Planning for a smooth landing on Mars

A U.S. mission to land astronauts on the floor of Mars can be not like every other extraterrestrial touchdown ever undertaken by NASA.

Though the space company has efficiently landed 9 robotic missions on Mars since its first floor missions in 1976 with the Viking Venture, safely bringing people to Mars would require new applied sciences for flight by means of the Martian environment. However these applied sciences and programs cannot be comprehensively examined on Earth beforehand.

Since 2019, a group of NASA scientists and their companions have been utilizing NASA’s FUN3D software program on supercomputers positioned on the Division of Power’s Oak Ridge Management Computing Facility, or OLCF, to conduct computational fluid dynamics, or CFD, simulations of a human-scale Mars lander. The OLCF is a DOE Workplace of Science person facility positioned at DOE’s Oak Ridge Nationwide Laboratory.

The group’s ongoing analysis mission is a primary step in figuring out easy methods to safely land a automobile with people onboard onto the floor of Mars.

“By its very nature, we do not have validation information for this. We are able to do worthwhile however restricted assessments in floor amenities like a wind tunnel or on a ballistic vary, however such approaches can not totally seize the physics that can be encountered on Mars. We will not flight-test within the precise Martian atmosphere—it is all or nothing once we get there. That is why supercomputing is so critically vital,” mentioned Eric Nielsen, a senior analysis scientist at NASA’s Langley Analysis Heart and principal investigator for the five-year effort on the OLCF.

In contrast to in current Mars missions, parachutes should not a part of the operation. As an alternative, the main candidate for touchdown people on Mars is retropropulsion—firing forward-facing rockets constructed into the craft’s warmth defend to decelerate.

“We have by no means flown something like this earlier than. The elemental query from the outset was, “Are we going to have the ability to safely management this automobile?'” Nielsen mentioned.

The explanation that NASA is investigating retropropulsion quite than standard parachutes is a matter of physics. Previous Mars landers have weighed about 1 ton; a automobile carrying astronauts and all their life-support programs will weigh 20 to 50 instances extra, or concerning the measurement of a two-story home. Mars’ skinny environment—about 100 instances much less dense than Earth’s—will not help a parachute touchdown for such a big craft.

“With a traditional automobile, we fly by means of a really clear, predictable atmosphere. All of that goes out the window with this idea, the place we can be touring by means of an especially dynamic atmosphere consisting of high-energy rocket exhaust,” mentioned NASA group member and CFD professional Gabriel Nastac.

With steerage from NASA mission planners, the group formulated a multiyear plan consisting of more and more subtle simulations aimed on the key query of controllability.

In 2019, the team conducted CFD simulations on the Summit supercomputer at resolutions as much as 10 billion components to characterize static automobile aerodynamics at anticipated throttle settings and flight speeds starting from Mach 2.5 all the way down to Mach 0.8, situations wherein the automobile’s rocket engines can be required for preliminary deceleration.

All through 2020, an intense code growth effort targeted on porting FUN3D’s normal reacting-gas capabilities to Summit’s graphics processing unit, or GPU, accelerators.

“Realizing environment friendly efficiency of an unstructured-grid CFD solver within the face of advanced physics-laden kernels is a gigantic problem in a GPU-based computing atmosphere. However we had been finally capable of restructure crucial segments of code to ship the efficiency we had been after,” mentioned NASA analysis pc scientist Aaron Walden, who leads the group’s multi-architecture software program growth.

The work set the stage for an vital 2021 marketing campaign that enabled the group to handle the advanced interactions of the liquid oxygen/methane rocket engines with the Martian environment, which consists of primarily carbon dioxide and nitrogen. A petabyte (equal to 1,000 terabytes) of output information for every simulation carried out utilizing 15,000–20,000 GPUs on Summit yielded key insights into crucial variations in automobile aerodynamics versus these noticed utilizing the prior simulation’s perfect-gas assumption.

For the 2022 marketing campaign, the group took a serious step ahead by incorporating the state-of-the-art NASA flight mechanics software program generally known as the Program to Optimize Simulated Trajectories II, or POST2, into the workflow. Shifting past simulations that assume a static flight situation, the group now sought to “fly” the automobile within the digital supercomputing atmosphere. This check would characterize a primary try to quantify and tackle crucial unsteady dynamics that may be encountered throughout an precise powered descent to the Martian floor.

The group enlisted key specialists from Georgia Tech’s Aerospace Techniques Design Laboratory; this group was led by Brad Robertson. These specialists had already spent a number of years growing a coupling algorithm to interchange the low-order aerodynamic fashions inside POST2 with real-time, physics-based FUN3D simulations to finally understand high-fidelity trajectory simulations that leverage subtle flight management algorithms.

“Coupling FUN3D and POST2 was fairly a problem. We needed to juggle 5 – 6 reference frames and the info transformations between them. However the reward was with the ability to undertake all of the laborious work carried out by different NASA engineers on detailed steerage, navigation, management and propulsion fashions and to convey all of them right into a single, unified, multiphysics simulation,” mentioned group member Zach Ernst, a Georgia Tech doctoral scholar on the time, who labored with NASA graduate intern Hayden Dean on the hassle.

Incorporating POST2 introduced a further problem. As a result of POST2 is topic to extra restrictive export-control laws than FUN3D, group member Kevin Jacobson was tasked with growing a distant coupling paradigm wherein POST2 would execute on a NASA facility whereas speaking in real-time with FUN3D working at management scale on the OLCF.

Establishing and sustaining this connection whereas accounting for firewalls, community interruptions and job schedulers offered quite a few challenges. This work required a few 12 months of planning and coordination with cybersecurity personnel and system directors at each amenities.

The extra effort paid off when the group achieved their long-term purpose of flying a considerable portion of the descent phase within the digital atmosphere.

The arrival of OLCF’s Frontier supercomputer couldn’t have come at a greater time for the mission. With exascale computing energy (a quintillion or extra calculations per second) now a actuality, the group might afford to reintroduce the specified bodily modeling and different classes realized over the lifetime of the mission.

In 2023, the group targeted on the final word simulation they’d hoped for years earlier: a really autonomous, closed-loop check flight leveraging the world’s strongest supercomputing system.

Whereas the eight predominant engines are used to manage pitch (up-and-down rotation) and yaw (side-to-side rotation) because the steerage system goals for the designated touchdown zone, POST2 additionally points instructions to instruct FUN3D to periodically fireplace 4 response management system, or RCS, modules organized circumferentially across the bottom of the lander to carry out roll corrections in flight.

“These capabilities can be crucial for assessing the controllability of future autos,” mentioned Georgia Tech’s Alex Hickey, who led growth of the RCS modeling.

The group’s long-term purpose grew to become a actuality in late 2023, as OLCF workers assisted in coordinating a cautious sequence of high-priority jobs over a two-week interval at scale on Frontier.

“For the primary time, we had been capable of return to the unique query of safely controlling one of these automobile in autonomous flight,” Nielsen mentioned. “In a typical aerospace CFD simulation, one would possibly compute a second or two of bodily time. Right here, Frontier enabled us to efficiently fly 35 seconds of managed flight, descending from 8 kilometers (about 5 miles) altitude to about 1 kilometer (0.6 miles) because the automobile approached its touchdown phase.

“The decision, bodily modeling and temporal length are past something we might try on a traditional high-performance computing system,” Nielsen added “The sheer pace of the GPUs applied at management scale is actually enabling, and we’re deeply grateful for the various alternatives and world-class experience that the OLCF has offered.”