Lubricant formulation effects on ultra-fine particles emissions from gas fuelled engines: experimental investigation using machine learning in data analysis

Formation and emission of solid particles from gas engines are strictly correlated with lube oil consumption and combustion phenomena, in turn linked to lubricant characteristics. Alongside with the research advancement in the understanding of the complex mechanisms behind particle formation, Machine Learning (ML) algorithms can be useful for an in-depth analysis of the cause-effect relationship between oil formulation and particle emissions, representing a powerful tool for the design of oil-engine coupling. The present work introduces a first approach on the use of ML algorithms to estimate particle emissions, varying lube oils formulation. An experimental campaign was performed on a Heavy-Duty gas engine running on World Harmonized Transient Cycle and multiple oils with different properties were tested. The results confirmed lubricant formulation as a strong tool in particles emissions control and provided a first database for the ML analysis. Convolutional Neural Network (CNN) algorithms were selected, trained through a set of oil data and tested to predict emissions in correspondence of different oils. The study also included the investigation of artificial algorithms for oil properties optimization, in terms of emissions and fuel consumption. Although in initial phase, the study revealed the CNN predictions quite accurate, well reproducing the experimental observed trends in particle emissions, varying the oil formulation. • Particle emissions from Heavy-Duty Compressed Natural Gas Spark Ignition engines. • Oil lubricants formulation as powerful tool in particles emissions control. • Machine Learning to analyze oil formulation and particle emissions relation. • Convolutional Neural Network algorithms for PN and Soot predictive analysis. • Lubricants optimization analysis by means of Monte Carlo Method.