Evaluation of control-oriented flame propagation models for production control of a spark-assisted compression ignition engine

Abstract

The drive to improve internal combustion engines has led to efficiency objectives that exceed the capability of conventional combustion strategies. As a result, advanced combustion modes are more attractive for production. These advanced combustion strategies typically add sensors, actuators, and degrees of freedom to the combustion process. Spark-assisted compression ignition (SACI) is an efficient production-viable advanced combustion strategy characterized by spark-ignited flame propagation that triggers autoignition in the remaining unburned gas. Modeling this complex combustion process for control demands a careful selection of model structure to maximize predictive accuracy within computational constraints. This work comprehensively evaluates a physics-based and a data-driven model. The physics-based model produces a burn duration by computing laminar flame speed as a function of test point conditions. The crank-angle domain is intentionally excluded to reduce computational expense. The data-driven model is an artificial neural network (ANN). The candidate models are compared to a one-dimensional engine model validated to experimental SACI engine data. Though both models capture the trends in burn rates, the ANN model has a root-mean square error (RMSE) of 1.4 CAD, significantly lower than the 10.4 CAD RMSE of the physics-based model. The exclusion of the crank-angle domain results in insufficient detail for the physics-based model, while the ANN can tolerate this exclusion.