Abd Latib, Muhammad Lut Liwauddin (2023) Development of under extrusion dataset using modified gcode in advance manufacturing of FDM 3D printer. Masters thesis, Universiti Tun Hussein Onn Malaysia.

Preview	Text 24p MUHAMMAD LUT LIWAUDDIN ABD LATIB.pdf Download (708kB) | Preview
	Text (Copyright Declaration) MUHAMMAD LUT LIWAUDDIN ABD LATIB COPYRIGHT DECLARATION.pdf Restricted to Repository staff only Download (476kB) | Request a copy
	Text (Full Text) MUHAMMAD LUT LIWAUDDIN ABD LATIB WATERMARK.pdf Restricted to Registered users only Download (21MB) | Request a copy

Abstract

Failures in 3D printers are common in the manufacturing industry, leading to wastage of materials and time. This thesis addresses the challenge of automatically detecting under-extrusion failures in Fused Deposition Material (FDM) 3D printers using training techniques. Existing studies primarily focus on stringing failure detection, disregarding the impact of extrusion rate on printer failures. Under-extrusion failures can result from insufficient extrusion speed, low melting temperature, nozzle blockage, or worn-out extruder gear, rendering printed models unusable. To overcome this issue, a training method based on the YOLOv5 models are proposed. A dataset comprising 2400 images is created by modifying and augmenting 200 under-extrusion samples. Raspberry Pi and a camera are employed to gather data on under-extrusion samples in real-time. The YOLOv5x model is trained on Google Colab, offering enhanced precision, recall, and mean average precision (mAP) compared to previous models. The experimental results demonstrate the effectiveness of the YOLOv5 training models in detecting under-extrusion failures. The YOLOv5x model achieves 99% precision, 100% recall, and 99% mAP, outperforming other models. This research contributes to the intelligence of 3D printers by reducing waste materials and time lost due to failed prints. Furthermore, the proposed training approach can be extended to identify other types of 3D printing failures

Actions (login required)

View Item