Machine-learning interatomic potentials (MLIPs) offer a powerful avenue for simulations beyond length and timescales of ab initio methods. Their development for investigation of mechanical properties and fracture, however, is far from trivial since extended defects—governing plasticity and crack nucleation in most materials—are too large to be included in the training set. Using TiB₂ as a model ceramic material, we propose a training strategy for MLIPs suitable to simulate mechanical response of monocrystals until failure. Our MLIP accurately reproduces ab initio stresses and fracture mechanisms during room-temperature uniaxial tensile deformation of TiB₂ at the atomic scale ( ≈ 10³ atoms). More realistic tensile tests (low strain rate, Poisson’s contraction) at the nanoscale ( ≈ 10⁴–10⁶ atoms) require MLIP up-fitting, i.e., learning from additional ab initio configurations. Consequently, we elucidate trends in theoretical strength, toughness, and crack initiation patterns under different loading directions. As our MLIP is specifically trained to modelling tensile deformation, we discuss its limitations for description of different loading conditions and lattice structures with various Ti/B stoichiometries. Finally, we show that our MLIP training procedure is applicable to diverse ceramic systems. This is demonstrated by developing MLIPs which are subsequently validated by simulations of uniaxial strain and fracture in TaB₂, WB₂, ReB₂, TiN, and Ti₂AlB₂.

机器学习原子间势 (MLIP) 为超越从头计算方法的长度和时间尺度的模拟提供了强大的途径。然而，他们对机械性能和断裂研究的开发绝非微不足道，因为大多数材料中控制塑性和裂纹成核的扩展缺陷太大，无法包含在训练集中。使用 TiB ₂作为模型陶瓷材料，我们提出了一种适用于模拟单晶直至失效的机械响应的 MLIP 训练策略。我们的 MLIP在原子尺度（约 10 ^{3 个原子）下准确地再现了 TiB} ₂室温单轴拉伸变形过程中的从头开始应力和断裂机制。纳米级（约 10 ⁴ –10 ⁶原子）更真实的拉伸测试（低应变率、泊松收缩）需要 MLIP 向上拟合，即从额外的从头开始配置进行学习。因此，我们阐明了不同载荷方向下理论强度、韧性和裂纹萌生模式的趋势。由于我们的 MLIP 经过专门训练来模拟拉伸变形，因此我们讨论了其在描述不同负载条件和具有各种 Ti/B 化学计量的晶格结构方面的局限性。最后，我们表明我们的 MLIP 培训程序适用于不同的陶瓷系统。这通过开发 MLIP 得到证明，随后通过模拟 TaB ₂、WB ₂、ReB ₂、TiN 和 Ti ₂ AlB ₂中的单轴应变和断裂来验证该 MLIP 。