Predictive Analytics for Hydrogen–Honge Oil Dual Fuel Engine Using Machine Learning

Abstract

Plant based fuels have been incessantly researched as a substitute of diesel. However, their use in compression ignition engine tends to deteriorate engine performance and increase the engine emissions. Thereby, offsetting the advantage of the fuel being carbon neutral. Hydrogen induction in intake manifold with the direct injection of biofuel enhances the engine performance and simultaneously reduce the emissions. This way diesel can be completely substituted with carbon neutral fuels. In the present study to replace diesel, honge oil was transesterified to its methyl ester and used as the direct injected fuel and the flow rate of hydrogen was varied. The results indicate an improvement in engine performance with higher in-cylinder pressure and heat release rate with 30 L per minute (lpm) flow rate of hydrogen. The brake specific energy consumption (BSEC) of the engine was reduced to 12.2kJ/kWh with the highest flow rate of hydrogen at full load condition. The unburnt hydrocarbon emission, carbon monoxide emission and smoke opacity reduced from 30 ppm, 0.8% & 59% to 19 ppm, 0.48% & 47%, respectively with maximum flow rate of hydrogen. However, due to improvement in combustion, the oxides of nitrogen emission increased from 1224 ppm to 1450 ppm with hydrogen premixing. For the same engine, if fuel is varied then extensive experimental study is required for analysing the engine performance which is costly, time consuming and may itself be a source of pollution. If the exhaust emissions can be accurately predicted with variation in fuels, then the previously mentioned issues can be resolved. In this regard novel features namely percentage of carbon, hydrogen and oxygen present in the fuel were used along with hydrogen flow rate. Various algorithms were compared for predicting the emissions. The results show that the lowest mean absolute percentage error (MAPE) and root mean square error (RMSE) of 0.78, 0.00647, 0.074, 0.00391, 0.155 and 0.013 was observed with support vector regression (SVR) for CO, HC and smoke emissions, respectively. While gradient process regression (GPR) algorithm resulted in lowest error of MAPE (0.22) and RMSE (0.65) for NO emission.