Numerical simulation of nitric oxide (NO) emission for HCNG fueled SI engine by Zeldovich’, prompt (HCN) and nitrous oxide (N2O) mechanisms along with the error reduction novel sub-models and their classification through machine learning algorithms

Abstract

The zero-carbon emission can be achieved with utilization of hydrogen fuel in transportation sector but NOx emission rises. The experiments have been conducted at different operating conditions: engine speed (700 to 2000 RPM), load (25 to 75 %), MAPs = 65 to 178 kPa, spark timings (0 to 60°CA bTDC), EGR ratios, (0 to 0.33), hydrogen percentages 0 to 40 % in CNG fuel at stoichiometric operating condition to measure the in-cylinder pressure and NOx emission. The in-cylinder temperatures have been obtained through calibrated quasi-dimensional combustion model of HCNG engine. The in-cylinder pressure and temperature have been applied in simulation of NO mechanisms such as: thermal, prompt, and nitrous oxide mechanisms. Furthermore, the parametric properties of in-cylinder temperature, pressure, and nitric oxide mechanisms have been studied at different operating conditions.In second part, the traces of nitric oxide mechanisms have been transformed into positive real number with inclusion of six different sub-models that are: arithmetic mean, trapezoid integration, and author’ established three zone models (TD = 50, 100, 150 and 200 K). The three zone sub-model (TD = 100 K) has better prediction accuracy as compared to other sub-models, but it is still quite high. The prediction accuracy is improved with combination of different sub-models: 6C2, 6C3, 6C4, 6C5 and 6C6 iterations. The optimized MAPE is 24.9471 % corresponding to the four combinations of sub-models: arithmetic mean, trapezoid integration, three zone models (TD = 50 & 100 K). The best of each combination has been trained through machine learning classifier with total 6-algorithims that are Tree, Discriminant, Naïve Bayes, Support vector machine, KNN & Ensemble along with 24-functions to identify the range of different sub-models in a specific combination. The best prediction accuracy is corresponding to Ensemble with sub-space KNN classifier.