Machine learning assisted modeling of ignition delay in a light-duty gasoline compression ignition engine

• Low Temperature Combustion (LTC): LTC is a promising approach to achieving carbon neutrality in transport through the use of renewable and synthetic fuels in hybrid and plug-in applications. • Machine Learning (ML) for Control Strategies: ML serves as a valuable modeling tool, offering cost-effective testing for control strategies in the field of engine research. • Ignition Delay in Sustainable Gasoline-like Fuels : Sustainable gasoline-like fuels often exhibit a long ignition delay (ID), which may compromise the benefits of LTC methodology. • Integration of Physical Knowledge and Machine Learning: The combination of physical knowledge with ML methodology is recognized as a key enabler for modeling complex physical phenomena. This work proposes an innovative artificial neural network-based (ANN) approach to predict the ignition delay (ID) in a Gasoline Compression Ignition (GCI) engine using the information coming from standard sensors mounted on the engine. Moving toward the carbon neutrality of transport by using renewable and synthetic fuels, GCI combustion is considered a promising technology to achieve high engine efficiency and ultra-low pollutant emissions. As with other auto-ignition combustion concepts, a deep understanding of compression ignition dynamics is crucial for maintaining a stable and controllable combustion process in different operating and environmental conditions and enhancing engine performance and durability over time. Machine learning (ML) offers a promising modelling tool to lower the cost for testing control strategies compared to traditional physical or empirical approaches. An experimental campaign was conducted with 293 steady-state engine operating conditions to establish relationships between ignition delay and engine parameters, such as engine speed, load, intake and injection pressures, exhaust gas recirculation (EGR), and injection parameters, in a light-duty GCI engine. Then, an ANN-based model was trained and validated using a combination of holdout and k-fold cross-validation methods, along with a Bayesian regularization algorithm. The ignition delay estimation through the ANN-based model has shown a NRMSE percentage of 2.16% and R 2 of 0.99 on test dataset, demonstrating to be accurate enough for engine control and diagnosis purposes. This work aims at inspiring innovative ANN-based control strategies for promoting the use of advanced combustion methodologies, such as GCI combustion, as technical solution for production engines.