Çѱ¹Åë½ÅÇÐȸ ³í¹®Áö (The Journal of Korea Information and Communications Society)
Current Result Document : 1,795 / 1,795
ÇѱÛÁ¦¸ñ(Korean Title) |
´ÙÁß ºí·Ï 2D-ÄÁº¼·ç¼Ç ½Å°æ¸ÁÀ» ÀÌ¿ëÇÑ È¿À²ÀûÀÎ 3D ÇÁ¸°ÅÍ Ãâ·Â °áÇÔ ºÐ·ù ±â¹ý |
¿µ¹®Á¦¸ñ(English Title) |
Efficient 3D Printer Fault Classification Using a Multi-Block 2D-Convolutional Neural Network |
ÀúÀÚ(Author) |
Made Adi Paramartha Putra
Ahakonye Love Allen Chijioke
Mark Verana
Dong-Seong Kim
Jae-Min Lee
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 47 NO. 02 PP. 0236 ~ 0245 (2022. 02) |
Çѱ۳»¿ë (Korean Abstract) |
º» ³í¹®¿¡¼´Â ½Å¼ÓÇÑ Ãß·Ð ½Ã°£°ú ³·Àº °è»ê º¹Àâµµ¸¦ °¡Áö´Â µö·¯´×(DL: deep learning) ¸ðµ¨À» »ç¿ëÇÏ¿© 3D ÇÁ¸°ÅÍ Ãâ·Â °úÁ¤¿¡¼ ¹ß»ýÇÏ´Â °áÇÔÀ» ºÐ·ùÇÏ´Â »õ·Î¿î ¹æ¹ýÀ» Á¦¾ÈÇÑ´Ù. Á¦¾ÈÇÏ´Â ºÐ·ù ±â¹ýÀº 3D ÇÁ¸°ÅÍ °áÇÔÀ» ºÐ·ùÇϱâ À§ÇØ ´ÙÁß ºí·Ï 2D-ÄÁº¼·ç¼Ç ½Å°æ¸Á(CNN: convolution neural network)À» »ç¿ëÇϰí CNN ºí·ÏÀ» »ç¿ëÇÏ¿© fused deposition modeling(FDM) 3D ÇÁ¸°ÅÍ¿¡¼ ¼öÁýµÈ À̹ÌÁö µ¥ÀÌÅÍ ¼¼Æ®¿¡¼ Ư¡À» ÃßÃâÇÑ´Ù. Á¦¾ÈµÈ ¸ðµ¨Àº MobileNet, AlexNet, VGG-11, VGG-16°ú °°Àº ±âÁ¸ À̹ÌÁö ºÐ·ù ¾Ë°í¸®Áò°ú ºñ±³ÇÏ¿© ¼º´ÉÀ» Æò°¡ÇÏ¿´´Ù. Á¦¾ÈµÈ ´ÙÁß ºí·Ï CNNÀº ±âÁ¸ ±â¹ý ´ëºñ Ã߷нð£ÀÌ 67.01% »¡¶óÁ³À¸¸ç 87.56% ³·Àº ¸Þ¸ð¸® »ç¿ë·®°ú ÃÖ´ë 9.36%±îÁö °¨¼Ò½ÃŲ ¸Å°³º¯¼ö¸¦ »ç¿ëÇÔ¿¡µµ ³ôÀº Á¤È®µµ¸¦ º¸¿©ÁØ´Ù. ÀÌ·¯ÇÑ ¼º´ÉÀº Á¦¾ÈµÈ 3D ÇÁ¸°ÅÍ °áÇÔ ºÐ·ù ¸ðµ¨ÀÌ ½Ç½Ã°£ ¸ð´ÏÅ͸µ Á¶°Ç¿¡¼ Á¤È®ÇÑ ºÐ·ù¸¦ Á¦°øÇϴµ¥ ÀûÇÕÇÔÀ» º¸¿©ÁØ´Ù. |
¿µ¹®³»¿ë (English Abstract) |
This paper proposes a novel fault classification method with an efficient deep learning (DL) model with fast inference time and lower computational complexity during the 3D printer printing process. Specifically, a multi-block 2D-convolutional neural network (CNN) is used to classify the 3D printer fault. In the proposed method, blocks of CNNs are used to extract the features from an image dataset that is gathered with a FDM 3D printer type. The performance evaluation of the proposed model is compared with existing image classification algorithms, such as MobileNet, AlexNet, VGG-11, and VGG-16. The results show that the proposed multi-block CNN classification model yields high accuracy with 67.01% faster inference time, 87.56% lower memory usage, and lower trainable parameters up to 93.36%. Furthermore, the proposed 3D model can provide an accurate classification in real-time monitoring conditions. |
Ű¿öµå(Keyword) |
3D Printing
CNN(convolutional neural network)
Efficient model
Fault detection
Manufacturing
|
¿ø¹® |
PDF ´Ù¿î·Îµå
|