Machine Learning Based Surrogate Models for Knock Prediction in Syngas (H2/CO) Added SI Engine Combustion

Abstract

Synthesis gas (syngas) co-combustion in spark ignition (SI) engines has been proven to greatly benefit greenhouse gas reduction and combustion quality improvement. There are, however, still differing opinions on the effect of syngas addition on the knocking phenomenon under boosted conditions: Some suggest that syngas reduces knock by inhibiting end-gas auto-ignition, while others claim that syngas increases knock due to the high pressure rise rate caused by the excessive flame speed of hydrogen. Therefore, this study examines the syngas addition effect on knock intensity and unveils the black box relation between combustion processes and knock intensity with syngas using machine learning based surrogate models. A boosted single-cylinder research engine equipped with a syngas injector and a gasoline port injector was utilized to gather the data over a range of boost levels and spark timing, and seven machine learning based surrogate models were trained on the combustion data. The selected regression model predicted knock intensity given an input space comprising engine control factors and calculated combustion phase data. The regression models were auto-tuned to find the best predicting model over 20,000 samples. The results show the trained model can explain knocking intensity over 70%. Syngas addition extends the knock limit in all boost conditions, and the effect is stronger as increasing the intake pressure. In addition, the model showed the initial flame kernel development phase has the strongest relation with knock intensity compared to other combustion phases.