Active Learning Discovery of High Temperature Oxidation Resistant Refractory Complex Concentrated Alloys

Akhil Bejjipurapu, Sharmila Karumuri, Joseph C. Flanagan, Victoria Tucker, Ilias Bilionis, Alejandro Strachan, Kenneth H. Sandhage, Michael S. Titus · 2025

Read Active Learning Discovery of High Temperature Oxidation Resistant Refractory Complex Concentrated Alloys by Akhil Bejjipurapu, Sharmila Karumuri, Joseph C. Flanagan, Victoria Tucker, Ilias Bilionis, Alejandro Strachan, Kenneth H. Sandhage, Michael S. Titus (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-15958.txt.

Introduction

Akhil Bejjipurapu and co authors use active learning to find heat resistant refractory alloys in a 2025 paper. These alloys suit extreme high heat uses, but their vast makeup space defeats standard trial and error search. The team focuses on aluminum containing four element systems and oxidation at 1,000 degrees Celsius.

Starting from 81 lab validated alloys, a Gaussian process model with Bayesian search picks batches of five alloys per round, balancing new ground and known ground. After six rounds, two alloys show mass gains under 1 milligram per square centimeter at 1,000 degrees in air. Both form adherent alumina scales with parabolic oxidation kinetics, and both show high hardness per weight, suiting them as bond coat candidates.

Catalog record

CREATORAkhil Bejjipurapu, Sharmila Karumuri, Joseph C. Flanagan, Victoria Tucker, Ilias Bilionis, Alejandro Strachan, Kenneth H. Sandhage, Michael S. Titus
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE78 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-15958

