Speaker
Description
Unique electromagnetic properties of steels with silicon and aluminum make them an excellent choice for the production of transformers and electric motors. These so-called electrical steels have enhanced energy efficiency due to reduced core losses and enhanced magnetic permeability. However, increasing the concentration of silicon and aluminum beyond a critical amount leads to significantly reduced ductility, which makes production very challenging.
The origins of these embrittlement effects from a first-principles perspective are still not fully understood, as ab initio modeling based on conventional Density Functional Theory (DFT) is computationally very expensive, due to the requirement of large supercells, for the treatment of disorder, and complex magnetic interactions. While bulk configurations can also be described by the Coherent Potential Approximation (CPA), real systems are never perfectly symmetric crystals, where in particular planar defects such as stacking faults are known to have a significant impact on mechanical properties of materials.
However, accuracy comparable to DFT can be achieved through the application of Machine Learned Interatomic Potentials (MLIPs). Using the recently developed workflow manager Autopot [1] we train Moment Tensor Potentials (MTPs) for the Fe-rich region of the ternary Fe-Si-Al system, containing both bulk configurations, and configurations containing a stacking fault. This potential is then used to predict properties relevant for the study of dislocation plasticity, namely lattice constants, elastic constants, and unstable stacking faults. Our work shows that an MTP-based workflow of this kind can pave the way for the description of failure mechanisms in multicomponent alloys.