Multi-modal radiographic imaging and tomography (MM-RadIT) through data fusion and deep neural networks*

Deep neural networks (DNNs) offer a generic platform for data fusion (DF) [1], which includes multi-instrument data fusion (MIDF), multi-experiment data fusion (MXDF), and simulation-experiment data fusion (SXDF). These features make DNNs attractive to nuclear fusion power plant applications, as well as multimodal (MM) radiographic imaging and tomography (RadIT) for non-destructive testing and plasma diagnostics, leveraging accelerated workflows through machine learning and artificial intelligence (AI). Here we first highlight recent advances in neutron-X-ray (NeuX) instrumentation and analysis [2-5]. We then summarize several possible new directions in MM-RadIT using the AI-enhanced framework of Physics-informed Meta-instrument for eXperiments (PiMiX) [1] including 1.) Advancing ‘mini-RadIT’ and their applications, addressing the gaps in spatial resolution, temporal resolution, high-repetition-rate, high-confidence (probability) and high-fidelity three-dimensional (3D), 4D (time-dependent 3D), and 4D+ tomographic and hyperspace reconstructions; 2.) Integrated RadIT data and information science approach, or data-driven RadIT towards AI-RadIT, through the state-of-the-art modeling, imaging hardware optimization, innovative RadIT modalities, advanced data algorithms, and uncertainty quantification (UQ). AI-RadIT may automatically harvest more information from experiments for carbon-neutral energy (e.g. nuclear fusion energy), and plasma processing of materials; 3.) Integrated ‘mini-RadIT’ and ‘AI-RadIT’ for mAI-RadIT; and 4.) Time-resolved positron imaging and tomography, as an emerging modality of time-resolved antiparticle RadIT, enabled by ultra-short-pulse high-power lasers and compact accelerators. Los Alamos Report number LA-UR-24-25903.