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Active Learning Discovery of High Temperature Oxidation Resistant Refractory Complex Concentrated Alloys Akhil Bejjipurapu, Sharmila Karumuri, Joseph C. Flanagan, Victoria Tucker, Ilias Bilionis, Alejandro Strachan, Kenneth H. Sandhage, Michael S. Titus Abstract Refractory complex concentrated alloys (RCCAs) are of significant interest for advanced high-temperature applications, owing to their broad compositional range and potential for attractive mechanical properties and oxidation resistance. However, their compositional complexity poses significant challenges to conventional alloy discovery methodologies. In this study, an active learning framework is introduced that integrates Gaussian process regression with Bayesian global optimization to accelerate identification of oxidation-resistant RCCAs. Focusing on aluminum-containing quaternary systems, alloy and oxide descriptors were used to predict oxidation performance at 1000°C. Beginning with a dataset of 81 experimentally validated RCCAs, this framework was used to iteratively select alloy batches (five alloys per batch) with optimization based on a balance between exploration and exploitation to minimize associated experimental costs. After six iterations, two alloys were identified (nominal Al 30 Mo 5 Ti 15 Cr 50 and Al 40 Mo 5 Ti 30 Cr 25 ) that exhibited specific mass gains less than 1 mg/cm 2 at 1000°C in air. Both of these alloys formed adherent external α \alpha -Al 2 O 3 scales and exhibited parabolic oxidation kinetics consistent with diffusion-limited scale growth. Furthermore, our multiobjective analysis demonstrates that these alloys simultaneously achieve high specific hardness ( > 0.12 >0.12 HV 0.5 m 3 /kg) and thermal expansion compatibility with thermal barrier coating systems, positioning them as promising bond coat candidates. This work underscores the efficacy of active learning in traversing complex compositional landscapes, and offers a scalable strategy for the development of advanced materials suitable for extreme environments. Keywords: Refractory complex concentrated alloys (RCCAs), High-temperature oxidation, Active learning, Bayesian optimization, Alumina forming RCCAs Introduction The development of materials capable of withstanding extreme environments, particularly at elevated temperatures, remains a critical challenge for the aerospace, energy, and defense sectors. Refractory complex concentrated alloys (RCCAs), which comprise a subclass of multi-principal element alloys that incorporate refractory metals such as Ti, Zr, Hf, V, Nb, Ta, Cr, Mo, and W, present promising alternatives to conventional alloys for high-temperature applications [ 1 , 2 , 3 ] . Nevertheless, the potential use of RCCAs is often curtailed by insufficient environmental resistance. In particular, the inability to develop dense, slow-growing, adherent (protective) oxide scales under oxidative conditions compromises their high-temperature performance. One critical application area where oxidation resistance is paramount is in thermal barrier coating (TBC) systems used in gas turbines and aerospace engines. In such systems, the bond coat plays a dual role of promoting adherence of the ceramic topcoat and of forming a protective oxide barrier [ 4 , 5 ] . MCrAlY (M = metal, Y = small elemental editions)-based bond coats are commonly used, but their limited compatibility with lightweight substrates, such as \gamma -TiAl, has prompted interest in alternative compositions [ 6 , 7 ] . RCCAs with tailored Al and Cr contents offer a promising new class of bond coat candidates with the potential to address both oxidation and substrate compatibility requirements. In contrast to typical RCCAs, advanced nickel-based superalloys with as little as 8–9 at.% Al effectively form protective \alpha -Al 2 O 3 scales for long-term, high-temperature environmental resistance [ 8 ] . Perhaps surprisingly, even with up to 20 at.% Al, RCCAs frequently fail to achieve similar protective behavior. Earlier efforts to develop CrTaO 4 -based oxide layers on RCCAs have only partially addressed oxidation resistance, as these scales are prone to internal oxidation, with deleterious species such as nitrogen penetrating into the substrate and contributing to embrittlement [ 9 , 10 , 11 , 12 ] . Hence, the identification of RCCAs capable of forming continuous, adherent, slow-growing, external \alpha -Al 2 O 3 scales for achieving high-temperature oxidation resistance remains a strong desire. A major challenge with RCCAs lies in their vast composition space, making traditional methods like trial-and-error and DFT-based calculations time consuming and inefficient. Although CALPHAD-based modeling has been attempted [ 13 , 14 , 15 ] , it remains costly and time intensive. Consequently, leveraging advances in artificial intelligence and machine learning, particularly active learning methods, has gained traction. These methods utilize a combination of machine learning (ML) surrogates and Bayesian optimization to efficiently navigate the expansive composition space, so as to identify a limited number of promising alloys for testing [ 16 , 17 , 18 , 19 , 20 , 21 , 22 ] . This methodology can accelerate the discovery process and thereby enhance the efficiency and impact of materials research. While machine learning (ML) has been successfully applied to predict properties such as yield strength [ 23 , 24 ] , hardness [ 25 , 26 ] , and low-temperature corrosion resistance [ 27 ] , the application of ML for identifying RCCAs with high-temperature oxidation resistance has not been as extensive. Several studies have highlighted the capability of ML to identify particular oxidation-related characteristics, including mass change [ 28 , 29 , 30 ] , oxide scale thickness [ 31 ] , effective activation energy (Arrhenius behavior) [ 32 , 33 ] , and the associated parabolic rate constant ( k_{p} ) [ 33 , 34 , 35 ] . For example, Kim et al. [ 36 ] applied an artificial neural network (ANN) to forecast mass gain in a complex Ni-based alloy system comprising Ni-Co-Cr-Mo-W-Al-Ti-Ta-C-B, using proprietary high-throughput experimental datasets. Duan et al. [ 29 ] developed a model to predict mass change in Ni-based superalloys within the Ni-Co-Cr-Mo-Al-Ti-Nb-Hf-Zr alloy system. Bhattacharya et al. [ 34 ] constructed a regression model to estimate k_{p} values for Ti-based alloys, and Dewangan et al. [ 37 ] utilized an ANN to predict mass gain in AlCrFeMnNiW x alloys (where x=0,0.05,0.1,0.5 ). Mishra et al. [ 38 ] developed an open-source computational workflow that systematically fits mass-gain data to multiple oxidation kinetics models and leverages Bayesian statistics for robust model selection . Despite these advancements, prior work has largely centered on optimizing oxidation resistance within restricted compositional domains or clarifying empirically observed trends between alloying elements and oxidation behavior. Notably, Gorsse et al. [ 39 ] recently employed Gradient Boosted Decision Trees (GBDT) to predict specific mass gain due to oxidation of refractory high-entropy alloys (RHEAs), and demonstrated significant accuracy while exploring an extensive compositional design space. However, most existing models do not incorporate oxide-specific thermodynamic features, and few studies have extended to broader design spaces or to iteratively-refined predictions through feedback from experiments. In this study, we introduce an innovative active learning framework that integrates Gaussian process regression (GPR) with Batch- Bayesian global optimization ( Batch- BGO) to efficiently and rapidly discover oxidation-resistant RCCAs. This methodology was adapted from an approach previously developed by Karumuri et al. [ 40 ] to discover alloys with high hardness. The workflow, illustrated in Figure 1 , synergistically combines alloy- and oxide-specific descriptors—including CALPHAD-derived phase stability and oxidation-related thermodynamic data—to predict oxidation resistance at 1000°C in air. The process involves defining the alloy design space, preparing the initial dataset, calculating descriptors, building a predictive GPR model, and optimizing the selection of alloys through BGO using Expected Improvement (EI) as

Read the complete text