Real-Time Vision-Based Fault Detection System for FDM 3D Printing Using Convolutional Neural Networks

We present a real-time vision-based fault detection system for Fused Deposition Modeling (FDM) 3D printing, utilizing multiple You Only Look Once (YOLO) Convolutional Neural Network (CNN) models-YOLOv4 Tiny, YOLOv5, YOLOv8, and YOLOv10-to identify defects such as blobs, cracks, spaghetti, stringing, under-extrusion, and warping. Unlike previous studies that use single models, our system integrates these YOLO variants to enhance detection accuracy and incorporates an adaptive learning module for continuous improvement based on real-time data. The system logs and displays faults on a live camera feed, significantly improving quality control in additive manufacturing. Comparative analysis shows that YOLOv8 achieves a 7% increase in detection accuracy and a 30% reduction in print errors. This novel integration of multiple YOLO models with adaptive learning advances automation and reliability in 3D printing.