CLOUD

By J. William Bell, NCSA Senior Science Writer -- NCSA and Alliance researchers provide Shell Oil Company with code and expertise for modeling refining and production equipment and in less than a week set up the first Linux cluster in Shell's oil products and chemicals research areas. Vacuum flashers -- they almost sound made up. Something a movie fighter pilot barks about being damaged in a dogfight. Something that allows a DeLorean to travel through time. Outside the movies, however, vacuum flashers are a critical, elementary part of many applications in the petroleum and chemical industry. These devices are nothing more than vessels in which varying atmospheric pressure allows products to be distilled from another substance. Vacuum distillation prevents decomposition at the point of vaporization, and the separation into components yields useful end products or elements that can be used in other processes. Regardless of the initial substance and the eventual products, vacuum flashers are home to a wicked fluid dynamics environment. Liquids and gases flowing among one another trigger dramatic fluctuations in pressure and velocity -- fluctuations we know as turbulence. Understanding this turbulence, its impact, and methods of combating it is not only a great research challenge. It will also impact the industrial bottom line by both reducing the deterioration of vacuum flashers and increasing the units' efficiency. In an effort to come to terms with turbulence, NCSA recently completed an extensive project with Shell Oil Company. Working under the auspices of NCSA's Private Sector Program, a research team modeled a vacuum flasher that Shell uses in chemical manufacturing. The team used an NCSA massively parallel computational fluid dynamics code known as GenIDLEST to capture with remarkable precision both the complex geometry of the vessel and the time-dependent characteristics of the turbulent flows. While they were at it, the team set up a Linux computing cluster dedicated to Shell computational fluid dynamics research. With standardized software that is now part of the Alliance's Cluster-in-a-Box software package and off-the-shelf computers, the cluster was up and running in a matter of days. "We spend a lot of money maintaining commercial tools for our research and technical service projects," says Raghu Menon, a Shell researcher who has collaborated with NCSA for years and is Shell's principal investigator on the application. "But problems like modeling time-dependent turbulence accurately and solutions like the new Linux cluster require relationships like that with NCSA. This relationship is very exciting. The power is apparent." ***A solution and a mystery*** Overhead lines carry vapors away from vacuum flashers. Researchers believe that the flow in these overhead lines cause structural vibrations seen in the units. These vibrations, if unmitigated, can cause fatigue failures in the vacuum flashers' nozzles and associated piping. Much of this vibration can be overcome with what is aptly called a vibration reducer. "It's just a pipe with holes in it that provides resistance to the flow in the flasher and dampens the resonance between the fluid and the flasher at the right frequency," says Danesh Tafti. Tafti was a research scientist at NCSA for 10 years before taking a faculty position in the mechanical engineering department at Virginia Tech in 2002. He continues research for the Alliance, developing GenIDLEST for the academic and industrial community at large. The team from NCSA and Shell used their model to isolate the most energetic modes in the flow. Then the team members could see the vibration reducer in action, suppressing these modes and thus allaying vibrations. They also found that the vibration reducer significantly affects the variation in pressure at points in the flow. The difference in pressure between two such points is known as a pressure drop. "The same phenomenon responsible for the vibrations also increases the pressure drop through the vacuum flasher. This increased pressure drop is detrimental to the process. It limits the capacity and requires additional energy input, which increases the company's energy costs. The vibration reducer, in addition to suppressing vibrations, decreases the pressure drop and significantly improves the process economics," Menon says. The vibration reducer also introduces a mystery. Researchers aren't sure how the simple little gadget works. "At this point, we don't know the exact mechanism, we only know the effects," Tafti says. "It's isolating the exact mechanism that we'd like to continue to pursue." ***Also RANS*** Until their work with NCSA, Shell relied on a common computational fluid dynamics method known as RANS or Reynolds Averaged Navier-Stokes. This approach has its benefits. It's cheap computationally, and it gives researchers an overall understanding of the behavior of the flow rather quickly. It does not, however, capture the wildly random, time-dependent nature of large-eddy, or very turbulent, flows like those found in a vacuum flasher. GenIDLEST, created by Tafti while NCSA, is designed for just such flows. Instead of time-averaging the flow and treating it as steady, GenIDLEST solves the time-dependent Navier-Stokes equations -- which are the Rosetta Stone for fluid dynamics, the base equations that make possible understanding and modeling a system -- and describes the flow at every moment in all its complexity. To do this work at the high flow velocities encountered in the vacuum flasher, researchers at Shell and NCSA had to use a fine-grained mesh that demanded that the calculations be run at about 1.6 million points throughout the model vacuum flasher's complex shape. This arrangement was demanding -- the initial run required about 10,000 hours on NCSA's SGI Origin2000 supercomputer -- but it showed the team many of the aspects of the flow with more precision than ever before. "We agreed that this had to be a time-dependent simulation, and that our standard methods wouldn't meet the requirements. It was well beyond our current capabilities and what we do typically," Menon says. ***Made to order in less than a week*** The model has already shown researchers many of the flow's characteristics -- that it is marked by large velocity and pressure fluctuations with a dominant frequency and that introducing a vibration reducer significantly calms that dominant vibration. Nonetheless, Shell researchers would like to know a lot more. To that end, Jeremy Enos, part of the NCSA team, helped install an eight-processor Linux computing cluster at a Shell research facility in Houston in January 2001. The Shell team plans to expand the cluster and install the Alliance's GenIDLEST software in early 2002. Getting the cluster up and running in less than a week was possible only because of OSCAR, the Open Source Cluster Application Resources. This suite of tools and software has everything needed to install, maintain, and use a modest-sized Linux cluster. OSCAR is now a part of the Alliance's larger Cluster-in-a-Box software package. "Danesh and Raghu have been talking about this for years," says Sundaresan Bala, who manages Shell's relationship with NCSA. "When the time came to update our in-house computational fluid dynamics capability, they saw an opportunity, and they took it. Now we have a cheap, easily installed system and the room to grow. With this system and our other eight or so clusters in other groups at Shell, PC Linux clusters are an important part of our high-performance computing strategy." Relevant URLs: --Access Story: http://access.ncsa.uiuc.edu/Stories/genidlest/ --NCSA Cluster-in-a-Box Website: http://www.ncsa.uiuc.edu/TechFocus/Deployment/CiB/