HPC simulation produces explosive results
In 2005, a truck carrying 35,000 pounds of explosives crashed in Spanish Fork Canyon in Utah. What made this a special incident was that the explosives, although volatile, should not have exploded. But the blast left a 30 by 70 foot crater on the Utah highway. The process, called deflagration-to-detonation transition (DDT, should not have happened in the case of these explosives.
Led by Professor Martin Berzins, the research team is performing large-scale 3D simulations on Mira, the ALCF’s 10-petaflop IBM Blue Gene/Q system, to study the physical mechanisms that led to the explosion.
Berzins said: ‘Ultimately, we hope our research will result in strategies that can prevent accidents like the one we are studying.’
In the case of an accidental fire, the explosive cylinders were supposed to burn rapidly in a type of combustion called deflagration. Limited by heat transfer, deflagration events spread at a velocity lower than the speed of sound. Detonation, on the other hand, occurs when the combustion spreads at a supersonic rate and triggers a high-pressure shock wave.
The simulations were a challenge due to the complex nature of DDT, which involves several strongly correlated processes, such as chemical kinetics, pressure waves and turbulence, all occurring in multiple spatial and temporal scales.
To achieve the high spatial and temporal resolution needed to carry out the multiscale, multiphysics simulations, the University of Utah research team optimised their highly scalable Uintah Computational Framework to take advantage of Mira's petascale performance. This involved leveraging the massively parallel capabilities of Uintah to develop a robust reaction model capable of simulating the different modes of combustion at large scales.
Jacqueline Beckvermit, a PhD student and research assistant at the University of Utah, said: ‘The main focus of our simulations is to determine why the fire escalated to detonation.’
‘We set out to simulate one-eighth of the actual semi-truck with the explosives in their original packing configuration, but it was not an easy feat’ Continued Beckvermit. ‘After two years of work and more than 100 million computing hours, we finally reached detonation this fall.’
The researchers recently achieved detonation with 3D simulations that ran on a significant portion of Mira’s 786,342 cores. The team now plan to scale Uintah to run on the entire supercomputer as a means to achieve even higher fidelity results in the future.
John Schmidt, co-principal investigator and adjunct assistant professor at University of Utah said: ‘We’re not quite to the point of doing production-level runs on the full machine, but we have completed proof-of-concept calculations to show that we can scale and take advantage of virtually all of Mira’s cores.’
Thus far, the team’s simulations have led them to two possible mechanisms for DDT in large arrays of explosives. One hypothesis points to inertial confinement, a process in which damaged cylinders of explosives form a barrier that traps product gases, creating a pocket of high pressure that could initiate DDT. The other proposed mechanism involves a shock-to-detonation transition caused by the impact of explosive cylinders colliding in a high-pressure environment. But more analysis is required before these theories can be confirmed or rejected.
In addition to investigating why DDT occurred, the researchers are using Mira to examine how different packing densities and configurations could be used to prevent such explosions.
‘The ultimate goal of our project is to propose ideas on how to package explosives for transport to make sure accidents like this don’t happen anymore,’ Beckvermit said.
This work coincides with an ongoing National Science Foundation PetaApps project aimed at using simulation science to explore ways to use the Uintah Computational Framework to solve problems in public health and public safety.
The project received computing time at the ALCF through the DOE Office of Science’s Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program.