Development and application of machine-learned interatomic potentials for the study of high-temperature TiAlNb alloys
Citation Link: https://doi.org/10.15480/882.17353
Publikationstyp
Doctoral Thesis
Date Issued
2026
Sprache
English
Author(s)
Advisor
Referee
Title Granting Institution
Technische Universität Hamburg
Place of Title Granting Institution
Hamburg
Examination Date
2026-05-08
TORE-DOI
Citation
Technische Universtität Hamburg (2026)
The lack of accurate and efficient interatomic potentials for technologically vital TiAlNb alloys has hindered their computational design and optimization for aerospace and energy applications. The thesis addresses the gap through a hierarchical investigation that begins by establishing the limitations of a conventional classical interatomic potential for the system. To address these limitations, the work then details the development and application of two state of the art machine learning interatomic potentials (ML-IAPs), the neural network-based deep potential molecular dynamics (DeePMD) and the linear regression-based moment tensor potentials (MTP), to investigate the influence of Niobium on the thermomechanical properties of TiAlNb alloys. Furthermore, the efficiency of the interatomic potential development process itself is systematically evaluated by contrasting a traditional passive learning approach with a data-efficient active learning strategy. The results demonstrate that the developed ML-IAPs achieve near-ab initio accuracy, revealing a clear trade-off between the higher fidelity of DeePMD and the greater computational and data efficiency of MTP. Critically, the active learning strategy is shown to reduce the required number of ab initio calculations by an order of magnitude without sacrificing model accuracy. Application of these potentials confirms that Nb addition enhances ductility while decreasing the strength of TiAl phases, providing a clear mechanistic link through simulated tension tests and the analysis of stacking fault energies. This research thus makes three principal contributions: it delivers a quantitative and mechanistic understanding of Niobium's influence on the thermomechanical properties of TiAl-based alloys; it provides a set of rigorously validated, high-fidelity machine learning (ML) potentials for the TiAlNb system; and it presents a detailed methodological guide for the efficient construction of such potentials via active learning. The benchmark datasets are made publicly available to accelerate the data-driven design of high-performance materials.
DDC Class
660: Chemistry; Chemical Engineering
Publication version
acceptedVersion