Emerging Progress and Feasibility Evaluation of Residual Stress Research in Heterogeneous Brazed Joints: Focus on Machine Learning-Integrated Numerical Simulation

Residual stress plays a decisive role in the mechanical performance, structural reliability, and service life of heterogeneous brazed joints. Owing to the pronounced mismatch in thermophysical properties between dissimilar materials, together with complex interfacial reactions and non-uniform cooling during brazing, residual stresses in these joints are typically highly heterogeneous and difficult to evaluate accurately. Although destructive and non-destructive techniques have been developed, their capability remains limited for full-field, high-resolution, and reliable stress characterization, particularly in narrow interfacial regions with steep stress gradients. In this context, this review provides a systematic and forward-looking assessment of residual-stress detection, prediction, and regulation in heterogeneous brazed joints, with particular attention to the prospective integration of machine learning and numerical simulation. It should be noted that a fully closed-loop framework integrating machine learning, numerical simulation, and brazing-specific residual-stress regulation has not yet been systematically established. Therefore, this review does not treat the integrated framework as a mature and fully validated paradigm. Instead, it evaluates its feasibility by distinguishing direct evidence from heterogeneous brazing studies from transferable methodological evidence obtained from related joining, manufacturing, and materials-design fields. The origins and evolution characteristics of residual stress are first discussed, followed by a critical assessment of detection methods and the physical origins of their limitations. The applicability, data requirements, failure modes, and interpretability limits of machine learning are then examined. Particular attention is given to simulation-assisted data generation, physics-guided learning, transferability boundaries, uncertainty awareness, and future dataset construction. This review is expected to provide a realistic roadmap for developing intelligent, physics-constrained, and experimentally validated approaches for residual-stress prediction and regulation in heterogeneous brazed joints.