Abstract
Different meta-heuristic optimization algorithms have been used in a variety of fields due to their intelligent behavior and fast convergence. However, use of these algorithms in the engine behavior optimization is limited. The development of so-called hybrid optimization technique when these algorithms are combined with experimental design technique is an upcoming method in the field of renewable energy. Hence in this research, meta-heuristic optimization algorithms and experimental design methods were combined to optimize the engine behavior. Additionally, artificial neural networks (ANN) were employed to forecast the performance and emission behaviors of a CI engine running on a novel hybrid biodiesel blend of Cucurbita pepo. L (pumpkin) and Prosopis juliflora, mixed with a novel Elaeocarpus ganitrus (Rudraksha) additive. To assess the success of the ANN, four statistical benchmarks (R2, MAE, RMSE, and MSE) were used. Experiments were designed according to Design of Experiments (DOE) rules with performance and emission parameters as outputs. Response surface methodology (RSM) was employed to find the effect of interaction factors. Single objective and multi-objective optimization using highly efficient hybrid RSM-particle swarm optimization (RPSO) and dragon fly algorithm (RMODA) were employed to optimize the response of the obtained RSM equations. The outcomes demonstrated that RSM and ANN were excellent modelling techniques for these kinds of situations, with good accuracy. In addition, ANN's prediction performance (R2 = 0.978 for BTE) was somewhat better than RSM's (R2 = 0.960 for BTE). On the other hand, the PJB20 blend with 5 mL additive increased BTE by 52.8% and reduced BSFC by 34.9% at maximum load. The smoke opacity was lowered by 7.1% when compared to pure diesel, without any engine modifications. CO2 emission was seen to be shortened by 19.14%. Finally, it can be concluded that this novel biodiesel can be possibly utilized in CI engines with no modification and the engine characteristics can be controlled by optimization and prediction models.
[Display omitted]