Three-Dimensional Printing Quality Inspection Based on Transfer Learning with Convolutional Neural Networks

Abstract

Fused deposition modeling (FDM) is a form of additive manufacturing where three-dimensional (3D) models are created by depositing melted thermoplastic polymer filaments in layers. Although FDM is a mature process, defects can occur during printing. Therefore, an image-based quality inspection method for 3D-printed objects of varying geometries was developed in this study. Transfer learning with pretrained models, which were used as feature extractors, was combined with ensemble learning, and the resulting model combinations were used to inspect the quality of FDM-printed objects. Model combinations with VGG16 and VGG19 had the highest accuracy in most situations. Furthermore, the classification accuracies of these model combinations were not significantly affected by differences in color. In summary, the combination of transfer learning with ensemble learning is an effective method for inspecting the quality of 3D-printed objects. It reduces time and material wastage and improves 3D printing quality.

Keywords

• fused deposition modeling
• image analysis
• quality inspection
• transfer learning
• ensemble learning