Source-linked AI summary
Design of a nickel-base superalloy using a neural network
B. D. Conduit, N. G. Jones, H. J. Stone, G. J. Conduit
TL;DR
Nickel-base superalloys must balance at least eleven target properties for demanding applications. The paper uses uncertainty-aware neural-network predictions to search for an alloy most likely to meet them. Experiments on the proposed polycrystalline V210A confirmed better yield stress and oxidation resistance than commercially available alternatives.
Problem
Nickel-base superalloys used in gas-turbine engines must simultaneously satisfy at least eleven target properties.
Source: §2 · p. 3Method
Results
Seven properties were experimentally verified, and V210A showed better yield stress and oxidation resistance than commercially available alternatives.
Source: §3 · p. 8Takeaways & Limitations
The tool identified V210A as an ideal compromise for disc applications.
Source: §3 · p. 8
Abstract
from arXiv · showhide
A new computational tool has been developed to model, discover, and optimize new alloys that simultaneously satisfy up to eleven physical criteria. An artificial neural network is trained from pre-existing materials data that enables the prediction of individual material properties both as a function of composition and heat treatment routine, which allows it to optimize the material properties to search for the material with properties most likely to exceed a target criteria. We design a new polycrystalline nickel-base superalloy with the optimal combination of cost, density, gamma' phase content and solvus, phase stability, fatigue life, yield stress, ultimate tensile strength, stress rupture, oxidation resistance, and tensile elongation. Experimental data demonstrates that the proposed alloy fulfills the computational predictions, possessing multiple physical properties, particularly oxidation resistance and yield stress, that exceed existing commercially available alloys.
1. Formalism
The formalism combines property-specific predictive models with uncertainty estimates, then searches compositions for the highest probability of meeting a multi-criteria specification.
1. Formalism
- The design workflow builds predictive models for each property, calculates the probability a composition meets targets, then searches for the most likely candidate.Source: §1 · p. 2
- Cost and density use weighted averages, while phase stability is screened with PhaComp and cross-checked for the final composition using CALPHAD.PhaComp estimates phase stability from transition-metal d-orbital energy levels; Md < 0.98 eV indicates acceptably low formation of topologically closed packed phases.Source: §1 · p. 2
- A committee of 64 neural networks predicts each property from composition and heat-treatment inputs, with variation across models estimating prediction uncertainty.The networks are trained on pre-existing experimental data; typically three hidden nodes minimize the reduced chi-squared statistic.Source: §1 · p. 2
- Maximizing log probability of meeting all targets preserves a gradient that favors improving the least-optimized property.The formalism combines individual property probabilities to evaluate overall conformance.Source: §1 · pp. 2–3
- Prediction uncertainty grows farther from existing data, bounding reliable extrapolation while allowing interpolation and exploration of property trade-offs.The authors contrast this likelihood-based search with principal component analysis and robust design, which do not account for likelihood.Source: §1 · p. 3
- Simulated annealing uses steps of about 0.1 wt.% per element and adjusts temperature to accept a target fraction of 0.267 of steps.The authors describe this search as rapidly exploring design space without getting stuck in local minima.Source: §1 · p. 3
2. Results and discussion
V210A was designed as a compromise across demanding superalloy targets, and its measured properties broadly matched predictions. In particular, it exceeded the yield-stress and oxidation-resistance targets that none of the exemplar commercial alloys met simultaneously.
2. Results and discussion
- The design sought better high-temperature yield strength and oxidation resistance while remaining competitive across eleven targets, including an elongation threshold of 11.6%.The γ′ content target kept solvus below 1150°C because higher-solvus alloys can be difficult to process and suffer quench cracking.Source: §2 · p. 3
2.0.2. Experimental verification
- V210A's measured properties agreed with predictions within uncertainty and exceeded the yield-stress and oxidation-resistance targets that no exemplar commercial alloy met together.The comparison covered cost, density, γ′ content, phase stability, yield stress, Cr activity, and γ′ solvus.Source: Figure 2 · p. 4, §2.0.2 · pp. 7–8
2.0.1. Analysis of proposed alloy
- V210A had higher overall target-conformance probability than compositionally similar Alloys A and B, despite all three being predicted to exceed the yield-stress target.Alloy A fell short on phase stability, while Alloy B had poor predicted phase stability and density.Source: §2.0.1 · pp. 4–5
2.0.2. Experimental verification
- V210A's predicted γ′ solvus was 1100°C; microscopy found 51% γ′ precipitates, less than 1% other deleterious phases, and a solvus within 10°C of prediction.The predicted oxidation resistance agreed with experiment and was superior to RR1000; measured yield stress exceeded RR1000's experimental value within expected uncertainty.Source: §2.0.1 · p. 5, §2.0.2 · p. 7, Figure 7 · p. 7
2.0.1. Analysis of proposed alloy
- At stress rupture of 850 MPa and phase stability of 0.95 eV, predicted joint success was generally highest; more rupture resistance with lower phase stability became increasingly unlikely.Increasing rupture resistance generally requires additional alloying elements, which decreases phase stability.Source: §2.0.1 · p. 5, Figure 4 · p. 6
- Refractory additions can improve thermal expansion and yield stress together, but raise density and reduce phase stability, making many compositions infeasible.After considering all targets, V210A had the highest probability among the plotted alloys despite its relatively high thermal expansion coefficient.Source: §2.0.1 · pp. 5–6, Figure 5 · p. 6
3. Conclusions
The paper develops an uncertainty-aware computational tool to design alloys with the greatest probability of meeting specifications spanning many properties. Applied to nickel-base superalloys, it proposed a polycrystalline alloy whose seven experimentally verified properties support better yield stress and oxidation resistance than commercially available alternatives.
3. Conclusions
- The tool incorporates uncertainty to identify alloys most likely to meet a specification containing many material properties.It was used to propose a nickel-base superalloy designed to fulfill eleven physical criteria.Source: §3 · p. 8
- Seven experimentally verified properties showed that the proposed polycrystalline alloy has better yield stress and oxidation resistance than commercially available alternatives.The tool predicted an ideal property compromise for disc applications.Source: §3 · p. 8
- The approach has also been used to design a nickel-base alloy for a combustor liner and two Mo-based forging-tool alloys.The authors conclude it could help engineers optimize bespoke materials for specific applications.Source: §3 · p. 8