Prediction of Engine Knock in a Gasoline Direct Injection Engine

Project Description

Increasingly strict fuel-economy and emissions regulations have challenged the automotive industry to invest heavily in innovative powertrain technologies. These advancements can result in a complex array of tunable control parameters to optimize engine performance over a range of operating conditions. Hence, the calibration process of a modern passenger-car Diesel engine consumes significant time, effort and resources, making it a bottleneck during the development process. Advanced modeling tools, such as CFD, are often used with the goal of streamlining portions of the calibration process. The usefulness of CFD simulations tools for in-cylinder engine combustion is often compromised by the computational overhead of detailed chemical kinetics, uncertainty in the combustion chamber wall temperatures, and complex interactions of fluid flow, chemistry, and heat transfer. Specifically, traditional Diesel engine CFD simulations consist of partial geometry sector mesh computations utilizing reduced-order kinetics mechanisms, fixed spatially uniform wall temperature boundary conditions, and a prescribed solid body swirl velocity field prior to spray injection in lieu of computing air induction with valve motion. This project seeks to leverage DOE’s leadership computing resources at OLCF and recent advancements in numerical simulation to improve the accuracy of CFD computations and accelerate engine calibration. First, higher-order kinetics will be solved using a GPU-based chemical kinetics solver (leveraged in past ALCC awards CMB119 and CMB124 [2016&2017]) integrated with the CONVERGE™ CFD modeling software. Full in-cylinder 3D spray, flow and combustion simulations will be undertaken considering conjugate heat transfer (CHT) to predict temporally and spatially varying wall-temperature boundary conditions. This year’s project will extend the practical state of the art by additionally employing a Large Eddy Simulation (LES) turbulence model with highly refined mesh to increase verisimilitude of the simulations. The results will be analyzed to compare differences in predicted combustion performance and emissions including nitrogen oxides (NOx), carbon monoxide (CO), unburned and partially burned hydrocarbons (HC), and smoke (i.e., soot) with actual engine measurements. The advances made in this project are anticipated to assist in bringing highly detailed simulations of engine combustion to new opportunities afforded by Summit and pave the way to applications on exascale computing resources.