Real-time defect detection for FDM 3D printing using lightweight model deployment

Abstract

FDM 3D printing is one of the most widely used additive manufacturing methods, bringing great convenience to production manufacturing. However, various printing defects may occur during the printing process due to human factors or printer-related issues. Timely detection of defects and halting printing becomes a scenario of significant practical importance. This paper first analyzes the causes of the five most common defects in FDM 3D printing, and a defect dataset is created by deliberately designing defects. Subsequently, a real-time defect detection system for FDM 3D printing, based on an improved YOLOv8 detection head, is developed. By employing an optimization method using Group Convolution to share parameters, the detection head is lightweight, resulting in better model performance. Experimental results demonstrate that the mAP50 of the improved YOLOv8 model reaches 97.5%, with an 18.1% increase in FPS and a 32.9% reduction in GFLOPs. This enhancement maintains comparable detection accuracy to the original model while achieving faster detection speed and lower computational requirements. The improved model is integrated into the detection system as the detection model, and through testing, the real-time detection system promptly and accurately identifies and alerts any occurring defects. The practical significance of this system lies in its ability to enhance production efficiency, reduce resource wastage due to defective printing, and improve product quality and manufacturing safety, thereby providing strong support for the application of visual inspection technology in FDM 3D printing.