Data Driven Materials Characterisation: Novel Synchrotron Experimental Strategies
Please login to view abstract download link
Virtual materials testing is becoming faster due to advancements in numerical methods, new model developments and the availability of open-source pipelines and computing resource. This increased speed allows virtual materials testing to be conducted within the typical timeframe of a synchrotron imaging experiment (3-5 days), enabling novel acquisition strategies. Here, we present three new experimental strategies making use of these developments: Automated stochastic image-based modelling pipeline [1]. We employ Python-based tomographic reconstruction, a U-Net classification network, and voxel-based simulation [2] within an open-source modelling pipeline. This combination provides real-time information on how representative elementary volume size and spatial resolution affect physical phenomena. We demonstrate this using the aqueous aluminium-ion battery, helping optimise experimental setup for optical magnification and detector settings. Super-resolution neural networks [3]. We use a Super-Resolution Generative Adversarial Network (SR-GAN) to up-sample a hierarchically structured, Zeolite 13X (Na86[(AlO2)86 (SiO2)106]• H2O) pellet by a factor of 4x. This network enables users to overcome detector volume-resolution physical limitations or reduce radiation exposure to beam-sensitive samples by a factor of 16. Multi-modal acquisition [4]. We use spatially correlated X-ray diffraction with X-ray computed tomography to resolve stress-strain fields in a soil system, adjacent to a plant root tip. Unexpectedly, the results show that stress appears to reduce at the highest penetration depths. By coupling these data with a digital twin solid mechanics model, we find that the measurements show the extent of the yielding inelastic zone, helping to explain the surprising result. References [1] Le Houx, James, et al. "Statistical effective diffusivity estimation in porous media using an integrated on-site imaging workflow for synchrotron users." Transport in Porous Media 150.1 (2023): 71-88. [2] Le Houx, James, and Denis Kramer. "Openimpala: open source image based parallisable linear algebra solver." SoftwareX 15 (2021): 100729. [3] Green, Calum, et al. “Three-Dimensional, Multimodal Synchrotron Data for Machine Learning Applications.” arXiv preprint arXiv:2409.07322 (2024). [4] Le Houx, James, et al. "Illuminating root-soil mechanics." bioRxiv (2024): 2024-10.