Machine Learning Predictive Model Applied to Detect Spaghetti Flaws in 3D Printing

The Industry 4.0 concept aims at greater flexibility in manufacturing, mass customization, higher quality, and improved productivity. These concepts can be achieved through three- dimensional (3D) printing. This type of process is still subject to several flaws, so developing a warning system for this event is fundamental. The same printer can produce different objects, making development of a general alarm system difficult as previous researchers focused on data collection of vibration or temperature of 3D printers. This work aims to create a computer vision model capable of detecting Spaghetti flaws in fused deposition modeling (FDM) 3D printing. More than 3,000 images of printing processes with this kind of failure were collected and labeled, and more than 10 models based on the YOLOV5 architecture, employing a single convolutional neural network (CNN) that divides the image into a grid, were trained using deep learning models. The best-performing model was tested against a test dataset classifying completely new images. The model’s performance achieved approximately 73% accuracy.