|
|
Open Access
Abstract: Non-destructive evaluation (NDE) is used to detect and characterise defects in safety-critical components. This paper focuses on manual pulse-echo ultrasonic testing applied to the sizing of surface-breaking crack-like defects. In the field, practitioners often use A-scans from single element, pulse-echo ultrasonic testing. Machine learning, and in particular neural networks, have huge potential for applications to NDE-type problems due to their ability to recognise patterns from signals. Accurate predictions require large, labelled databases yet there is a paucity of such ultrasonic data for surface-breaking thermally fatigued cracks. To address this deficit, this work demonstrates that a database composed entirely of simulated, synthetic A-scans provides promising predictions on measured data. Crack height predictions with a mean absolute error of around 0.124 mm are achieved for measured data (on cracks between 0.5 mm and 4 mm). When tested on simulated A-scans, both height and tilt angle predictions are extremely accurate (to within 0.035 mm and 0.41°, respectively). To achieve such accuracy the in-situ inspection techniques were mimicked, with particular attention paid to the characteristics of the thermally fatigued crack species, and the single element transducer and its input signal. The details of the finite element simulations that generated the database are presented here. The work outlined in this paper shows that A-scans can be used to size defects using neural networks trained entirely on synthetic data, informed by experimental measurements, with potential applications for improving the efficiency of sizing cracks in-situ.
|
Oct 2026
|
|
|
|
Open Access
Abstract: With their high capacity and energy density, Li metal anodes have been intensively researched over the last few decades. Dendrite growth and electrolyte degradation, however, remain obstacles for their commercial implementation. Designing a stable solid-electrolyte interphase (SEI) is an important step in passivating the degradation reactions and mitigating dendrite growth. The advent of machine learning (ML) force fields has accelerated simulations with accuracy similar to quantum mechanical methods. Here we explore the predictions from a fine-tuned MACE-MP-0 foundation model for simulating the interface between Li metal anodes and carbonate-based liquid electrolytes. We find the fine-tuned MACE-MP-0 foundation model quantitatively agrees with DFT calculations, that the force field is remarkably stable, and that the decomposition products at the SEI fall within what is expected to be possible. We find that bare Li reacts with the liquid electrolyte giving typical electrolyte degradation products. We then study Li metal anodes passivated with model inorganic SEI components and find that the presence of these coatings significantly reduces electrolyte decomposition. While certain coatings are more stable and passivate further surface reactions, other coatings allow electrolyte reduction to proceed. In general, (100) coatings appeared more stable than (111) coatings. LiF (100) is found to be the most structurally and chemically stable coating even at high temperatures, fully suppressing electrolyte reduction reactions. The influence of the coatings on interfacial stability is further studied with the ONETEP program for large scale electronic structure calculations. Results show that electrolyte reduction occurs via electron transfer from the surface of coatings. We also quantify the Li self-diffusion coefficient in the metal, coating and electrolyte from the fine-tuned MACE-MP-0 foundation model for the interface simulations. It is found that the more stable coatings also suppress Li diffusion. Designing a stable SEI which is electronically insulating and ionically conducting requires optimizing the trade-off between these various opposing factors.
|
Oct 2026
|
|
I24-Microfocus Macromolecular Crystallography
|
Abstract: Glycosphingolipids (GSLs) are essential components of biological membranes, with several functions in both normal physiology and disease pathology. Glycosphingolipids are synthesised by lipid glycosyltransferase enzymes through the sequential addition of sugar groups onto the growing glycan headgroup. Perturbations in GSL biosynthesis can cause severe neurological disease and are implicated in a range of cancers. B4GALNT1 is one such glycosphingolipid biosynthetic enzyme, where loss-of-function results in a complex form of hereditary spastic paraplegia, and its dysregulation is associated with cancer, particularly paediatric neuroblastoma. B4GALNT1 is a key enzyme of ganglioside biosynthesis, a sialylated subclass of GSLs. In this work we present the first experimental structure of a GSL biosynthetic enzyme, namely B4GALNT1. We demonstrate dynamic remodelling of the B4GALNT1 active site upon substrate binding and identify the functional consequence of disease associated mutations. We show that B4GALNT1 possesses hydrophobic surface exposed loops that flank the active site that are essential to B4GALNT1 activity, with mutation of hydrophobic residues to polar resulting in severely ablated activity. Molecular dynamics simulations offered mechanistic insight, providing evidence that these surface exposed loops can insert into lipid bilayers