3D Printing Failure Detection, A Machine Learning Extension Architecture

Additive manufacturing is experiencing a surge in popularity, with affordable 3D printers utilizing Fused Deposition Modelling (FDM) technology becoming commonplace among hobbyists. While once a niche technology, these printers are now accessible at reduced costs. However, the quality and reliability of 3D-printed objects are susceptible to numerous variables, posing risks of harm, material wastage, and time loss in case of failures. Addressing this, the integration of computer vision and machine learning for failure detection is crucial. Existing 3D print failure datasets face limitations in detecting failures in diverse environments and lighting conditions. To overcome this, our proposed architecture framework enhances modularity within applications, enabling customization for machine learning models and datasets. This allows end-users to train the model with their specific printing environment dataset and seamlessly integrate it with currently available control software. Our architecture aims to bridge the gap between theoretical concepts and real-world applications, ultimately improving the overall reliability and effectiveness of 3D printing technology. In experimental trials, the detection model trained with a custom dataset demonstrated superior performance. Using the same printing configuration, the pretrained model achieved 100^% stringing test failure detection, 80% for boats, 42.86% for name plates, and 12.5% for action figures. With additional custom data, detection rates improved to 100% for stringing tests, 90% for boats, 71.43% for name plates, and 62.5% for action figures.