Parts | Days | Selection | Search | Updates | Downloads | Help

MM: Fachverband Metall- und Materialphysik

MM 38: Mechanical Properties and Alloy Design: e.g. Light-Weight, High-Temperature, Multicomponent Materials III (joint session MM/KFM)

MM 38.2: Talk

Wednesday, March 20, 2024, 12:00–12:15, C 230

Training strategies for machine-learning potentials suitable to simulate mechanical response of ceramics — •Shuyao Lin^1,2, Zhuo Chen³, Luis Casillas-Trujillo², Ferenc Tasnadi², Zaoli Zhang³, Lars Hultman², Paul H. Mayrhofer¹, Davide G. Sangiovanni², and Nikola Koutna^1,2 — ¹Institute of Materials Science and Technology, TU Wien, A-1060, Vienna, Austria — ²Department of Physics, Chemistry, and Biology (IFM), Linköping University, SE-58183, Linköping, Sweden — ³Erich Schmid Institute of Materials Science, Austrian Academy of Sciences, A-8700, Leoben, Austria

Machine-learning interatomic potentials (MLIPs) offer a powerful avenue for simulations beyond length and timescales of ab initio methods. In particular, MLIPs enable investigations of mechanical properties and fracture behaviour of materials with supercell sizes, loading geometries and temperatures relevant for real operation conditions. Using the example of hard TiB₂ ceramic, in this talk we propose a strategy for fitting MLIPs suitable to simulate mechanical response of monocrystals from atomic to nanoscale, including strains until fracture and deformation-induced phase transformations. After validation, the best-performing MLIP is employed to carry out molecular dynamics simulations of various loading conditions, with main focus on tensile and shear deformation. Consequently, we derive size-dependent trends in theoretical strength, toughness, and crack initiation patterns of TiB₂. To approach experimental observations, we additionally apply our MLIP to models containing a pre-crack and/or grain boundaries.

Keywords: Machine Learning Interatomic Potentials; Diborides; Ab initio Molecular Dynamics; Mechanical Properties; Size effect