Scholarlink

<p>Hypersonic vehicle engine designs require addressing potential metallurgical failures due to operation in extreme environments: sustained high temperatures (1200 to 2000&#xa0;K), aggressive oxidative atmospheres, ultrafast thermomechanical cycling, and high strain-rate deformations. High-entropy alloys (HEAs), first introduced by Yeh and Cantor in 2004 as alloys with 5 to 13 elements, have emerged as candidates for such extreme environments. Specifically, AlCoCrFeNi HEAs offer low density and high-temperature yield strength (YS), making them suitable for hypersonic scramjet environments. However, the vast number of elemental and preparation combinations make purely experimental exploration prohibitely expensive. Deep Learning (DL) models can predict HEA properties from millions of compositions, but current methods often overfit by relying on original dataset values, failing to capture the complex, nonlinear relationships between structural, thermodynamic, and processing parameters. More importantly, DL models fail to clearly disaggregate which parameters dominate in governing alloy strength. FORGEall addresses the limitations of these HEA discovery pipelines through feature transposition while maintaining model accuracy. Devising latent space projections through a novel semi-supervised sparse autoencoder and incorporating K-Nearest Neighbor regression, FORGEall improves generalization, achieving 98% accuracy in YS predictions. Unlike prior work, the semi-supervised component of the framework disentangles and isolates the governing properties of alloy performance. FORGEall’s top two HEAs (FORGEAlloy-1: Al10Cr40Fe20Co10Ni10Cu10 and FORGEAlloy-2: Al13.33V6.67Cr26.67Fe20Co13.33Ni6.67Cu6.67Mo6.67) achieve YS values of 1850&#xa0;MPa, ideal for hypersonic engines. Independent verifications using LAMMPS molecular dynamics simulations validate the predictions within 0.6% and 3.8%. Additionally, simulations using Ovito attribute this performance to high lattice distortion, while ThermoCalc-CALPHAD indicates high-temperature phase stability. These HEAs exhibit high strength-to-weight ratios (densities less than 9&#xa0;g/cm3),low cost, and outperform current industry alloys. Thus, FORGEall enables efficient exploration of compositional spaces for developing fatigue-resistant hypersonic engine materials.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

FORGEall: A Deep Learning Framework to Reliably Predict High Yield Strength High-Entropy Alloys for Hypersonic Engines

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Avni Iyer

摘要

Hypersonic vehicle engine designs require addressing potential metallurgical failures due to operation in extreme environments: sustained high temperatures (1200 to 2000 K), aggressive oxidative atmospheres, ultrafast thermomechanical cycling, and high strain-rate deformations. High-entropy alloys (HEAs), first introduced by Yeh and Cantor in 2004 as alloys with 5 to 13 elements, have emerged as candidates for such extreme environments. Specifically, AlCoCrFeNi HEAs offer low density and high-temperature yield strength (YS), making them suitable for hypersonic scramjet environments. However, the vast number of elemental and preparation combinations make purely experimental exploration prohibitely expensive. Deep Learning (DL) models can predict HEA properties from millions of compositions, but current methods often overfit by relying on original dataset values, failing to capture the complex, nonlinear relationships between structural, thermodynamic, and processing parameters. More importantly, DL models fail to clearly disaggregate which parameters dominate in governing alloy strength. FORGEall addresses the limitations of these HEA discovery pipelines through feature transposition while maintaining model accuracy. Devising latent space projections through a novel semi-supervised sparse autoencoder and incorporating K-Nearest Neighbor regression, FORGEall improves generalization, achieving 98% accuracy in YS predictions. Unlike prior work, the semi-supervised component of the framework disentangles and isolates the governing properties of alloy performance. FORGEall’s top two HEAs (FORGEAlloy-1: Al10Cr40Fe20Co10Ni10Cu10 and FORGEAlloy-2: Al13.33V6.67Cr26.67Fe20Co13.33Ni6.67Cu6.67Mo6.67) achieve YS values of 1850 MPa, ideal for hypersonic engines. Independent verifications using LAMMPS molecular dynamics simulations validate the predictions within 0.6% and 3.8%. Additionally, simulations using Ovito attribute this performance to high lattice distortion, while ThermoCalc-CALPHAD indicates high-temperature phase stability. These HEAs exhibit high strength-to-weight ratios (densities less than 9 g/cm3),low cost, and outperform current industry alloys. Thus, FORGEall enables efficient exploration of compositional spaces for developing fatigue-resistant hypersonic engine materials.