Abstract: Considering the limitations of traditional non-destructive testing techniques in terms of cost and identification accuracy, and to better ensure the safety of engineering structures, this study systematically investigates structural internal defect inversion methods that reverse-engineer internal structural defects (such as cracks, voids and inclusions) from observational data, aiming to clarify their research progress, technical characteristics, and future directions. The research progress of structural internal defect inversion methods was reviewed from three aspects: optimization methods based on objective function iteration, data-driven methods based on machine learning, and physics-data dual-driven methods, analyzing the technical advantages and application limitations of each approach. The research demonstrates that optimization methods based on objective function iteration have broad theoretical applicability but suffer from low computational efficiency and susceptibility to local optima. The data-driven methods possess advantages of fast identification speed and high accuracy, but depend on large amounts of labeled data with limited generalization capability. The dual-driven methods can effectively integrate physical knowledge with data advantages and show promising development prospects. Based on current technological trend analysis, the future of this field will continue to enhance accuracy and reliability to strengthen real-time monitoring capabilities for various structures.