Hot‐Tearing Mechanisms and a Machine‐Learning‐Assisted Criterion for Fracture‐Mode Discrimination in 42CrMo Steel

Hot tearing in 42CrMo steel is governed by the coupled evolution of thermal–mechanical loading, solidification, and the resistance of the semisolid skeleton, and its fracture response depends on both process conditions and the evolving semisolid structure. In this work, a mechanism‐based framework was established by combining dynamic‐loading thermal simulation, in situ confocal observation, fracture‐surface analysis, cellular‐automaton reconstruction, and an interpretable threshold criterion. The results show that increasing strain rate shifts crack propagation to earlier solidification stages, as reflected by a higher crack‐initiation temperature and a lower corresponding solid fraction, while the crack‐initiation stress varies only within a narrow range. In situ observation and fractography reveal a transition from metallurgical‐bridge‐dominated resistance to liquid‐film‐ and mechanical‐bridge‐dominated crack propagation. CA‐derived descriptors quantify this transition through the extent and connectivity of the solid–liquid interfacial network. On this basis, a machine‐learning‐assisted threshold criterion was established to discriminate brittle and nonbrittle fracture modes. The full model achieved a leave‐one‐out cross‐validation accuracy of 0.95, decreasing to 0.85 and 0.80 after removing connectivity‐related and interfacial‐fraction descriptors. These results show that hot tearing is controlled by the interaction between process variables, interdendritic connectivity, and bridge‐controlled crack resistance.