Speaker
Description
Traditionally, high precision muon scattering tomography has required expensive detectors that are hard to deploy and often highly sensitive to environmental conditions. Furthermore, muon tomographic inversion software optimised to handle large amounts of data and high spatial resolution models, on the order of tens-to-hundreds of seconds, are not readily available. Muodim holds the keys to these issues within both the hardware and software domains. Scintillators with centimetre-scale bars on each observation axis are cheap to produce, easy to deploy and maintain, and may provide sufficient resolution for many practical imaging scenarios. Furthermore, recent developments in machine learning make algorithms designed for GPUs a trackable, scalable and flexible solution to muography inversion problems. Muodim deployed two scintillator detectors aligned vertically, above and below multiple targets, both for controlled laboratory experiments and client projects. Results confirm that centimetre precision scintillating bars are sufficient for centimetre precision imaging of target objects. Our inverse problem framework is scalable to problems of ~1 million events with ~1 million voxels deployed on a 4Gb laptop GPU and is executed on the order of minutes. The model can be extended to process events with an “out of memory” model, allowing the treatment of larger datasets at the cost of speed. In the Muodim framework, the scattering problem based on the MLSD algorithm (Schultz, 2007) can be interchanged for a bespoke method developed to solve the absorption tomography problem with comparable performance characteristics.