Ultra low engine-out particle emissions from gas fuelled engines through lube oil formulation: evidence from an experimental campaign and machine learning based approach for data analysis

The transportation industry has undergone major changes in the past years, switching from traditional fossil fuels supply to alternative and cleaner ones. This is true especially in the field of heavy-duty (HD) engines, where Compressed Natural Gas is currently the most valid substitute of Diesel for off-road and transport applications. The upcoming stringent limits on sub-23 nm particles emissions will regard also gas fuelled engines; their compliance with future regulations requires great technological effort. The present work is addressed to the investigation of strategies to reduce particles emitted from HD gas engines. In particular, focus of the activity is the study of the potentiality offered by lubricant oils formulation in particle number (PN) emissions control. Indeed, particle emissions from gas engines come mainly from the combustion of lubricant oil leaking into the engine combustion chamber. An extensive experimental campaign on an HD SI engine with different lubricants was performed, identifying the formulation of oils as a strong tool for PN and soot reduction. Considering that Machine Learning (ML) is one of the most promising approaches in solving real-world problems by means of available data, the second part of the activity, still ongoing, was devoted to the evaluation of ML algorithms in PN and soot emissions predictions. To this aim, a Convolutional Neural Network (CNN) was trained through a set of oil data and tested to predict emissions in correspondence of different oils from the training ones, which were compared with the acquired data. Although the study is still ongoing, first results show that the CNN predictions proved to be accurate, as they place close to the mean value of a standard statistic distribution of the experimental data.