentration is found to be ∼20% and ∼40% for satisfying the different objectives concerning the overall engine characteristics. Finally, the outputs for the optimised parameters are compared to the validated multi-zone model predictions within the maximum error of ∼3% and 7.9% for performance and emission parameters respectively.
Read full abstract