Integration, improvement, and validation of the ACID model in KIVA3V CFD simulation for predicting SOC in HCCI engines

Homogeneous Charge Compression Ignition (HCCI) is an emerging technology, which improves the emissions and efficiency of the internal combustion engine by combining advantages of both Spark-Ignition and Direct Injection Compression-Ignition engines. However, controlling the Start of Combustion (SOC) in HCCI is the toughest challenge due to the lack of a direct control mechanism. HCCI combustion is mainly dominated by chemical kinetics instead of fluid dynamics, which makes numerical research in HCCI more convenient to develop control strategies for HCCI applications. Although HCCI features well mixed charge, inhomogeneities within the HCCI engine still need to be included. Hence, the Computational Fluid Dynamics (CFD) model, KIVA3V, can be applied to compute the influence of inhomogeneities in HCCI engines with detailed chemistry. The most difficult issue arising from the use of KIVA3V, however, is the expensive computational cost involved, which slows down the analysis and development for HCCI control strategies. The main objective for this research is to investigate how to improve the efficiency of HCCI computation using KIVA3V involving chemistry without losing prediction accuracy by utilizing the Artificial Neural Network (ANN) Combined with Ignition Delay (ACID) model and also by seeking improvements to the ACID model.; In this study, the ACID model combined with KIVA3V is explored. The combined model is found able to reduce computational time without sacrificing combustion characteristics because the essential chemistry is kept by ANN. In order to improve the SOC predictions of the original ACID model for more accurate HCCI predictions, a 3-section ACID model is developed and validated. The SOC predictions given by the 3-section ACID model agree better than the original ACID model because the 3-section ACID model lowers the average error of ANN. This work also explored if a reduced chemistry for a large chemical mechanism, i.e. isooctane with 857 species and 3,606 steps, can be utilized to improve the computation efficiency. A skeletal mechanism, Skeletal291, is derived from the LLNL detailed isooctane mechanism, displays good agreement on both SOC and emissions predictions. It is demonstrated that reduced chemistry is useful for improving the computational expense and therefore helpful in developing control strategies for HCCI research.