Cover Story
Answering the Questions: ARL MSRC Provides Support to Columbia Space Shuttle Accident Investigation
By Brian J. Wagner
Just 16 minutes before it was scheduled to touch down, the Columbia space shuttle tore apart upon re-entry into the earth's atmosphere and claimed the lives of its seven crew members on Feb. 1.
The question that needed to be answered was, How could anything so tragic happen? At the time, it was speculated that the primary cause of the physical failure of the Columbia was that during takeoff insulating foam on the liquid fuel tank broke loose and breached the thermal protection on the left wing. Then, under the extreme conditions of re-entry, this breach allowed hot gases to burn through the protection on the wing, causing the destruction of the space shuttle.
 |
| Shuttle assembly colored by pressure at the time of ramp separation. |
Within two hours of the accident, NASA started its investigation. Seven months later, after rigorous investigation, the Columbia Accident Investigation Board (CAIB) confirmed the earlier speculation: "The physical cause of the loss of Columbia and its crew was a breach in the Thermal Protection System on the leading edge of the left wing, caused by a piece of insulating foam which separated from the left bipod ramp section of the External Tank at 81.7 seconds after launch, and struck the wing in the vicinity of the lower half of Reinforced Carbon-Carbon [RCC] panel number 8."
Even though the hundreds of scientists and engineers were not always in agreement, the enormous amount of data gathered from all fi ve areas of review - aerodynamic, thermal dynamic, sensor data timeline, debris reconstruction, and imaging evidence - leaves little doubt that the board's conclusions follow the preponderance of evidence so painstakingly gathered by these researchers and thousands of individuals who collected debris.
Dr. Robert L. Meakin is an Army Civil Servant with the US Army Aeroflightdynamics Directorate (AMCOM) at the NASA Ames Research Center in Moffett Field, Calif.
He works at Ames under an Army-NASA joint agreement. In this instance, this relationship leads to the development of software that is of mutual benefi t to both agencies - the Army's interest being largely focused on prediction tools for rotary-winged aircraft. Dr. Meakin has been working in the area of unsteady aerodynamics for problems involving relative motion between component parts since 1987. He developed many of the tools he now uses when he was a member of a Space Shuttle computational fluid dynamics (CFD) team, comprised of NASA Ames Research Center and NASA Johnson Space Center (JSC) personnel in the wake of the Challenger disaster.
So it was quite logical when, only one day after the Columbia accident, Ray Gomez of the NASA JSC contacted a small group of scientists, including Dr. Meakin, at NASA Ames and asked for CFD support for an investigation of STS-107 ascent aerodynamics. In addition to Dr. Meakin, the group at NASA Ames consisted of Drs. Michael Aftosmis, Stuart Rogers, and William Chan, all NASA Civil Servants in code INA, and Dr. Scott Murman, a NASA contractor with Eloret, Inc.
Dr. Meakin graciously allowed me to interview him on the role that he and this group played in the investigation. The group, part of the larger transport analysis team mentioned in the CAIB Final Report, Vol. II, Appendix D, Part 8, was asked to support the analysis of the Columbia STS-107 during ascent/takeoff at the time that the piece of foam broke away from the left bipod ramp. To do this the group provided high-end computational dynamic support, using "high-fidelity, numerical simulation techniques to get an understanding of the aerodynamic fields during the incident, to compute the speed of the foam debris at impact, and to gain insight into the possible mechanics of debris separation."
Some questions that members of this group sought to answer were what the likely impact velocity of the debris might have been and its size and mass properties. They did not have much to go on because there was very little information on the debris. JSC engineers provided geometric definitions for the pieces of debris that ranged from the size of the entire bipod ramp to smaller pieces based on earlier flights and best estimates. They also provided the flight conditions of Columbia at the time that the debris broke away.
Using this information, the CFD group at Ames evaluated a variety of mass properties and made its simulations using the codes CART-3D and OVERFLOW-D, the latter a product originally developed through DoD HPC Modernization Office support via a project called CHSSI CFD-4. The code uses overset structured grids to accommodate arbitrarily complex geometries that, on a component-wise level, retains the computational advantages inherent to structured data.
 |
| Impact with the leading edge of the lifter, a silvery blue tube has been superimposed to help visualize the patch of the insulating foam. |
The Ames CFD group began by employing the existing JSC grid system for the integrated Space Shuttle launch vehicle - the Orbiter, the external tank, the solid rocket boosters, and all of the attached hardware.
Dr. Meakin explained how they proceeded. "We replaced the left bipod ramp, just removed what was there and then regenerated grid systems for the piece of debris and the remaining ramp so that, initially, they would be in coupled position, so it would look like the entire ramp."
Then they carried out the high-fidelity simulations of the foam debris incident by allowing the debris to break away and move relative to the launch vehicle. This procedure resulted in time-accurate simulations of the event, providing "the debris position, speed, and orientation as a function of distance down stream and also a function of time."
The group encountered significant problems in its analysis because initial conditions for the incident were not known, and the debris trajectory is highly dependent on the initial conditions. Thus, they did not know what the piece of foam that broke off looked like or its mass properties.
The CAIB confirmed how difficult their task was when it stated in the final report that the nature of the foam's structure makes it "extremely difficult to model analytically or to characterize physically. The great variability in its properties makes for difficulty in predicting its response in even relatively static conditions, much less during the launch and ascent of the shuttle."
