AIsteels – Materials Intelligence Stack

Skip to content

Materials Intelligence Stack

AIsteels is a pipeline made of three tools forming a coherent stack. Austenitic handles environmental degradation and coating protection at the surface. The FM steel tool handles bulk mechanical behaviour and remaining life in the structural substrate. The HEA tool handles compositional design and property prediction for next-generation alloys that will eventually need both coating and structural assessment. They share the same epistemological architecture — physics-based priors, explicit uncertainty quantification, honest applicability domain boundaries, experimental anchoring targeted at the minimum necessary measurement — so outputs from one can in principle feed inputs to another without methodological inconsistency.

The three tools

Austenitic + coating

A physics-first coating lifetime optimiser built originally for FeCrAl laser cladding on 316L in LBE nuclear service at 600°C.

The core engine uses Arrhenius corrosion kinetics, Miner’s rule damage accumulation, parabolic oxide growth, Griffith spallation sigmoid, and Bayesian Sobol UQ. The architecture is clean, modular, and genuinely reusable. The LBE nuclear case has a coherent plan and a named experimental facility.

Three adjacent case studies — Oil & Gas HP/HT, supercritical CO2 Brayton cycles, and CSP thermal barrier coatings — have been added with correct physics structure but no experimental grounding whatsoever. It is one validated-pathway tool plus three physics-complete sketches.

Ferritic, FM and ODS

A physics-based digital twin for 9-12Cr steels that reduces experimental iteration.

1) Forward design: give it a composition + service condition → it returns properties, degradation state, creep life, risk scores.

2) Inverse design: give it a target property profile → it searches composition space and returns the alloys that hit it.

3) Multi-objective optimisation: give it competing targets (maximise creep life, maintain toughness, etc.) → it returns the Pareto front of compositions.

4) Classification/screening: it flags which degradation mechanisms are active, ranks candidates, and scores HAZ risk before experimental work.

Ni-based and HEAs

Already useful as a screening and Pareto tool for conventional Ni-base development programmes.

With post-calibration on refractory or exotic compositions it becomes genuinely novel, combining physics-based creep and yield prediction with Bayesian optimisation, CALPHAD, and explicit model form uncertainty flagging for this alloy class.

For compositions such as, say, a Mo-Nb-Ta-W-Zr refractory HEA, the model will give physically self-consistent but quantitatively unreliable absolutes until doing roughly 5-8 targeted experiments covering the T/σ extremes of intended service. Below that, even the best optimisation cannot distinguish model bias from aleatoric scatter.

The framework status

A physics-based, uncertainty-quantified, multi-domain computational platform that demonstrably reproduces known literature across four alloy and environment classes, with honest applicability domain boundaries and a targeted experimental ask for each remaining gap. No equivalent public or commercial tool exists spanning ODS structural assessment, conventional FM steel remaining life, HEA compositional design, and surface coating degradation on a consistent UQ architecture (from case study #4, for LBE-cooled fast reactor applications as an example).