Quantification of Uranium holdup using Compton Imaging

The Generalized Geometry Holdup (GGH) method is the most commonly used method for estimating uranium holdup from basic point, line, and area sources. However, GGH can have errors upwards of ±50% depending on the distribution of material in process equipment. Traditional Compton methods create relative intensity images mapped onto 2D unit spheres or planar surfaces and fused with visual camera images. To improve localization and uncertainty quantification a 3D scene of the spatial and mass distribution of uranium holdup was created using Compton Imaging of the weighted ratios of the 186keV and 1001keV emissions. A PHDS Co. GeGI™ gamma imager with a 9cm diameter, 1.1cm thick, HPGe crystal with double-sided strip readout was used for data acquisition. Strip signal data in binary form was imported using Python into an efficient, portable, labeled, and indexed SQLite database and a table of all photoelectric and Compton interactions was generated. Because distance and distribution are necessary to estimate mass, Compton reconstruction is performed into a 3D cloud of point sources each with an energy bin structure. Point clouds can be generated randomly or initialized with a [ToF-]LiDAR scene to optimize reconstruction effort. Each point was weighted by its angular position in the angular uncertainty of each Compton cone, its distance to the detector, and the current mass estimate. Total mass is determined by integrating the spatial regions of holdup. For computational efficiency, C+OpenMP was used to project all Compton cones from every detector position into the 3D scene in parallel. The current Compton reconstruction performance for making 3D scenes of uranium holdup mass is 103.0 cones per second per 1e6, 120 energy bin, points on a mobile Intel Core i7-9750H. Initial results indicate that good statistics for the highest energy events improves the mass estimation image. Future development will improve mass quantification accuracy, improve computational efficiency by migrating to C+CUDA on a GPU, and run on a 20W embedded ARM64 Nvidia Jetson platform paired with the GeGI™.