An In-Situ Deep Learning-Based Defect Detection Technology for Additive Manufacturing Process

A trend in the additive manufacturing (AM) area is to deliver full-functional end-use products rather than discrete parts. Therefore, multiple manufacturing processes, additive or traditional, need to be combined to enable complex structure building, various material processing, and end-use functionality actualization. 3D electronics is a type of important product that such AM technologies can produce. However, the whole manufacturing procedure is much more complex than ever before, and any defect that occurs during this period may fail the product. To address the challenge, this paper proposes a deep learning-based defect detection technology to monitor the AM fabrication procedure in situ. An improved YOLOv8 algorithm is developed to train the defect detection model to identify and evaluate the defect image. A dataset of 3550 defects in four categories was created by selecting typical defects of the extrusion 3D printing process to testify to the practicability of this method. Experimental results showed that the improved YOLOv8 model achieves a mean average precision (mAP50) of 91.7% with a frame rate of 71.9 frames per second. In the future, this technology can be deployed on AM apparatus for real-time quality monitoring during the fabrication period. Moreover, depending on the detection results, printing settings can be adjusted real-timely to enhance the success rate of the printing process.