Speaker
Description
Muon tomography is commonly divided into absorption and scattering modalities. Absorption approaches are well suited for very large structures, while muon scattering tomography (MST) targets medium scale, heavily shielded objects such as dry storage casks and cargo containers. MST leverages multiple Coulomb scattering (MCS), where the distribution of angular deflections and lateral displacements depends strongly on material composition, enabling inference of changes in effective density and atomic number. A central limitation is the low cosmic muon flux, which often requires hours to days of data collection, especially when using simplified trajectory models such as straight-line path (SLP) or point of closest approach (PoCA) that effectively collapse MCS into a single interaction. This work presents μTRec, a physics-informed reconstruction algorithm that models the cumulative nature of MCS through a Bayesian framework. Using a Gaussian approximation to the bivariate scattering process (angle and displacement with covariance) and a constant average energy loss model, μTRec reconstructs a stepwise curved muon trajectory constrained by measured upstream and downstream tracks. In GEANT4 based studies of dry storage casks, μTRec achieves missing fuel assembly detection with approximately 20× fewer muons than PoCA for comparable signal-to-noise and contrast-to-noise ratios, substantially reducing required acquisition time and enabling near-real time imaging. We also present the first MST study for sealed microreactors, which are intended for remote deployment and long lifetimes (about 5 to 10 years) with limited physical access. Non-intrusive monitoring is therefore essential for safe operation, safeguards assurance, and maintaining public trust. Including muon momentum measurements improves μTRec detectability for missing fuel scenarios by about a factor of two. Moreover, Momentum-informed μTRec resolves key microreactor components and increases detectability by about 350%, with higher confidence than momentum-integrated PoCA. Finally, we demonstrate a Python based analysis pipeline for cargo security, applying machine learning based (unsupervised and one class) anomaly detection to identify concealed uranium from muon scattering derived features.