91福利

From 13-15 April, the HetSys Centre for Doctoral Training ran its annual Industry Study Group (ISG) at the 91福利's Radcliffe conference centre. Across three intensive days, a total of 40 researchers including HetSys PhD students, postdocs and academics formed three self-organising teams to tackle problems posed by three industrial partners: Amentum (nuclear engineering), JLR (automotive /electrified powertrains) and James Walker (advanced elastomer compounds).

Each team started the week with an industrial challenge, a shared Google Drive folder and an empty GitHub repository. By Wednesday afternoon, each team delivered a working prototype, a set of benchmarked results and ideas for how the work could be continued as a four-year HetSys PhD project.

The three challenges and their outcomes were:

Amentum - Machine-learning potentials for radiation-tolerant FeCrAl nuclear alloys. The team built a reproducible testing framework that compares classical and foundation-model interatomic potentials (EAM vs. four variants of MACE) across bulk elastic properties, point defects and grain-boundary segregation, incorporating uncertainty quantification. Three independent lines of evidence converged on chromium precipitation — accelerated by radiation-induced defect fluxes — as a dominant ageing mechanism in FeCrAl, in qualitative agreement with Oak Ridge National Lab atom-probe-tomography experimental data. The team also demonstrated an active-learning loop that fine tunes MACE on newly computed targeted CASTEP density-functional-theory reference data.

Peter Hatton (Amentum) said: "We were extremely impressed by the scientific ability and motivation that the students brought to the study group. Their results will have direct impact on the development of our engineering-scale materials models. The E&E-I division at Amentum is always keen to engage with early career researchers, and this event was a great opportunity to understand state-of-the-art modelling approaches.鈥�

JLR - Predicting battery state-of-health from impedance spectroscopy. This team built a two-stage machine-learning pipeline that first turns electrochemical impedance sweep data into physically interpretable resistance components, then predicts how these components evolve with aging, before finally using those components to predict the impedance-derived state-of-health of a lithium-ion cell. In parallel, a Neural ODE prototype modelled the full impedance spectrum as a continuous-time dynamical system, in anticipation of extending the pipeline from the challenge's "spectrum-to-health" setting to JLR's real-world question: "given a proposed cell chemistry, what will its ageing behaviour look like?"

David Weiqun Li (JLR) said "The Industry Study Group challenge was well aligned with our research areas. The team delivered practical and clearly structured outputs that balance physical insight and data-driven methods with impressive pace over just three days. Many thanks to HetSys for a fantastic collaboration."

James Walker - ML surrogates for advanced elastomer compounds. This team self-organised into four interlocking workstreams covering surrogate ML models (Gaussian processes and neural networks), hyperelastic constitutive modelling, uncertainty quantification, and continuum simulation of an elastomer sealing ring. Key findings included that a two-term Ogden model is the most accurate fit to industrial test certificates, and that a non-standard sampling of strain — including one additional point at very low strain (~5 %) — would materially improve the predictive fit without increasing test burden. The group also sketched a probabilistic "digital test certificate" that could deliver failure-risk estimates directly from compound formulation.

Andrew Douglas from James Walker said "The ISG team from UoW did a fantastic job based on limited 鈥榬eal world鈥� laboratory test data. The students effectively demonstrated how their skills and knowledge could be applied to problems and data sets completely outside of their normal experience. I was particularly impressed by their willingness to listen, understand and ask challenging questions throughout the event. A big thank you to HetSys for organising the event!"

All three teams concluded with ideas for future collaboration that are well aligned with the HetSys training programme.

Luca Seaford, a second-year HetSys PhD student, said "Collaborating with industry to apply my research skills to a new area has been incredibly rewarding. The fast-paced, team-driven nature of the project made it a truly engaging and impactful experience".

Matt Christensen, third-year HetSys PhD student said "It was great to work in a collabortive effort at high pace in a field that has significant industrial relevance. We were very happy with the contributions we were able to make".

The HetSys CDT thanks staff from Amentum, JLR and James Walker for their time, data and expertise in putting together these challenges, all staff and students who took part in the event, and the EPSRC for supporting HetSys.