In August 2005, in a remote part of Spanish Fork Canyon, Utah, a tractor-trailer overturned. A fire started in the engine compartment and spread to the load – 8,400 cylindrical devices containing the mining explosive pentolite. The blast injured some people nearby but killed no one. To the surprise even of explosives experts, the explosion left a crater 24 meters wide and 10 meters deep.
The eruption was orders of magnitude more powerful than anyone would have expected. The current understanding of explosions holds that the event should have been a case of deflagration, which would have limited damage.
In deflagration, burning material heats and ignites adjacent material; fire progresses through a flammable or explosive substance at an even pace. In slow-burning materials such as wood and wax, deflagration gives the steady flame of a campfire or a candle. With faster-burning materials, such as pentolite, the material burns so fast that it creates an explosion – a mass of rapidly expanding gases.
But the Utah accident didn’t stop with deflagration. The size of the crater showed it progressed to a far more violent type of explosion: detonation, in which a pressure wave forms inside the fire and expands throughout the material faster than the speed of sound. The high pressure of this shock wave can ignite nearly all of the explosive material in milliseconds, before the thermal explosion can throw that material outward and away from the ignition site. This more violent type of explosion is harder to establish when flammable gases are the primary fuel, yet detonation is suspected in several disastrous industrial accidents and coal mine explosions.
This phenomenon, deflagration-to-detonation transition (DDT), transforms limited explosions into catastrophic ones, with detonation releasing about 100,000 times more energy than deflagration. Yet much of the process remains a mystery.
Using supercomputers at national laboratories, two research groups are making major contributions to the understanding of DDT.
Martin Berzins of the University of Utah in Salt Lake City and his colleagues simulated the Spanish Fork Canyon explosion using the Oak Ridge Leadership Computing Facility’s Titan, a Cray XK7, and Mira, an IBM Blue Gene/Q at the Argonne Leadership Computing Facility.
“It took five and a half years to actually model the detonation, to make it possible to model at a scale needed to see what was actually happening,” Berzins says. A major hurdle: ensuring the Uintah code could simulate the event’s full scale in three dimensions, a task requiring between 64,000 and 500,000 CPU cores.
Graduate student Jacqueline Beckvermit carried out the simulation. She and Berzins collaborated with University of Utah colleagues Andrew Bezdjian, Todd Harman, Justin Luitjens, Qingyu Meng and Joseph R. Peterson; Scott Bardenhagen of Mesomechanics LLC, Santa Fe, New Mexico; and Charles A. Wight of Weber State University in Ogden, Utah. The group used the Uintah Computational Framework, a simulation code originated and developed at C-SAFE, the Department of Energy’s Center for the Simulation of Accidental Fires and Explosions at the University of Utah.
The simulation recreated one-eighth of the tractor-trailer’s load of explosives. It revealed that the transition to detonation occurred as the deflagration compacted the explosive material, forming a high-density barrier that trapped the expanding gases.
“As each cylinder expands, they push outward and into each other,” Beckvermit says. “That leads to inertial confinement, a high-density barrier that’s not letting gases escape.” When the pressure reached the critical limit of 5.3 gigapascals, a shock wave formed and moved outward through the bulk of the explosives, detonating nearly the entire load.
Based on their simulations, the team has devised alternative ways of packaging that leave spaces between explosive devices. “That would allow the gases to escape and not build up to the pressures needed for detonation,” Beckvermit says. The group has run simulations of their suggested packaging schemes, which have not yet been tested experimentally.
Gases in motion
A separate research team, led by Alexei Khokhlov at the University of Chicago, is probing the causes of DDT in gas explosions. The problem is different from modeling solid explosives because, in a gas explosion, literally every part is in motion.
Khokhlov and his colleagues have a long history of creating and refining simulation codes modeling DDT in supernovae plasmas as well as earthly gases. Colleagues in this work have included Joanna Austin and Andrew Knisely at the University of Illinois at Urbana-Champaign, Charles Bacon at Argonne.
The collaborators have revealed many of the transition’s secrets. They have recreated processes that cannot be seen in experiments involving actual explosions. In particular, they have discovered the formation of detonation-triggering hot spots that arise when a pressure wave hits burning gas.
“Now the interest is in how the shock is born inside the flame,” Khokhlov says. “That is the question we’re trying to answer.”
The project recently received an award from the Department of Energy INCITE program (Innovative and Novel Computational Impact on Theory and Experiment) of 150 million processor hours on Mira. To run the simulations, Khokhlov, Austin and Marta Garcia at Argonne will use the high-speed combustion and detonation (HSCD) code, which the larger group developed. The team validated the code through an earlier INCITE award.
The group will focus on the most important flammable gases for industrial and public safety, beginning with the simplest case and building in complexity. First, they’ll take on hydrogen, large volumes of which have formed and exploded in nuclear reactor accidents. Second, they’ll study syngas, a mix of hydrogen and carbon monoxide used as a source of clean energy. Finally, they’ll focus on ethylene gases such as ethylene oxide. Ethylene is one of the most widely used hydrocarbons in industry.
The group has succeeded in simulating a transition to detonation. “But there are also discrepancies with the experiments,” Khokhlov says. One reason for disagreement between the model and experiments is that few people have done highly detailed gas-explosion experiments. “We can reproduce the process itself, but not the details of it. The simulation has to match the experiment.”
Khokhlov is confident that the group will make important contributions, now that high-performance computers such as Mira enable them to build models that include all the necessary physics.
“Right now the machines are at the level where you can start to do this modeling in earnest,” he says, adding that the simulations will keep improving as computing power increases. “The beauty of it is that it will become easier and easier over time.”