AINEWS
A 900°C chamber at Berkeley Lab pairs micro CT (high resolution X ray 3D scanning) with AI reconstruction, closing the simulation gap that surprised NASA's uncrewed Artemis I Moon test in 2022.
A spacecraft coming back from the Moon hits the top of Earth's atmosphere at roughly 1,650°C, hot enough to keep steel liquid. For sixty years, the only way engineers could see what that heat does to a heat shield was to look at the ablative material before launch and after landing. The middle stayed dark.
A new 900°C chamber at Lawrence Berkeley National Laboratory's Advanced Light Source, paired with X-ray micro-computed tomography and AI-assisted reconstruction, lets researchers watch that middle happen for the first time. The work, published in npj Materials Degradation as "Super-resolved microstructure of pyrolyzing superlight ablators," is being framed as the diagnostic NASA did not have when its uncrewed Artemis I test flight came home in 2022 with charring the ground simulations never predicted.
Artemis I was the uncrewed 2022 test of the rocket and capsule NASA wants to fly astronauts on for the next crewed lunar return. The Orion capsule flew past the Moon, came back, and landed. Engineers opened the heat shield and found uneven loss of ablative material, the kind of pattern that means the protective layer did not burn off the way the models said it would.
The standard workflow before that flight, used through Mercury, Apollo, and the Space Shuttle, was to test ablators in arc-jet facilities and high-heat ovens, then fly them and inspect the recovered hardware. That workflow is enough to know a heat shield works. It is not enough to predict how the char layer peels, where it cracks first, or which microstructural feature decides whether a region survives reentry or fails early.
The Advanced Light Source is a Department of Energy synchrotron, a kind of particle accelerator that produces X-ray beams roughly a billion times brighter than a hospital scanner. Inside it, a specialized chamber heats a small sample of ablator material to temperatures up to about 900°C. That is the onset of heat-shield breakdown, the moment the material's resin begins to break down thermally into char and gas.
While the sample heats, the X-ray beam sweeps through it and the detector records a micro-CT volume, a three-dimensional scan at micron resolution that shows the pores, fibers, and resin inside the material. Two honest limits: real reentry peaks around 1,650°C, and the chamber currently reproduces heat without air velocity. The paper is explicit that the conditions are deliberately sub-flight rather than a full reentry. What the chamber does deliver is the part of the failure that ground testing has handled worst: the microstructure evolving as the ablator loses mass.
"AI" in the press coverage is easy to read as a new model architecture. The honest description is narrower and more useful. The micro-CT volumes come back noisy, because the samples are hot, sometimes outgassing, and the scan windows are short. The team trained a reconstruction and segmentation model to clean those volumes up and to separate the solid char from the pores and cracks as they form. The AI is a noise-floor and labeling problem, not a discovery engine.
That distinction changes what the result means. The imaging is the science: the first time-resolved, three-dimensional view of an ablating heat shield's interior. The machine learning makes that view legible at the frame rate and resolution engineers need.
Engineers can now watch the inside of an ablator fail in real time, change the recipe, and watch it again. The method applies to the Orion heat shield, to Mars-class entry vehicles, and to the longer-term question of whether any part of a thermal protection system can be made reusable without losing the safety margin of an expendable layer.
The ceiling is concrete. The chamber does not yet reproduce the full 1,650°C peak, the air velocity, or the mechanical loading of a real reentry. It reproduces the part of the failure that was a black box, at a fidelity that lets a materials team iterate before flight rather than after.
For the next crewed lunar return, that is a quieter and more useful kind of progress than a bigger rocket. It is the instrument an engineer can use.