to facilitate binding of the lipid substrate into the active site. We extended this work through further molecular dynamics simulations of predicted structures for other ganglioside synthetic enzymes, that such a mechanism involving membrane insertion loops may be exploited by other GSL biosynthetic enzymes. To step more broadly into the pathological consequence of GSL biosynthesis we developed a suite of computational tools to facilitate gene expression informed metabolic simulations of the GSL biosynthetic network. These tools enable simple and rapid gene knockout simulations whilst also permitting patient or cell-specific simulations of GSL metabolism. Given disrupted GSL metabolism is associated with malignancy in several cancers and the potential mechanistic roles of GSLs in different tumour types, we sought to leverage these tools in a cancer setting. We focused on Lower Grade Glioma (LGG) due to existing evidence for GSL related malignant phenotypes in Glioblastoma Multiforme (GBM) coupled to transcriptional similarities in severe LGG and GBM. We utilised patient specific metabolic simulations to enrich GSL biosynthetic enzyme transcriptomic datasets which were then utilised for training an ensemble-based machine learning model. This approach yielded strong risk-stratification for LGG patients whilst also implicating several biological pathways as potential GSL-related drivers of severity. These included cell motility, division and Wnt signalling. Together this work establishes molecular mechanistic detail on the function of the B4GALNT1 ganglioside biosynthetic enzyme and extends to the in silico modelling of GSL metabolism which was leveraged through machine learning to model survival and make biological insights in LGG.
|
Oct 2026
|
|
I12-JEEP: Joint Engineering, Environmental and Processing
|
Abstract: High-pressure reverse osmosis concentrates (HPROC) generated from coal chemical zero-liquid-discharge processes accumulate nitrate in high-salinity matrices that deteriorate downstream salt recovery and nitrogen management. This necessitates selective nitrate conversion before evaporation-crystallization. Electrocatalytic reduction offers a promising route to convert nitrate into ammonia, yet electrocatalyst design remains largely empirical, complicating the development of manufacturable electrodes for practical applications. Herein we developed a literature-informed interpretable machine learning (ML) framework to screen the CuNi@FeO electrocatalyst toward highly selective nitrate-to-ammonia conversion under high-salinity conditions. The self-supporting CuNi@FeO was fabricated by spontaneous deposition on sacrificial iron foam and it could realize spatially and energetically nitrate activation, hydrogen supply, and hydrogen evolution reaction (HER) suppression, accounting for a Faradaic efficiency (FE) of 98.74% and an ammonia yield of 19.56 mg h−1 cm−2 at −1.0 V vs. reversible hydrogen electrode (RHE). Operando spectroscopy and theoretical calculations revealed a synergistic multisite pathway involving the sequential *NO3→*N→*NHx conversion. The membrane-electrode assembly reactor could sustain 200 mA cm−2 for over 100 h and ammonia FE of 91.46% at cell voltage of 2.5 V during treatment of real industrial wastewater concentrate. This work establishes a paradigm shift to ML-powered screening of electrocatalyst toward scalable nitrate valorization from industrial concentrates.
|
Sep 2026
|
|
|
|
Terence
Tan
,
Oliver J.
Clark
,
Matthew J.
Derry
,
Renaud
Duyme
,
Guilherme Abreu
Faria
,
Robert
Farla
,
Ralf
Flaig
,
Yogal Prasad
Ghimirey
,
Anushka
Ghosh
,
Miguel A.
Gomez-Gonzalez
,
Ellen L.
Heeley
,
Anna
Herlihy
,
Annette
Kleppe
,
Paul
Millar
,
Melanie
Nentwich
,
Josh
Pickston
,
Nick
Terrill
,
Armin
Wagner
,
Andrew C.
Walters
,
Matthew
Watson
,
Philippe
Rocca-Serra
,
Susanna-Assunta
Sansone
,
Stephen P.
Collins
Open Access
Abstract: Experiment proposals at synchrotron facilities serve as the primary gateway for instrument access. They currently lack the standardized and granular topic metadata necessary for tasks such as classification and review, and, broadly speaking, reuse. This paper defines and tests the feasibility of a real-time topic classification service for experiment proposals using an open-source machine-learning model and domain experts for the evaluation phase. We applied the OpenAlex topic classification model to 5384 experiment proposals and selected 209 of them to each be independently evaluated by three domain experts to assess the performance and utility of the model. Analysis of the evaluations reveals a general consensus among the reviewers regarding the model's predictions, with a Krippendorff's alpha of 0.572. We also find that 74.2% of the proposals had at least one topic that was unanimously deemed relevant, which suggests that the model performs well enough to be used in a live setting with real-time verification. However, we do not recommend using it in automated environments without human oversight, given the proposal-based precision score of 56.0%. By aligning the data infrastructure of photon and neutron facilities with the OpenAlex ecosystem, we also lay the groundwork for the eventual inclusion of proposals and experiment reports into OpenAlex, which is necessary for a complete record of a research activity.
