Coupling physics in machine learning to predict properties of high-temperatures alloys

Abstract

High-temperature alloy design requires a concurrent consideration of multiple mechanisms at different length scales. We propose a workflow that couples highly relevant physics into machine learning (ML) to predict properties of complex high-temperature alloys with an example of the 9–12 wt% Cr steels yield strength. We have incorporated synthetic alloy features that capture microstructure and phase transformations into the dataset. Identified high impact features that affect yield strength of 9Cr from correlation analysis agree well with the generally accepted strengthening mechanism. As a part of the verification process, the consistency of sub-datasets has been extensively evaluated with respect to temperature and then refined for the boundary conditions of trained ML models. The predicted yield strength of 9Cr steels using the ML models is in excellent agreement with experiments. The current approach introduces physically meaningful constraints in interrogating the trained ML models to predict properties of hypothetical alloys when applied to data-driven materials.

A machine learning aided interpretable model for rupture strength prediction in Fe-based martensitic and austenitic alloys

“…From the table, it is evident that the model is not overfitting the data as the training and testing performance are quite similar. However, considering the performance of the state-of-the-art model ( > 0.95) for yield stress prediction reported in the literature 33 for a similar FMA dataset and the high variance of the , this performance is deemed as unsatisfactory for accurate prediction of rupture strength. It is worth noting that in that study, in addition to the chemical composition and processing parameters, synthetic alloy features generated using CALPHAD were incorporated to capture microstructural and phase transformation related information.…”

Materials Informatics for Mechanical Deformation: A Review of Applications and Challenges

“…Pattern detection in the research of glassy matter is generally regarded as a supervised learning process where appropriate machine learning features (and/or regressors) are trained based on a large set of atom-wise dynamical measures (atomic trajectories, mobility, or rearrangements) and corresponding local structural information (e.g., average pair correlation function, coordination number, bond orientation etc.) [ 17 , 115 , 116 , 117 ]. The concept was largely put forward in a series of important papers applying machine learning to describe the interplay between structure and dynamics in several glass formers as well as polycrystals [ 42 ].…”

Data analytics approach to predict high-temperature cyclic oxidation kinetics of NiCr-based Alloys

“…The MIC method can identify the strength of both linear and nonlinear relationships, but only the magnitude without a sign. Both approaches are expected to provide insights into the correlation between input features and k p from differing statistical aspects, which may inspire alloy design experts to generate alloy hypotheses 23,26 . Moreover, correlation analysis can facilitate the training of high-fidelity ML models using highly ranked features.…”

“…Recently, data analytics approaches have been successfully applied to predict the mechanical properties of multi-component high-temperature alloys 18,[22][23][24][25][26][27] . While high-temperature oxidation also has scientific and practical importance, to the best of the authors' knowledge, very limited effort has been made to predict the oxidation kinetics of complex multi-component alloys by ML.…”

“…From a statistical perspective, a good agreement between the predicted and experimental k p was achieved. However, their dataset was collected from multiple literature sources; thus, the data may exhibit mixed oxidation mechanisms due to the difference in experimental protocols (i.e., atmospheres and isothermal or cyclic exposures) 26 . Moreover, their work focused only on building predictive models without analyzing the relationship between input features and k p .…”