Research
We build the software infrastructure to hardwire UK Exascale computing directly to beamlines. Our work targets the stochastic precursors to material failure, replacing deterministic testing schedules with autonomous, AI-driven control loops that navigate the search space of degradation in real-time.
Self-Driving Microscopes
Stochastic failure events — dendrite initiation, crack propagation, phase decomposition — are rare, fast, and invisible to pre-programmed experiments. We are building autonomous platforms where reinforcement learning agents steer beamlines in real-time, using physics-based digital twins to predict and capture these events as they happen.
This requires solving several coupled problems: training Graph Neural Networks to distinguish benign aging from true failure precursors, building simulation engines fast enough for real-time inference, and designing the control architecture that connects prediction to instrument action. We defined the operational requirements for this in the BASE Scale, a taxonomy for deploying autonomous agents at large-scale facilities, and the practical framework in the ESME architecture.
Key papers: BASE Scale (arXiv, 2026) · Heuristic Operando Experimentation (arXiv, 2025)
Operando & Correlative Synchrotron Characterisation
We design and run multi-modal operando experiments at UK national facilities (Diamond Light Source, ISIS, Central Laser Facility) and international sources (Sirius, ESRF, ILL). Our methods combine X-ray computed tomography with diffraction, fluorescence, and scattering to capture both structure and chemistry simultaneously. While batteries are the primary application, these transferable methods are deployed across a range of scientific problems: planetary science, biomechanics, soil physics, and CO2 absorption for carbon capture.
This includes leading the world’s first in-situ dark-field X-ray microscopy experiments for battery research (ID03, ESRF) and developing new correlated imaging and diffraction methods at Diamond Light Source.
As PI, Le Houx leads the UK’s Battery Imaging BAG at Diamond Light Source and has run 9 flagship experiments as Principal Investigator. Recent highlights include the first observation of fluid inclusions in cryogenic natron — a mineral predicted on Saturn’s moon Enceladus — and coupled XRD-CT/XCT of soil mechanics during analogous root growth.
Key papers: Natron / Enceladus (Nature Portfolio, in review) · Operando XCT of All-Solid-State Batteries · Coupled X-ray Imaging/Diffraction in Soil Mechanics
Computational Imaging & Machine Learning
We develop open-source computational tools to extract more from 3D imaging data than conventional analysis allows:
- Super-resolution: Generative adversarial networks (EXG-GAN / OpenLSR-X) that decouple the resolution/field-of-view trade-off in X-ray tomography, enabling high-resolution imaging at reduced dose and acquisition time.
- Data fusion: Methods to combine multi-modal, multi-resolution datasets (XCT + XRD-CT) into unified 3D maps of structure and chemistry.
- Image-based simulation: OpenImpala, our open-source massively parallel solver (AMReX/C++) for computing effective transport properties directly from tomographic images, achieving 100x speedups over commercial solvers.
- Data-driven mechanics: D2IM-Strain (with Tozzi group, Greenwich), a deep learning framework that predicts strain fields directly from undeformed XCT images, bypassing numerical differentiation and its associated noise.
- Generative microstructures: Physics-constrained diffusion models (PorousDiff) for generating synthetic 3D electrode microstructures for surrogate modelling.
Key papers: Multimodal Synchrotron Data for ML (Scientific Data, 2025) · D2IM-Strain (bioRxiv, 2026) · OpenImpala (SoftwareX, 2021)
Quantum-AI for Scientific Imaging
An emerging direction exploring whether quantum computing can address the computational bottlenecks in high-dimensional imaging. We are investigating quantum-native representations of volumetric image data and hybrid quantum-classical architectures for segmentation, classification, and image correlation — including the conceptual development of Quantum Image Correlation (QuIC) as a future extension of digital volume correlation.
Software & Data Stack
Our code is open-source and developed for Tier-1 National Supercomputing facilities (e.g. Isambard-AI). All repositories are hosted under the BASE-Laboratory GitHub organisation.
| Repository | Domain | Description |
|---|---|---|
| OpenImpala | Physics | Massively parallel solver for transport physics (AMReX/C++). SoftwareX, 2021 |
| PorousDiff | GenAI | Conditional 3D diffusion for mechanical surrogate modelling. |
| OperandoCellLibrary | Hardware | Streamlit library of operando sample environments for battery research. |
| LiionDB | Data | Community standard database for Li-ion battery parameters. Progress in Energy, 2022 |
| OpenLSR-X | Vision | SRGAN implementation for synchrotron XCT super-resolution. |
| BatteryExplorer | Viz | 4D interactive dashboard for operando failure analysis. |
| MultimodalBenchmark | Data | Official code for the 3D multimodal synchrotron dataset. Sci. Data, 2025 |