Los Alamos National Laboratory releases first public dataset for training AI models on explosion and shock physics

Los Alamos National Laboratory published the HEAT (High Explosives and Affected Targets) dataset, a collection of 2D physics simulations showing how explosive shocks travel through different materials. This is the first publicly available training data for this type of simulation.

The dataset captures complex physics that normally requires expensive supercomputer simulations: how materials deform, change phase (solid to liquid), break apart, and interact when hit by explosive shocks. AI models trained on this data can predict these outcomes in seconds instead of hours, making early-stage design testing faster and cheaper.

Engineers and researchers in defense, mining, aerospace, and materials science can now train AI "surrogate models" (simplified AI versions of expensive physics simulations) without building their own datasets—cutting development time for safety analysis, weapon design, and accident prevention. Previously, each organization had to generate their own proprietary data.