军事知识服务平台
基于深度学习的难熔高熵合金密度预测
Prediction of Density of Refractory High-Entropy Alloys Based on Deep Learning
摘要 难熔高熵合金(refractory high-entropy alloys,RHEAs)因其优异的高温性能、力学性能和抗氧化性能,在航空航天等领域应用前景广阔,但高密度特性限制了其在重量敏感领域的应用.构建了一种新型合金嵌入网络,结合深度学习方法,实现了对RHEAs 密度的高精度预测.与传统机器学习方法的对比结果显示:该网络在预测双相和多相合金密度时性能最优,决定系数(R2)超过0.95,能有效揭示成分与密度之间的非线性关系.单相合金的密度主要受成分控制,可通过晶胞参数的线性叠加计算得出,机器学习的优势相对有限;而在复杂的多相体系中,合金嵌入网络通过可学习的元素表示和加权机制,实现了更精准的预测,并展现出较强的泛化能力.
Abstract:
Refractory high-entropy alloys(RHEAs)have broad application prospects in aerospace and other fields due to excellent high-temperature performance,mechanical properties,and oxidation resistance.However,the high density restricts their application in weight-sensitive fields.A novel alloy embedding network was constructed,in combination with deep learning techniques,therefore the high-precision density prediction for RHEAs was achieved.Comparison results with traditional machine learning methods indicated that the proposed network exhibited superior performance in predicting the density of dual-phase and multiphase alloys,with the coefficient of determination(R2)exceeding 0.95,and could effectively reveal the nonlinear relationship between composition and density.The density of single-phase alloys was primarily determined by composition and could be derived from the linear superposition of lattice parameters,the advantages of machine learning approaches were relatively limited.In contrast,for the complex multiphase systems,the alloy embedding network achieved more precise predictions through learnable element representations and weighting mechanisms,and demonstrated strong generalization capability.
刊 名 上海金属 ISTIC
年,卷(期): 2026, 48(5)
分类号: TG132 TP181
机标分类号: F830.9 TP301.6 U616
在线出版日期: 2026年9月30日
基金项目: 52003150:国家自然科学基金,