← Back to NASA Technology Projects
Completed TRL 2 (started at 2, targeting 3)
SONTRAC is designed to detect incident solar neutrons within an energy range that fill a current gap in the energization process of flare ion acceleration. SONTRAC tracks recoil protons (from neutron interactions) as they traverse the fiber bundle volume, which deposit ionization energy along their path.
Currently, the reconstruction involves determining the energy deposited and the direction (e.g., the momentum vectors) of the recoil protons. In many cases, there is significant ambiguity in how to best identify the tracks properly. Kinematics can be used to eliminate certain configurations, but the effort is fully manual and laborious. We will take advantage of PIML to dramatically simplify the reconstruction effort, resulting in a significant improvement in the number of neutron interaction events that can be reconstructed in an autonomous manner and thus improve the instrument efficiency. We will embed physics within the training of the model via sophisticated custom loss function terms to utilize established physical principles and derived formulas. Using PIML also allows for enhanced generalization to unseen scenarios due to the embedded knowledge of physical phenomenon.
By simulating SONTRAC we enable the production of adequate amounts of data for training, and combining this with ML, which gives us novel insights, we can embed the insights within classical physics-based equations, we can usher in a new state-of-the-art (SOA).
We will also explore using PIML to improve the efficiency of the Wang-Sheely-Arge (WSA) model of the near-solar environment. WSA, currently built with empirical evidence, is used to predict space weather and is vital for assessing the impact of solar winds on satellite operations, communication systems, and astronaut safety. It is used worldwide and is currently standard and SOA in its field, but enhancements have not been made in many years.
The end goal of this effort is two-fold. The first end goal is to significantly improve the use cases targeted in this work, which are the SONTRAC instrument, which would be improved via enhanced autonomous neutron interaction event reconstruction, and the current WSA model, by utilizing PIML to achieve better accuracy and/or efficiency. The second end goal is to advance the field of PIML in Heliophysics, which would enhance many current efforts, such as (but not limited to) the Magnetospheric Multiscale Mission (MMS), Cluster mission, by automatically detecting and labeling plasma waves, or current Heliophysics models like ENLIL, by enhancing its accuracy and efficiency.
Our approach has three primary tasks: (1) training dataset generation, (2) loss function development, and (3) model training. To use our proposed ML models, we collect Geant4 simulator data into an ML-ready format. Then, we will process the individual neutron paths into a voxel representation resembling the physical construction of SONTRAC. Since we will have the compressed 2D readout from the simulated SONTRAC instrument as inputs and the true 3D paths through the instrument as labels, we can train an ML model, such as a physics-informed neural network (PINN), to directly predict these particle paths and collisions through generation of the 3D voxel representations. We will then be able to construct a loss function that ensures predicted paths do not violate the hard physical constraints that limit these particle interactions. This loss function will have multiple parameters, and their relative weighting will be a subject of investigation. We will then attempt to generate experimental data to prove and benchmark our ML model for SONTRAC.
We will do a similar process with WSA. First we will generate simulation data based on WSA, as well as gather any in situ data that might be available. Then, we will embed the physical and empirical aspects of the current WSA model with PIML, achieving this goal using the aforementioned primary tasks, to create a new and improved WSA model.
This process will be directly applicable to instruments and models beyond SONTRAC and WSA, and will provide a blueprint for others to implement their own physics-based strategies. The developed loss functions will be iterated on as part of a standard hyperparameter search completed during the training of out-path generation models. This results in model that accurately interprets SONTRAC data with a strong physics rationale, and predicts solar wind speed with greater accuracy and/or efficiency.
This mission will directly benefit the SONTRAC instrument, by enhancing its capabilities of tracking neutron incident tracks and energy deposits via protons.
This effort directly aligns and supports:
Listed on TechPort itself — the most direct way to ask about this specific project.
This is early/mid-stage (TRL 2) — the most realistic path in is NASA SBIR/STTR, which funds small businesses and research institutions to develop technology aligned with NASA's needs (equity-free, phased funding). Check whether a current SBIR/STTR solicitation topic overlaps with this project's technology area, or contact the project directly (above) to ask.
None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.