Fault Detection in 3D-Printing with Deep Learning

Fused filament fabrication is a popular 3D printing process where layers of fusible plastic filament are used to build a workpiece. However, the high manual maintenance needs limit its potential. Given the common incidence of defects in the printed components, automated solutions detecting these errors while printing are of great benefit to every practitioner. Current research primarily focuses on monitoring and optimizing manufacturing processes and printing parameters or using additional sensors to directly observe the printing process and identify errors. However, there is no established procedure for assessing print quality in real time with minimal additional sensors. Our proposed computer vision-based deep-learning system overcomes previous limitations by detecting errors in 3D printing and assessing the quality of printed parts in real time using only a single camera. We record and provide a multi-class image dataset that encompasses different printed geometries, error classes, and printing condition variations. Our extensive evaluation shows that considering various geometries and printing conditions is vital for detecting printing errors. Our proposed computer vision-based deep-learning system enhances 3D printing by automating error detection, enabling practitioners to increase efficiency and print high-quality workpieces.