Fused Deposition Modeling (FDM) is a widely preferred additive manufacturing technology due to its design flexibility and cost advantages. However, structural and visual defects such as stringing, layer shifting, warping, or off-platform issues frequently occur in FDM-based three-dimensional (3D) printing processes. This results in losses of raw materials, energy, and time. This study aims to classify common defect types occurring in FDM-based 3D printing using image-based deep learning algorithms. In this study, a transfer learning approach is adopted using the EfficientNetB0 and MobileNetV2 architectures, both pre-trained on ImageNet, with a dataset of 1912 images. The model’s generalization capability is enhanced through data preprocessing and enhancement techniques. The model can successfully classify defect types using print images without additional sensors or hardware components. The results show that the EfficientNetB0 model has an overall accuracy of 87.7%, while MobileNetV2 achieves 97%. The MobileNetV2 architecture demonstrates strong performance, particularly in error classes such as Layer Shifting and Stringing, with high F1 scores. The proposed architectures have the potential to reduce losses in FDM-based 3D printing processes by providing a low-cost and accessible visual quality control system.
Aktepe et al. (Tue,) studied this question.