Application of a GAN model incorporating transfer learning methods for the detection of forming defects in metallic materials

Defect detection is essential for quality assurance in metal forming, but traditional methods often suffer from process complexity and imbalanced datasets. This study proposes a GAN-based defect detection framework integrating transfer learning to improve accuracy and robustness. A pre-trained GAN is first employed to augment non-defective metal samples, enhancing data diversity, followed by transfer learning-based defect classification. Experimental results demonstrate that the improved auxiliary classifier GAN achieves an average classification accuracy of 95.98%, outperforming traditional auxiliary classifier GAN (91.08%) and ResNet-50 (88.65%). Moreover, the proposed model obtains a Mode Score of 1.386, higher than Res-Wasserstein GAN (1.374) and Res-ACGAN (1.361), indicating improved quality and diversity of generated samples. The model is further applied to defect detection of aluminum profiles and hot-rolled steel strips, where it effectively identifies and classifies various defect types with enhanced detection efficiency and reliability. These results demonstrate that the proposed AI-driven approach can improve intelligent quality control in metal processing, reduce potential safety risks caused by undetected defects, and provide an effective solution for advanced manufacturing applications.