The samples sat in a specially controlled environment that independently adjusts temperature, pressure and gas mixture, recreating realistic, evolving reentry conditions. The team heated samples to 1,652 degrees Fahrenheit — the upper end of the range where these materials begin to decompose — and imaged them at the micrometer scale at multiple time points.

They focused on two commercial ablators used in different parts of spacecraft backshells and with different chemical compositions:

• SLA-561V, which contains cork — the same natural material used in wine bottle stoppers — as a structural filler
• SLA-220, which has no organic filler and relies on a rubber-like silicone matrix

The scans tracked how reentry-like heating drives real-time, multiphase chemical decomposition and changing porosity inside both materials. These direct measurements feed predictive models, reduce uncertainty about heat shield performance, and improve mission planning and crew safety.

Where does AI come in?

The researchers faced a trade-off. Fast scans of large, representative sample areas came at low resolution; high-resolution images captured fine structural detail but only of static samples before and after heating. The team resolved this with an AI-based super-resolution method built on generative adversarial networks — machine learning systems that generate realistic images from training data.

The bright, broad-spectrum X-ray beam from the light source let the team quickly capture large-volume, lower-resolution scans at short intervals, tracking fast structural changes during heating. A single-wavelength beam then produced crisp, high-resolution snapshots of the samples before and after heating. By training the AI on those before-and-after images, the researchers enhanced the entire real-time image sequence into a complete high-resolution record of each material's transformation.

"We can perform 3D imaging of samples under different extreme conditions such as heat, cold, pressure, and tension, and we can watch the internal structure of materials evolve and give a deep look into internal structure in high detail as their properties change under these conditions," said ALS scientist Liz Clark.

What did the images show?

The AI-enhanced imaging revealed a striking difference between the two materials. When heated, the cork in SLA-561V chemically breaks down and disappears, leaving open, empty pockets distributed throughout the material. SLA-220 responds differently: its silicone matrix forms a dense, branching network of interconnected channels.

These are not just visual quirks. Open, isolated pockets behave differently from an interconnected channel network when heat and gas move through a material under reentry conditions — and those differences affect how each material performs as a heat shield.

"This is a great example of how years of working together to customize the imaging technique and integrate AI are helping us generate high-quality data and facilitate analysis, leading to more detailed scientific insights in a fraction of the time," Clark said.

What changes for future missions?

After the first Artemis mission, when heat shields did not perform as NASA's computational methods predicted, the agency used the ALS to examine shield materials and better understand how their internal structure evolves over time, Clark noted.

The new findings give engineers something they have not had before: direct microscopic measurements of heat shield materials changing under thermal conditions similar to atmospheric reentry. The results remain tied to laboratory simulations of reentry rather than flight tests, so real-world validation will continue. Still, the measurements bolster past observational studies, sharpen model development, and help missions return safely home.