Synergistic Alloy Steel Development using Multi-scale Computational Modeling and Machine Learning
Advanced steel alloys have attracted extensive interest from researchers in academia and industry owing to their applications in a wide range of fields such as automobile, aerospace, and defense industries. As these industries constantly evolve to meet technological demands, the need for accelerated steel design has surged. Numerous steels with enhanced properties have emerged through different generations of development. However, the traditional experimental method to design steel alloys, because of its intensive and costly nature, is unable keep pace with the current requirement. To accelerate the steel alloy development, we aim to demonstrate a framework that utilizes Integrated Computational Materials Engineering (ICME). The framework focuses on two key aspects of next-generation steel, specifically High Entropy Steels (HES): a) mechanical properties prediction and b) phase prediction. The mechanical properties prediction models were built on the accuracy of the DFT calculations to develop the interatomic potential for Molecular Dynamics (MD) and further Machine Learning (ML) to expedite the properties prediction. An interatomic potential for Fe-C was developed by fitting on forces, energies, and stress tensor data from ab-initio calculations. Subsequently, an extensive dataset was generated using the potential for various ordered and disordered Fe-C alloys. Seven individual machine learning algorithms and an ensemble approach were trained on this dataset, and two methodologies, multi-variate prediction, and independent prediction, were evaluated based on the error in prediction. Expanding upon this groundwork, Mn was introduced into the system to investigate the properties of various Fe-Mn-C compositions. Two interatomic potentials, MEAMfit and DeePMD, were developed and used to reproduce mechanical properties from literature, with their prediction errors compared. Additionally, a phase prediction study for HEAs was conducted, as the phases formed (e.g., SS or IM) significantly influence alloy mechanical properties. Four machine learning algorithms were trained on collected datasets in conjunction with sampling techniques. The most accurate model was applied to a limited HES dataset to determine the phases they formed. Hence, a framework for studying advanced steel materials’ properties was established. This framework aims to accelerate the exploration of compositional and temperature space, facilitating the development of innovative advanced steel alloys.