Machine learning-based model predictive control for balancing of combustion in hydrogen/diesel dual-fuel engine

Utilizing hydrogen as a secondary fuel in internal combustion engines is a promising method for significantly reducing greenhouse gases in the transportation sector. However, injecting secondary fuels with port injection in multi-cylinder engines introduces variability in combustion metrics, such as peak pressure (PP) and maximum pressure rise rate (MPRR), which increases emissions and leads to reduced engine durability. This variability appears as either cycle-to-cycle or cylinder-to-cylinder variation, ultimately resulting in decreased engine performance. This study presents a machine learning-based nonlinear model predictive control strategy for achieving real-time combustion balancing in a multi-cylinder hydrogen/diesel dual-fuel engine. Experimental results demonstrate a mean absolute error of 0.2 bar for tracking the indicated mean effective pressure (IMEP) reference. Differences in IMEP between cylinders are reduced by up to 87% compared to the benchmark. The coefficients of variation for PP and MPRR have decreased by 29.6% and 5.5%, respectively, among the six cylinders. The results show that the proposed controller effectively minimizes cylinder-to-cylinder variations while maintaining all combustion and safety constraints.