Optimization of ANN Models Using Metaheuristic Algorithms for Prediction of Tailpipe Emissions in Biodiesel Engine

Abstract

Machine learning techniques are gaining momentum in the present-day context in most engineering applications due to their versatility and accuracy. They facilitate faster data processing coupled with a high degree of accuracy. They are extensively used in understanding and modeling engine combustion and emissions. Engine emissions significantly contribute to environmental degradation. In the current study, an effort has been made to compare the emissions recorded from a four-stroke single-cylinder biodiesel engine with those obtained using artificial neural network (ANN) models, where the hyperparameters have been optimized using nature-inspired metaheuristic optimization algorithms like JAYA, WOA, ROA, and WaOA. The study was conducted using diesel and cardanol-methanol-diesel blends of B10M10, B20M10, and B30M10, by varying the fuel injection pressure from 180 bar (standard injection timing) to 220 bar with an interval of 20 bar. Furthermore, experiments were conducted with oxygen enrichment at concentrations of 3%, 5%, and 7% w/w on the standard oxygen concentration of air. The study showed a remarkable reduction of 59% in CO emissions at 220 bar fuel injection pressure with 7% w/w oxygen enrichment for the B30M10 blend as compared to 180 bar without oxygen enrichment. A similar reduction of 32.6% and 16.6% were observed for HC emissions and smoke opacity for the same operating conditions. However, a rising trend of 50% was observed for NO_x emissions for the same blend and operating conditions. The findings indicate that the data recorded conforms with that obtained by using the ANN model optimized through these metaheuristic algorithms.