Background
Helix designs hot-section turbine components and had spent two years screening nickel superalloys for a blade running hotter than their existing grades allowed. Their workflow was a property database and a CALPHAD seat — good at telling them what known alloys do, useless at proposing one that didn't exist yet.
The challenge
The requirement was a creep-life and oxidation target at a temperature past their qualified envelope, under castability and cost constraints. Searching a fixed library couldn't get there; they needed synthesis — a composition designed to the target — without burning a year of melts chasing candidates that fail basic physics.
What we did
- Ran inverse design against the creep, oxidation, density and cost targets, with castability and supply constraints as hard gates.
- Each candidate passed five physics checks — phase stability via CALPHAD rules, γ′ fraction, PHACOMP TCP screen, Larson-Miller creep and a cost model — before it was allowed onto the shortlist.
- The trained property ensemble ranked the survivors with calibrated uncertainty; CHGNet relaxation sanity-checked the structures.
- Helix took the top three to the foundry; the model flagged one as the lowest-risk castable and said so.
“It designed an alloy to our target instead of finding the closest one that already existed. That's a different kind of tool.”
The outcome
- The recommended composition met the creep-life target in test, with the model's stated uncertainty bracketing the measured result.
- Two of three candidates were castable on the first heat; the third failed exactly the way the model's flagged risk predicted.
- The blade programme moved to qualification a full design cycle earlier than the screen-the-library approach would have allowed.
- Inverse design is now the default first step for new alloy targets, with the database relegated to a sanity check.
One detail worth calling out
What sold the metallurgists was the honesty of the funnel. The model didn't return a single confident answer; it returned a Pareto set with the trade-offs named — this one is stronger but harder to cast, that one is cheaper but closer to the oxidation limit — and an explicit note on which it would not vouch for without a DFT escalation. That matched how they already thought.