Comparison of library-based and detailed chemistry models for knock prediction in spark-ignition engines

The present engine development pathway for increased specific power and efficiency is moving Spark-Ignition engines towards unprecedented levels of mean thermo-mechanical loading. This in turn promotes undesired abnormal combustion events in the unburnt mixture (also called “engine knock”), leading to solid parts failure and constituting a severe upper constraint to engine efficiency. In this context, CFD simulations are regularly used to investigate the fluid-dynamic reasons for engine knock and to address knock suppression strategies, using dedicated models to simulate the chemical reaction rate of the fuel/air/residual mixture at the same thermodynamics states as those encountered in engines. In this paper three different approaches are coherently compared to simulate knock occurrence on a turbocharged GDI engine, representing some of the most popular choices for modelers in the RANS framework. The first one considers the on-the-fly solution of chemical reactions, which represents the state-of-the-art knock modelling approach albeit its problematic computational cost for industrial turnaround times. The other two methods consider pre-calculated libraries of ignition delay times (calculated at constant pressure and volume, respectively) for the same fuel model, and knock timing is predicted using a classical Livengood-Wu approach coupled to the same main combustion model. All the analyzed models for the end-gas reaction rate are coupled with a dedicated combustion model for propagating flame (G-equation). A comprehensive analysis of computational cost and of knock prediction accuracy is carried out for library-based methods against the detailed chemistry model. Finally, results are critically discussed and explained using combined ignition delay time maps and traces for thermodynamic in-cylinder states, and guidelines for the a priori choice for constant pressure- or volume-generated libraries are provided. In this context, the use of a synthetic knock model combined with libraries of ignition delays calculated at constant volume emerges as an accurate and efficient modelling strategy. The study outlines a method for the well-supported use of simplified CPU-efficient models, with a promoted confidence in simulation results from the comparison with detailed chemistry.