In this paper, the corrosion environment of a railway coal truck was simulated with 1.0%H₂SO₄ + 3%NaCl solution. The effect of weld toe Tungsten Inert Gas (TIG) remelting on the microstructure and corrosion resistance of welded joints of Q450NQR1 high-strength weathering steel was studied. The results show that the weld toe melts to form a remelting area after TIG remelting. After TIG remelting, the weld geometry was improved, and the stress concentration factor decreased from 1.17 to 1.06 at the weld toe, a decrease of 9.4%. TIG remelting refines the microstructure of the weld toe and improves the corrosion resistance of the welded joint. The surface of the TIG-remelted sample is uniformly corroded with no “deep and narrow” pits after the removal of corrosion products. The weight loss rate and corrosion rate of remelted welds are lower than those of unremelted welds. The structure of corrosion products is loose at the initial stage of corrosion, and the corrosion products are transformed into Fe₃O₄ and Fe₂O₃ protective rust layers with a dense structure after 480 h of corrosion. With the extension of corrosion time, the tensile strength and percentage elongation of the specimen decreased linearly. The decreasing rates of tensile strength of remelted and unremelted specimens were 0.09 and 0.11, respectively, and the decreasing rates of elongation after fracture were 0.0061 and 0.0076, respectively.

Numerical simulation of metal microstructure evolution is essential for material design and performance optimization. This paper reviews major simulation methods for key evolution mechanisms, including recrystallization, grain growth, slip, twinning, and phase transformation. The reviewed methods are classified into atomistic models, discrete-field models, and continuous-field models. Molecular dynamics (MD) is discussed as an independent atomistic approach, with emphasis on its role in revealing atomic-scale mechanisms, calibrating mesoscale parameters, and bridging atomistic, mesoscale, and continuum simulations. Discrete-field methods, including Monte Carlo, cellular automata, and vertex models, are compared with continuous-field methods, including artificial neural networks, phase field models, finite element methods, and level-set methods. Furthermore, a semi-quantitative evaluation matrix based on accuracy, computational cost, scalability, and applicability is established to clarify the practical trade-offs among different methods. The results show that no single method is universally optimal; instead, method selection should depend on the dominant physical mechanism, target length scale, required accuracy, and available computational resources. This review provides methodological guidance for multiscale microstructure simulation and supports future applications in precision machining, additive manufacturing, and process parameter optimization.