Metal magnetic memory testing (MMMT) is a nondestructive testing technique that can detect early signs of damage in components. Many scholars have studied the effect of uniaxial stress on the self-magnetic-leakage field (SMLF)’s strength. Nevertheless, there is still insufficient research on the combined action of bending and shear. We studied the law of distribution of the magnetic signal,$\Delta H_{\mathrm{SF}}\left(\mathrm{y}\right)$, at different stress parts of a steel box girder and the quantitative relationship between the magnetic characteristic parameters and the external load. The results showed that the MMMT could accurately detect the early stress concentration zone (SCZ) and predict the final buckling zone of steel box girders. It could be judged that the corresponding parts of the steel box girder had entered the elastic-plastic working stage by the reverse change of the $\Delta H_{\mathrm{SF}}\left(\mathrm{y}\right)$-F and${\left|H_{\mathrm{SF}}\left(\mathrm{y}\right)\right|}_{a }$-F curve trends, this feature could be used as an early warning sign before the steel box girder was deformed or destroyed. The fitted${\left|H_{\mathrm{SF}}\left(\mathrm{y}\right)\right|}_{ave }$-F linear expression could be used as the expression between the magnetic signal and the shear capacity. All the evaluation methods were expected to provide a basis for effectively evaluating the stress state of steel box girders with the MMMT method.

It is valuable to conduct non-destructive testing of steel box girders in order to evaluate their working status. The metal magnetic memory inspection method can effectively identify early damage and the stress state of ferromagnetic materials. However, applying this technique in the inspection of steel components is difficult due to insufficient research on magnetic memory signals under complex stress states. This study analyses the magnetic memory effect for a steel box girder under four-point bending. It is shown that the normal component H_p(y) of the magnetic signal can effectively locate the stress concentration area. The average absolute value H_m of H_p(y) can identify the yielding state and predict the occurrence of failure. H_m changes roughly quadratically with the average strain ɛ_m in the elastic stage of the specimen, which is consistent with the theoretical result. The ratio D of H_m to the equivalent stress σ_eqv changes continuously with the applied load, which can be used to estimate the stress state of the steel box girder.

Melting of crystalline material is a common physical phenomenon, yet it remains elusive owing to the diversity in physical pictures. In this work, we proposed a deep learning architecture to learn the physical states (solid- or liquid-phase) from the atomic trajectories of the bulk crystalline materials with four typical lattice types. The method has ultra-high accuracy (higher than 95%) for the classification of solid–liquid atoms during the phase transition process and is almost insensitive to temperature. The atomic physical states are identified from atomic behaviors without considering any characteristic threshold parameter, which yet is necessary for the classical methods. The phase transition of bulk crystalline materials can be correctly predicted by learning from the atomic behaviors of different materials, which confirms the close correlation between atomic behaviors and atomic physical states. These evidences forecast that there should be a more general undiscovered physical quantity implicated in the atomic behaviors and elucidate the nature of bulk crystalline melting.