Comparative study using RSM and ANN modelling for performance-emission prediction of CI engine fuelled with bio-diesohol blends: A fuzzy optimization approach

Abstract

Abstract This present investigation focuses on prediction of engine responses of a single cylinder CI engine powered by bio-diesohol (diesel-palm biodiesel-ethanol) blends using RSM (response surface methodology) and ANN (artificial neural network) model. RSM combined with multi-level general full factorial design (FFD) is used for the prediction of brake thermal efficiency (BTE), brake specific fuel consumption (BSEC), and nitrogen oxides (NOx). The engine experimental data is trained in ANN model using Levenberg-Marquardt back propagation training algorithm with logistic-sigmoid activation function. Different statistical measures are calculated to quantify the errors and correlations of the predicted models. Comparatively lower prediction error and higher correlation have been observed from the ANN model compared to RSM. The range of overall mean square error (MSE) and correlation coefficient are found (0.0003–0.00059) & (0.99403–0.998) and (0.00019–0.00035) & (0.99943–0.99971) from RSM and ANN model respectively. The range of overall mean absolute percentage error (MAPE) from ANN model (3.13–4.55%) is found lower compared to RSM model (3.97–6.6%). Thereafter, RSM and ANN predicted responses are introduced in fuzzy logic system for the optimization of engine operating parameters. At 100% load, the D75B20E5 (75% diesel\xa0+\xa020% palm biodiesel\xa0+\xa05% ethanol) blend has been found best for the optimization of BTE, BSEC and NOx emission. Finally, after the confirmation test, it has been revealed that the performance of D75B20E5 blend is as comparable to diesel.