|The simulation computed profiles of temperature (left) and black carbon (right) formation for a bio-jetfuel flame. Click to enlarge.|
Researchers at the University of Toronto, collaborating with colleagues in the fields of fuel, aerospace, and combustion, have performed what they say are the largest yet simulations of bio-jet fuel combustion, making it now possible to identify fuel mixtures which reduce soot formation while minimizing greenhouse gas emissions.
Meghdad Saffaripour, Seth Dworkin, and Murray Thomson, of the Mechanical and Industrial Engineering department, are using the computing facilities of Canada’s SciNet to study the effect of using bio-jetfuels in place of conventional fuel. Renewable bio-fuels, made of plants which absorbed carbon dioxide as they grew, can emit significantly less net greenhouse gasses than fossil fuels; but they are more complicated chemically.
They announced their results yesterday at the 24th HPCS, Canada’s largest supercomputing conference.
These simulations had previously been possible only for simple fuels. The more complex the fuel, the harder it becomes to simulate it accurately.
Combustion simulation for the last 25 years has largely been confined to the realm of simple fuels—such as those with one to three carbon atoms per fuel molecule, including methane, ethylene and propane, the researchers said. This is partly due to the difficulty in developing the libraries of chemical reactions that are involved in combustion, and to the relationship between fuel and computation cost; the more complex the fuel, the larger the chemistry library, and the more computationally intensive the calculation becomes.
The researchers, working with combustion chemistry experts at the CNRS in Orléans, France, generated a chemistry library that tracks 304 distinct chemical species—twice the size of that used in comparable calculations to date, they said.
The research group has spent nearly a decade developing algorithms that would be specifically well-suited to parallelization. They achieved this by coupling a detailed computation fluid dynamics (CFD) module to an implicit chemical reactor simulator wherein the flame is subdivided into thousands of tiny cells (the reactors) in which combustion occurs and particulate matter may be formed.
The physically accurate coupling of combustion and fluid dynamics is achieved via an iterative process that globally refines the fluid properties, and permits network communication to share the results of each chemical reactor simulation. The resulting semi-implicit algorithm is scalable and can effectively handle the stiffness of the highly non-linear set of partial differential equations that govern fluid dynamics and combustion.
|SciNet IBM TCS|
The group’s results required 192 processors on SciNet’s IBM Power-6 TCS (tightly coupled system) for three months continuously. By lessons learned from the original calculation, and subsequent optimization techniques they developed, computation time has been cut in half, down to four to six weeks per computation.
The group is currently performing complimentary computations, investigating the effects that bio-jet fuel components might have on the efficiency, viability and emissions of commercial airline travel.
SciNet is Canada’s largest supercomputer center, providing Canadian researchers with the computational resources and expertise necessary to perform their research on scales not previously possible in Canada, from the biomedical sciences and aerospace engineering to astrophysics and climate science.
Clearing the Air: How SciNet is Helping Improve our Air Quality (Dworkin, Saffaripour & Thomson; UofT)
Meghdad Saffaripour, University of Toronto; “Distributed-Memory Parallel Computation of a Laminar Sooting Coflow Jet-A1 Diffusion Flame” (HPCS)