Single Cylinder Research Engine Combustion Model with Integrated Laminar Flame Speed Neural Network MetaModel and Knock Induction Time Integral Evaluation

This study develops a 0D combustion model for a single-cylinder research engine, integrating a neural network-based laminar flame speed model. Experimental data were obtained with RON95E10 fuel at 23 stoichiometric engine operating points. The neural network, trained on a grid of pressure, temperature, equivalence ratio, and EGR, was coupled with the combustion model, whose turbulent flame speed parameters were calibrated using 18 points and tested on the remaining 5. The model reproduces the centre of combustion and crank angle at maximum pressure with a root mean square error of 1°, while the maximum pressure deviation is approximately 3 bar for the testing dataset. Knock tendency was assessed using the knock induction time integral with ignition delays from the C3MechV3 mechanism, identifying knock-prone conditions at a threshold of 0.9. Results demonstrate the method’s accuracy and highlight the value of machine learning for combustion modelling of RON95E10 fuel.