So, as Dr. Meakin put it, they dealt with this problem by using "a shotgun approach." They ran a lot of cases on five different sized pieces of foam with a range of mass densities, bracketing "the maximum and minimum that ... [the pieces] could have weighed and then a range of separation conditions and evaluated the most likely combinations that led to the strike on the left wing."
Another problem was the sheer size of the computational task at hand. The data that had to be gathered, stored, and moved about was very great. An immense amount of computer time also had to be obtained. Each simulation required thousands of processor hours and about 30 GBs of disk space for solution storage. Dr. Meakin alone ran some 50 simulations. Other members of the group ran nearly 400 additional simulations. So, even storage and post-process analysis and handling of the data became a problem.
To address this problem, Dr. Meakin first thought of requesting machines from NASA, but he didn't have a current account and wanted to begin immediately. Because he had been doing all of his research at DoD HPC Major Shared Resource Centers and had a current account with the ARL MSRC, he thought that he might save time by asking to use some of the MSRC's computer resources. So he called Larry Davis at the HPC Modernization Office, explained the situation, and asked for computer resources. He also called Charles J. Nietubicz, director of ARL MSRC.
"Within a day, I had 40,000 hours of time at my disposal, a dedicated queue on Zornig (Origin 3000) and Brainerd (IBM SP) with 32 processors available on each machine for my exclusive use, and was able to start immediately."
So the ARL MSRC got his project off to a quick start. Then he completed the simulations using an additional 150,000 hours of processor time on NASA's machines.
There were some difficulties with the initial porting of Dr. Meakin's code to the MSRC. To resolve these, he called on Tom Kendall, government chief engineer, and Phil Matthews and Jeffrey Robinson of the ARL MSRC Project Team. Dr. Meakin said that because the investigation was so urgent, these specialists "were really helpful. They fielded phone calls in the middle of the night, weekends, and holidays to keep the systems running and working."
The MSRC also saw to it that Dr. Meakin received the huge storage space necessary for his solutions.
Dr. Meakin ran five simulations at the ARL MSRC. Given the urgency of the investigation and the continuing need for more simulations to be carried out on the NASA HPC systems, the flow visualization and simulation analysis - necessary to allow easy communication of the significance of the simulations - became a very big challenge. So, Dr. Meakin asked Raytheon's Mark Bolstad, Scientific Visualization Team, for help.
Mark ran some of Dr. Meakin's solutions through an application called EnSight to generate the geometry that was then read by Houdini, a digital content creation tool that converted those geometries into multiple-angle, multiple-light-source, 3D, animated videos. (Figures in this article show some of the images that Mark made from the solutions.)
His animated presentation reveals the most likely trajectory of the piece of foam as it breaks away from the left bipod ramp, traces a trajectory to the impact area, and hits the wing. He can also show all of this in slow motion and with various colors demarcating such things as stress, flow, and thermal conditions. These are factors that the code numbers alone do not show.
Dr. Meakin said, the animations "were helpful to me in confirming that things were carried out nominally in the simulations."
They were likely helpful to NASA JSC, too because, as Dr. Meakin says, "I do know that the JSC folks looked at them and used them to advantage ... I believe that they were helpful."
The primary data provided by the CFD simulations to NASA JSC included debris trajectories as functions of debris mass, shape, and mechanism of separation; debris velocity at impact; debris rotational velocity at impact; and animations. This data was part of the data used by NASA JSC to establish the range of conditions for ground-based testing to assess the likelihood of foam debris damage to Columbia.
The CAIB contracted with the Southwest Research Institute to make impact tests on Shuttle wings because of its extensive test and analysis experience with ballistic impact modeling methods, including work on the Orbiter's Thermal Protection system. For the tests, Southwest Research Institute redesigned a nitrogen-powered gun to accommodate foam samples that would be fired at previous shuttle wings to determine the damage that might result from such impact.
The CFD simulations and animations, along with data gathered by many other researchers, helped to determine the best size and density of the foam used and the velocity, rotation, and trajectory of the debris used in the impact tests. These tests confirmed that "foam can breach the RCC" given the "best-estimate" properties used in some of the tests.
The CFD simulations represented a significant contribution to the CAIB's attempt to determine how the foam broke away and whether it was capable of starting a chain reaction that led to the destruction of the Columbia. In addition to providing valuable data for the design of the impact tests, however, the CFD simulations also provided some insight into the actual mechanism of debris breakaway from the bipod ramp and confirmed that, "... embedded pressure sensors in the Orbiter recorded the signatures of the debris as it passed by."
All of this is in a report called STS-107 Transport Final Report in Support of CAIB, August 2003, a subset of which is in the Final Report of the CAIB.
The CAIB, taking into consideration the aerodynamic data produced by CFD simulations and the Southwest Research Institute's impact tests, along with the data from the other areas (thermal dynamic, sensor data timeline, and debris reconstruction), arrived at the conclusion seen earlier in this article that foam separated from the external tank's bipod ramp, hit the left wing in the vicinity of RCC panel 8, making a breach that allowed hot gases during re-entry to melt supporting structures of the wing.
Thus, the Columbia accident investigation represents the careful work and countless hours of numerous scientists and engineers. Among these was the CFD group at NASA Ames Research Center, including the Army's Dr. Meakin, supported by the HPC and the dedicated people at the ARL MSRC and countless other Americans who contributed in very important ways to establish the cause of this tragic accident.