A digital firework
Researchers from ETH Zurich have simulated autoignition in a turbulent flow using a supercomputer with up to 65,000 processors in one of the largest reactive flow simulations to date. The results could help to develop better models and reduce the high cost of real experiments.
A team of researchers from the Laboratory of Aerothermochemistry and Combustion Systems (LAV) headed by Professor Konstantinos Boulouchos recently presented initial results of one of the largest reactive “Direct Numerical Simulations” (DNS) to date. The simulated autoignition of hydrogen in a turbulent hot air coflow is essentially a laboratory experiment performed on the computer. The fuel ignites by itself when its pressure and temperature is high enough. “The main advantage of such a numerical simulation over a “real” experiment is that data like the flow velocity, temperature, pressure and concentration of the individual chemical components can be measured accurately with a high temporal and spatial resolution. It would simply be impossible to obtain such detailed information in a real experiment”, explains Christos Frouzakis, head of the DNS group at LAV.
100 terabytes of data for 11 milliseconds
The software used for the simulation is the result of a long collaboration between the LAV group and the Argonne National Laboratory (ANL) near Chicago, and had already been used successfully in less demanding computations. In order to make the simulation on tens of thousands of processors possible, Stefan Kerkemeier, a doctoral student at LAV, and Paul Fischer from ANL optimized and further expanded the code over the last two years. . The simulation was performed on the IBM BlueGene/P at ANL, one of the world’s largest high-performance supercomputers. It required about 12 days on up to 65,000 processors to calculate 11 milliseconds of real time. According to Kerkemeier, at the time of the computations (December 2008), such a simulation was not possible in Switzerland due to lack of computational resources. . Over 100 terabytes of data were collected with a temporal resolution of 3 microseconds and a spatial resolution of 30 micrometers. “The wealth of data is enormous. The next big challenge is to analyze the results of the simulation and understand the processes involved”, stresses Kerkemeier.
For more efficient engines and energy from hydrogen
A better understanding of autoignition is necessary for the development of more efficient engines. So far, the simulation used hydrogen as the fuel. “We used hydrogen because the computational time would be too long with petrol or diesel due to their complex chemical composition”, explains Konstantinos Boulouchos. However, hydrogen could become increasingly important in the future, according to Boulouchos. For example when coal or natural gas is converted into hydrogen in order to be burnt in a gas turbine, while carbon dioxide is separated before the combustion chamber.