Predicting Mode-I/II fracture toughness and crack growth in diboride ceramics via machine-learning potentials
Predicting Mode-I/II fracture toughness and crack growth in diboride ceramics via machine-learning potentials
Fracture toughness and strength are critical for structural ceramics, which are prone to brittle failure. However, accurately characterizing these properties is challenging, especially for thin films on substrates. In-situ microscopy often fails to resolve crack initiation, while fractured samples provide limited insight into fracture modes and crack paths. Here, we employ stress intensity factor ($K$) controlled atomistic simulations of fracture to characterize the crack-initiation properties of hard but brittle diboride ceramics. Our molecular statics calculations are based on moment-tensor machine-learning interatomic potentials (MLIPs) trained on {\it{ab initio}} information collected for a variety of atomistic environments. TMB$_{2}$ (TM$=$Ti, Zr, Hf) lattice models with six distinct atomically-sharp crack geometries subjected to Mode-I (opening) and/or Mode-II (sliding) deformation serve as platforms to illustrate the capability of the approach. The Mode-I initiation toughness $K_{Ic}$ and fracture strength $\sigma_{f}$ -- within ranges 1.8-2.9~MPa$\cdot\sqrt{m}$ and 1.6-2.4~GPa -- are extrapolated at the macroscale limit by fitting the results of finite (up to 10$^{6}$ atoms) cracked plate models with constitutive scaling relations. Our simulations show that most diboride lattice models fail by extension of the native crack. However, crack-deflection on the $(1\overline{1}01)$ plane is observed for the $(10\overline{1}0)[\overline{1}2\overline{1}0]$ crystal geometry. As exemplified by TiB$_{2}$, varying Mode-I/II loading ratios have little influence on crack propagation paths, which overall occurs by decohesion of low-energy fracture planes or combined sliding. Our predictions are supported by cube-corner nanoindentation on TiB$_{2}$ samples along the [0001] direction, revealing the same fracture plane as observed in simulations.
Shuyao Lin、Zhuo Chen、Rebecca Janknecht、Zaoli Zhang、Lars Hultman、Paul H. Mayrhofer、Nikola Koutna、Davide G. Sangiovanni
材料科学
Shuyao Lin,Zhuo Chen,Rebecca Janknecht,Zaoli Zhang,Lars Hultman,Paul H. Mayrhofer,Nikola Koutna,Davide G. Sangiovanni.Predicting Mode-I/II fracture toughness and crack growth in diboride ceramics via machine-learning potentials[EB/OL].(2025-03-23)[2025-04-28].https://arxiv.org/abs/2503.18171.点此复制
评论