|
Sep 2026
|
|
|
|
Open Access
Abstract: Equipping robots with a physical, brain-inspired computing unit is a key step towards achieving true autonomy in intelligent machines; however, the practical realization of such neuromorphic hardware remains a major challenge. Because both software- and hardware-based spiking neural network approaches demand substantial training and energy consumption, an architecturally simple neuromorphic platform is required to enable scalable and energy-efficient robotics. Here, we experimentally demonstrate robotic control driven by a neuromorphic hardware based on physical wave interactions. Using a water-based experimental system that provides direct visual access to wave-network dynamics, we establish a wave-based reservoir computing framework that achieves near-perfect accuracy in robotic vehicle obstacle recognition, followed by real-time autonomous motion with obstacle avoidance controlled entirely by this wave-based hardware. Through micromagnetic simulations, we extend this concept to electrically excited spin waves in a magnetic nanodevice operating at gigahertz frequencies, outlining an approach toward solid-state robotic chips based on wave computing. These results position wave-based computation as a scalable and energy-efficient alternative to conventional digital neuromorphic architectures for next-generation intelligent machines.
|
Sep 2026
|
|
|
|
Christopher D. O.
Cooper
,
Evgeny
Tankhilevich
,
Kim
Remans
,
Ahmed
El Marjou
,
Aled M.
Edwards
,
Andrew
Leach
,
Andrew
Quigley
,
Anja
Schutz
,
Anne
Zemella
,
Ario
De Marco
,
Åsa
Sivertsson
,
Barbara
Borgonovo
,
Ben
Kemp
,
Bjørn
Voldborg
,
Carlo
Carolis
,
Carter
Mitchell
,
Cecilia
Wickstrand
,
Dafydd R.
Owen
,
David
Drechsel
,
David
Jones
,
Diego A.
Oyarzún
,
Dominic
Esposito
,
Eric
Geertsma
,
Ewa
Krupinska
,
Frank
Bernhard
,
Frederico
Ferreira-Da-Silva
,
Hanna
Tegel
,
Ian
Hunt
,
Imre
Berger
,
James
Love
,
Jan
Dohnalek
,
Jelena
Thies
,
Jerome
Basquin
,
Jesse
Coker
,
Johannes
Buyel
,
Joop
Van Den Heuvel
,
Jurgen
Haustraete
,
Katharina L.
Dürr
,
Kelvin
Lau
,
Kristina
Hedfalk
,
Kristof
Bozovicar
,
Lihua
Liu
,
Lisa
Bamber
,
Luigi Angelo
Scietti
,
Malin
Bäckström
,
Maren
Schubert
,
Mark
Elvin
,
Martin
Pelosse
,
Matthew
Todd
,
Matthieu
Schapira
,
Miroslava
Alblova
,
Nicholas S.
Berrow
,
Nikolay
Marinchev Dobrev
,
Oleg
Brodsky
,
Ondřej
Vaněk
,
Opher
Gileadi
,
Pam
Dossang
,
Paola
Storici
,
Patrick
Celie
,
Paul
Wan
,
Peter A.
Loppnau
,
Rachel
Harding
,
Ray
Owens
,
Renaud
Vincentelli
,
Richard
Altman
,
Rick
Davies
,
Rob
Meijers
,
Robbert
Kim
,
Ruth
Saxl
,
Sabine
Suppmann
,
Sameer
Velankar
,
Sebastiano
Pasqualato
,
Sergio
Martinez Cuesta
,
Stephane
Petres
,
Steven
Harborne
,
Susanne
Witt
,
Svend
Kjaer
,
Tamar
Unger
,
Timothy
Craig
,
Tsafi
Danieli
,
Ugis
Sarkans
,
Wolfgang
Knecht
,
Wolfgang
Kuttenlochner
,
Yoav
Peleg
,
Lovisa
Holmberg Schiavone
,
Nicola A.
