Session: 06-01 SI Engine Modeling

Track: Track 6: Modeling and Simulation

Paper Number: 91016

91016 - Predicting Combustion Variability Using Machine Learning From the Flow Field Data at Spark Timing for a Gasoline Direct Injection Engine 

Machine learning techniques are used to predict the variability of peak cylinder pressure using the flow field data at spark timing. A gasoline direct injection engine is simulated using computational fluid dynamics (CFD) with the Large Eddy Simulation (LES) approach at two operating points (with and without exhaust gas recirculation). Such engines are susceptible to large cycle-to-cycle variability (CCV). Multi-cycle engine simulation data whose results statistically compare well with those of the experiments in a prior validation is used in this work. It is first shown that the flow field just before spark has a deterministic relationship with the subsequent peak pressure. The Pearson Correlation Coefficient (PCC) was computed for a selection of flow field variables right before spark with the subsequent peak cylinder pressure. Variables with significant PCC were selected as features for the machine learning algorithm. The machine learning algorithm was then able to predict the peak cylinder pressure for engine cycles that the ML algorithm had never seen before (10% of the data was reserved for testing). Using the above ML prediction, one can determine the CCV of the engine without having to run expensive combustion simulations by running multiple CFD simulations (30-100) only until the spark timing and the ML algorithm will then be subsequently used to predict the spread of the peak pressure.

Presenting Author: Nitesh Attal Convergent Science, Inc