Artificial intelligence-driven defect detection and localization in metal 3D printing using convolutional neural networks

Metal additive manufacturing (AM) has attracted significant interest in high-value industries due to its ability to produce complex parts flexibly, but the reliance on costly manual monitoring remains a major burden for quality control. Artificial intelligence (AI)-driven models for automated defect detection are emerging as promising solutions. This study contributes a new annotated dataset for AI research in AM and evaluates the performance of four widely used convolutional neural network (CNN) models in detecting powder bed morphology defects, based on layer-wise images acquired by the EOSTATE PowderBed system during the metal laser-based powder bed fusion process. The models were trained through transfer learning methods with manually labeled and pre-processed data. Results demonstrated that ResNet50 and EfficientNetV2B0 achieved over 99% accuracy in defect classification, while YOLOv5 outperformed Faster region-based-CNN in defect detection and localization. However, lower average precision values in object detection tasks were attributed to variability in defect scales and annotation quality. This study confirms the potential of AI-based models for defect identification in AM, with YOLOv5 demonstrating clear advantages in managing complex, multi-scale defects. Future improvements will focus on expanding the dataset and refining annotation strategies to further enhance model robustness.