NpjComput.Mater.：前沿機器學習：更新傳統氧化機理，加速新型渦輪材料開發

海歸學者發起的公益學術平臺

分享資訊，整合資源

交流學術，偶爾風月

在現代渦輪系統中，其關鍵部件渦輪葉片，通常透過在鎳基高溫合金基體上噴塗陶瓷基熱障塗層（TBC）來實現熱防護。為提高塗層與基體之間的結合強度，並在高溫環境中提供抗氧化效能，需在兩者之間引入金屬粘結層，其透過形成穩定的氧化鋁層來提供防護。目前工業上廣泛採用的金屬粘結層材料是 MCrAlY 合金（M 通常為 Ni 和/或 Co）。然而，隨著燃燒室溫度的不斷提升，開發具備優異抗氧化效能、可替代 MCrAlY 的新型材料已成為行業研究的重點方向。

來自美國西弗吉尼亞大學機械、材料與航空航天工程系的胡山山教授團隊，提出了一種創新性的材料開發框架，結合先進的機器學習技術與傳統熱力學理論，可利用已有實驗資料構建預測模型，全面提取 Ni-Co-Cr-Al-Fe 高熵合金在高溫氧化環境下的行為特徵。

該框架融合了機器學習演算法與 CALPHAD 熱力學計算方法，在複雜的 Ni-Co-Cr-Al-Fe 多元高熵成分空間中，成功篩選出若干種氧化效能優於傳統 MCrAlY 的新型金屬材料。值得注意的是，傳統氧化機理中被廣泛接受的“第三元素效應”在簡單合金體系中已被充分驗證，但在高熵合金這樣高度複雜的成分系統中，氧化行為更加複雜和多變。該研究團隊所建立的機器學習框架，恰恰在繼承和融合傳統氧化機理的基礎上，實現了對以動力學機制為主導的複雜氧化行為的有效預測。除了開發出可根據成分精準預測氧化速率的模型，該研究還揭示了高熵合金在氧化過程中的若干關鍵行為機制：

在多相合金體系中，富含有益元素（如鋁和鉻）的各個相需保持成分均衡，避免區域性元素富集或偏析，方能穩定形成緻密、黏附性強、長期無剝落的氧化鋁保護膜；
具有奈米晶粒結構的多相合金顯著提升了整體抗氧化效能，為新一代高溫結構材料的設計提供了新思路。

Figure 1 Material screening workflow. Design framework for high-temperature oxidation resistance Ni-Cr-Co-Al-Fe based high-entropy alloy.

Figure 3 Machine Learning Model Performance Parity plot of the predictive kp value versus the literature reported kp values of 7 machine learning approaches: (a) artiﬁcial neural network (ANN), (b) gradient boosting regression (GBR), (c) kernel ridge regression (KRR), (d) support vector regression (SVR), (e) gaussian process regression (GPR), (f) decision trees regression (DTR) and (g) multilayer perceptron regression (MLPR) machine learning algorithm. Blue and red plots represent train, and test set, respectively. Mean squared error (MSE), coefﬁcient of determination (R2) and mean absolute error (MAE) are computed to estimate the prediction errors. Black dashed line is the ideal fitting line. (h) UMAP projections of predicted kp using the GBR model at 1150 °C overlay the complete Ni-Cr-Co-Al-Fe design space.

該文近期發表於npj Computational Materials11, 93(2025)，英文標題與摘要如下，點選左下角“閱讀原文”可以自由獲取論文PDF。

Machine learning and high-throughput computational guided development of high temperature oxidation-resisting Ni-Co-Cr-3 Al-Fe based high-entropy alloys

Xingru Tan, William Trehern, Aditya Sundar, Yi Wang, Saro San, Tianwei Lu, Fan Zhou, Ting Sun, Youyuan Zhang, Yuying Wen, Zhichao Liu, Michael Gao & Shanshan Hu

Ni-Co-Cr-Al-Fe-based high-entropy alloys (HEAs) have been demonstrated to possess exceptional oxidation resistance, rendering them promising candidates as bond coats to protect critical components in turbine power systems. However, with the conventional time-consuming alloy design approach, only a small fraction of Ni-Co-Cr-Al-Fe-based HEAs, focusing on equiatomic compositions, has been explored to date. In this study, we developed an effective design framework with the aid of machine learning (ML) and high throughput computations, enabling the rapid exploration of high-temperature oxidation-resistant non-equiatomic HEAs. This innovative approach leverages ML techniques to swiftly select candidates with superior oxidation resistance within the expansive high-entropy composition landscape. Complemented by a thermodynamic-informed ranking-based selection process, several novel non-equiatomic Ni-Co-Cr-Al-Fe HEA candidates surpassing the oxidation resistance of the state-of-the-art bond coat material MCrAlY have been identified and further experimentally demonstrated. Our findings offer a pathway for the development of advanced bond coats in the realm of next-generation turbine engine technology.