Livermore Helps Reduce the Development Time of Cleaner Burning Engines
Reaction models, simulation software, and prediction tools illuminate the complexities of combustion chemistry, and assist designers in their quest for engine efficiency.
The Need for Cleaner Burning, More Efficient Engines
Combustion engines move billions of people around the globe everyday, but they do not operate as energy-efficiently as possible. Engines used for transportation range in efficiency from 15 percent to as high as 50 percent among the most advanced vehicles, but many engines manufactured today are only about 30 percent efficient on average. Optimizing control of combustion through better engine and fuel design could improve the fuel economy of vehicles and reduce their harmful emissions. However, building a large number of prototype engines would be cost prohibitive and time consuming. Researchers at Lawrence Livermore National Laboratory are using their expertise in fundamental chemistry, machine learning, advanced algorithms, and predictive simulation to put improved tools in the hands of the automotive industry's engine designers.
Clean Combustion Supports Energy Security Mission
The Laboratory innovates science and technology solutions for reliable energy supply, cybersecurity and resilient infrastructure, and homeland and defense systems security. Clean combustion research and development helps fulfill this mission by reducing harmful emissions and fuel consumption of engines for transportation.
Better Simulations of Combustion
The chemistry of combustion is complex and still not fully understood. Burning a fuel in air results in a set of combustion products that include water vapor, carbon dioxide and carbon monoxide, and nitrogen oxides. However, the process that transforms a common transportation fuel such as gasoline into these products involves several thousand intermediate molecule species and chemical reactions. For decades, scientists at Livermore have conducted research on combustion chemistry, working to shed light on the many reaction pathways that fuels take as they burn inside engines (see combustion.llnl.gov) and creating the most accurate models of this process in the world. Although highly accurate, few industry designers have the computational resources to run simulations that could take weeks to resolve tens of milliseconds of an engine test. Livermore researchers are closing that simulation gap with their Zero Order Reaction Kinetics (Zero-RK) software package, which simulates combustion more efficiently and brings predictive fuel chemistry to the design process.
Zero-RK is built to take advantage of the massively parallel computing architectures that will be deployed as part of the Department of Energy's Exascale Computing Project. Toward this goal, the Laboratory has been collaborating with Oak Ridge National Laboratory and General Motors to demonstrate a completely virtual engine calibration using Zero-RK on the Laboratory's Titan supercomputer.
Zero-RK can also run on workstations and computer clusters operated by industrial users. By applying Zero-RK in conjunction with off-the-shelf computational fluid dynamics software, engine designers are reducing their simulation time by orders of magnitude.
Fuel Optimization and Discovery with Machine Learning
The Laboratory's predictive transportation fuel models and high-performance combustion solvers allow researchers to look for new fuel blends that improve engine performance, are produced by lower energy paths, and reduce costs. For example, the Laboratory created an artificial neural network model that predicts the octane rating of complex gasoline surrogates. This new model incorporates data from more than 700 experimental tests published in the last 10 years and uses ignition predictions from detailed reaction models as inputs combined with molecular structure and thermophysical properties to improve prediction accuracy.
By applying this approach, researchers can model realistic transportation fuels using a variety of almost 50 components including promising renewables. Previously, typical fuel models would be limited to fewer than 5 compounds, and octane predictions could be off by as much as 10 octane numbers. The resulting neural network can provide accurate predictions for an array of compounds, even those with limited test data. For example, the model was tested on 10 blind predictions for new gasoline blends and was accurate to within 2 octane numbers. This unprecedented level of accuracy for wide-ranging fuel components enables the design of new fuel blends and fuel-engine co‑optimization.
The central feature of the Combustion Chemistry project at LLNL's Physical and Life Sciences Directorate is the development, validation, and application of detailed chemical kinetic reaction mechanisms for the combustion of hydrocarbon and other types of chemical fuels.