GE Global Research Center is one of the world’s most diversified industrial research labs, providing innovative technology for all of GE’s businesses. Global Research has been the cornerstone of GE technology for more than 100 years, developing breakthrough innovations in areas such as medical imaging, energy generation technology, jet engines, and lighting. In the energy industry, GE Global Research is developing new technologies to reduce the cost of renewable energy, make turbines more efficient and fuel-flexible, and build a 21st century smart grid that can support a more diversified energy portfolio.
Improving Models of Spray Breakup in Liquid Fuels Combustion
Stringent emission requirements and increasing emphasis on reducing fuel consumption to stay competitive in the marketplace have forced automobile and aviation engine manufacturers to seek out new technologies for their engines. These increasingly complex technologies consist of subsystems and components that have to follow an intensive design-optimization process, in which prototypes are designed, fabricated, and comprehensively tested. Frequently, for components such as fuel injectors, this process must be repeated several times before an optimal design is obtained — an iterative approach that can increase development cycle times and associated costs.
The need to increase engine thermal efficiency while simultaneously driving down emissions makes the underlying technologies for the combustion of liquid fuels increasingly important. Chief among these challenges is a detailed understanding of unsteady fuel spray phenomena. Detailed numerical simulations are a promising approach to quantify unsteady spray breakup but are computationally expensive because of the wide range of length and time scales that characterize the coupled, nonlinear spray dynamics. These simulations require robust and accurate numerical methods to ensure discrete mass, momentum, and energy conservation. With the power of supercomputers, these simulations can accurately capture liquid fuel behavior, opening new pathways to better engine technology.
In this collaboration GE Global Research and LLNL through its HPC Innovation Center will demonstrate that high-performance computing (HPC)—coupled with robust, accurate numerical methods for studying liquid breakup—can potentially reduce the number of design iterations needed to create advanced fuel injectors, thereby giving engine manufacturers a competitive edge. These advantages are expected to reduce development costs for new engine designs and time to market. This project will pave the way to adopting HPC for optimizing fuel injector design to achieve high overall combustion efficiency, reduced fuel consumption, and reduced emissions. The potential long-term benefits of using HPC in fuel injector optimization include reducing the use of foreign oil and reducing the environmental impact of next-generation automobile and aviation engines.
Ongoing Work
Simulating fuel injector sprays using high performance computing allows researchers to better understand the physics underlying the breakup of liquid fuel in combustion chambers. Researchers from GE Global Research, Arizona State University, Cornell University and LLNL are testing the benefits of using HPC to simulate two different injector geometries. One injector geometry corresponds to the classical liquid jet in crossflow configuration, shown below, while the other is a generic dual orifice injector geometry. The goal of these simulations is to capture the phenomena occurring at very fine scales, with the current smallest mesh size (or distance between calculation points) of around 20 – 40 micrometers (microns).
For the round-edged and sharp-edged injectors, the calculations require six hours to complete one full flow through interval on 1440 cores each. Experimental datasets from Georgia Institute of Technology [Gopala, Y., Ph. D. Thesis, 2012, Lubarsky et al. 2010] are being employed to validate the simulations and ensure that applicable physics associated with liquid jet breakup are being captured by the computational codes.
For the generic dual orifice injector, it took two days of calculations on 696 cores to complete calculations from the start of injection to reach a steady state. These simulations have shed insight into phenomena that are challenging to visualize using experiments alone. For example, the simulations shown on the left shed insight on the transient formation of the central air core which then leads to the formation of the liquid sheet from the central primary liquid circuit of the dual orifice injector.
Sensitivity of the computational results to mesh size is being investigated. Simulations for both the liquid jet in crossflow configuration and the dual orifice injector geometry on about 5 times the current mesh size (decreasing the distance between calculations by a factor of 1.6) on 6,000-10,000 processor cores are being planned.
Prior to the hpc4energy incubator
Fuel injector design historically relies heavily on experimental testing. Typically, prototypes for each design of the fuel injector are built and measurements are undertaken to study the behavior of the spray. While measurements are mostly conducted external to the injector, the region of interest where the initiation of the liquid breakup occurs is hidden within the confines of the injector. By employing accurate, robust and scalable numerical methods for spray breakup from the fuel injector on massively parallel HPC resources, the physics underlying fuel spray breakup can be better understood and fuel injector designs can potentially be optimized using HPC resources. An ability to predictably simulate liquid fuel breakup on a computer has implications on being able to study fuel injectors at conditions where experiments can be challenging to undertake.
The two software codes that have been employed for this project have been applied to fuel spray break up in earlier projects. Examples of fuel spray breakup from simulations on other computation systems produced prior to the hpc4energy collaboration are shown below.
Arizona State University code (Prof. Marcus Herrmann): This is an 
unstructured, finite volume code that employs a body-fitted mesh to model complex geometries. The figure to the right shows a liquid jet in crossflow simulation performed using the ASU code. Details of the code architecture and other test cases studied can be obtained at http://multiphase.asu.edu/.
Cornell University code – NGA (Prof. Olivier Desjardins): This is a structured, finite 
difference code that employs an “immersed-boundary” technique to model complex geometries. The figure to the right shows a liquid jet issuing into a quiescent ambient. Details on the code architecture and other test cases related to liquid breakup that have been studied using NGA are available at http://ctflab.mae.cornell.edu