Burgess-Brown
Open Access
Abstract: As machine learning (ML) becomes increasingly important in protein engineering, synthetic biology, and bioprocess optimisation, the lack of consistent, complete, and standardised experimental reporting remains a major barrier to building effective models. EXPERT (EXpression and Purification Experimental Reporting Template) defines a structured framework for capturing protein expression and purification data across the community, to increase reproducibility across laboratories and support downstream ML applications. We propose here the essential data that should be recorded during expression and purification workflows by both novice and experienced protein production scientists. Its recommendations are informed by expert contributions from across academia and industry. The reporting requirements were distilled to a focused set of parameters deemed sufficient for workflow reproducibility and constructing robust ML models aimed at improving protein production. By deploying these guidelines, we hope to enable the collection of larger data sets to support future model building.
|
Sep 2026
|
|
I23-Long wavelength MX
|
Open Access
Abstract: Preventing collisions during automated sample exchange is critical for synchrotron beamlines, particularly for complex cryogenic in-vacuum endstations where recovery from hardware damage may take days. GoniOwl, a compact convolutional neural network (CNN) model, classifies sample-pin presence on the goniometer from a live camera feed on the long-wavelength macromolecular crystallography beamline I23 at Diamond Light Source. Trained on over 8700 manually verified images spanning two years of routine operation and augmented for robustness to illumination changes, camera shifts and occlusions, the model achieves >99% accuracy with millisecond-level inference. A confidence-gating mechanism routes uncertain predictions to a fail-safe path requiring operator confirmation, ensuring suitability for machine-protection control. Integrated via Experimental Physics and Industrial Control System (EPICS) process variables, GoniOwl runs in real time within the automated sample-change sequence. In shadow-mode deployment, the CNN matched or exceeded both the legacy histogram method and operator confirmations, which each achieved 96% accuracy. A closed-loop disagreement-audit workflow automatically collects divergent cases for targeted retraining and verification. The approach is readily transferable to other beamline environments where camera-based vision systems can provide an additional software machine protection layer.
|
Sep 2026
|
|
I11-High Resolution Powder Diffraction
|
Diamond Proposal Number(s):
[13284]
Open Access
Abstract: Crystal structure prediction (CSP) from powder diffraction data is a central challenge in materials chemistry. Machine learning (ML) models show promise, but most are trained on idealized simulated data, limiting reliability on real experiments. Here, we assess real-world behaviour using the previously published deCIFer model as an example of PXRD-conditioned generative CSP. deCIFer is an autoregressive transformer that conditions each step of structure generation on encoded PXRD data, guiding token-wise predictions of space group, lattice parameters, and atomic positions. Using controlled robustness tests, we quantify performance under realistic artefacts (noise, background, peak asymmetry, and Scherrer broadening) and introduce metrics for accuracy and predictive uncertainty. deCIFer adapts smoothly to signal distortions and improves over unconditioned baselines when diffraction features remain informative, while expressing appropriate uncertainty as the PXRD pattern becomes underdetermined. Experimental PXRD tests recover the known structures of Si and CeO2 and expose the expected limitations for lower-symmetry Fe2O3 and nanocrystalline CeO2. Overall, ML-based CSP is fundamentally limited by the information content of PXRD, but can accelerate expert workflows by rapidly generating chemically plausible candidates and quantifying uncertainty, making such models valuable human-in-the-loop tools for real-world structure determination.
|
Aug 2026
|
|
I03-Macromolecular Crystallography
I04-Macromolecular Crystallography
|
Emily J.
Radley
,
Alessia C.
Andrews
,
Indrek
Kalvet
,
Yunling
Deng
,
Elizabeth L.
Bell
,
Colin W.
Levy
,
Mary
Ortmayer
,
Derren J.
Heyes
,
Clare F.
Megarity
,
Reyes
Núñez-Franco
,
Amy E.
Hutton
,
Yi
Lu
,
David
Baker
,
Anthony P.
Green
Diamond Proposal Number(s):
[38021, 31850]
Abstract: Modern protein design methods based on deep learning allow generation of customized protein scaffolds with diverse geometries and functionalities. Here we capitalize on these recent advances to develop hyper-thermostable de novo CO2 reductases featuring a cobalt porphyrin IX (CoPPIX) cofactor. CoPPIX-containing enzymes were assembled in vivo through media supplementation with cobalt salts and assessed for photocatalytic CO2 reductase activity. We identified two cysteine-ligated designs that exhibit high activity (>1000 turnovers at rates of up to 25 min–1) while suppressing competing hydrogen evolution pathways. A 2.1 Å crystal structure shows close agreement to the design model with the Co–Cys bond programmed as intended. This study showcases the power of computational protein design in developing artificial enzymes to activate challenging molecules such as CO2.
|
Jul 2026
|
|