µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
Attention U-Net ½Å°æ¸ÁÀ» È°¿ëÇÑ À¯Ã¼ÀÇ ¹Ì·¡ »óÅ ¿¹Ãø ±â¹ý |
¿µ¹®Á¦¸ñ(English Title) |
A method for predicting the future state of fluids using Attention U-Net neural networks |
ÀúÀÚ(Author) |
ÀÌâ°Ç
±èÀÏÁÖ
Changgeon Lee
Iljoo Kim
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 37 NO. 12 PP. 0003 ~ 0015 (2021. 12) |
Çѱ۳»¿ë (Korean Abstract) |
Àü»ê À¯Ã¼ ¿ªÇÐ ½Ã¹Ä·¹À̼ÇÀº Ç×°ø±â, °Ç¹°, ÀÚµ¿Â÷ µî À¯Ã¼¿Í °ü·ÃµÈ ´Ù¾çÇÑ µðÀÚÀÎ ºÐ¾ß¿¡¼ È°¿ëÀÌ µÇ¾îÁö°í ÀÖÀ¸³ª, ¿À·£ ½ÇÇà ½Ã°£°ú ¸¹Àº ºñ¿ëÀÌ ¹ß»ýÇÏ´Â ³ªºñ¿¡-½ºÅäÅ©½º ¹æÁ¤½ÄÀÇ »ç¿ëÀ¸·Î ÀÎÇØ °³¹ß¿¡ ¸¹Àº ½Ã°£ÀÌ ¼Ò¿äµÈ´Ù. ¶ÇÇÑ À§Çè ¹°ÁúÀÌ È®»êµÇ´Â ȯ°æ°ú °°ÀÌ Áï°¢ÀûÀÎ À¯Ã¼ È帧ÀÇ ºÐ¼®ÀÌ ÇÊ¿äÇÑ ºÐ¾ß¿¡¼´Â È°¿ëÀÌ ¾î·Æ±â ¶§¹®¿¡ ÃÖ´ëÇÑ Á¤È®µµ¸¦ º¸Á¸ÇÏ¸é¼ ºü¸£°Ô ¿¹ÃøÇÏ´Â ±â¼úÀÌ ¸Å¿ì Áß¿äÇÏ°Ô ¿©°ÜÁö°í ÀÖ´Ù.
À̸¦ À§ÇØ º» ¿¬±¸¿¡¼´Â Attention U-Net ½Å°æ¸ÁÀ» ±â¹ÝÀ¸·Î À¯Ã¼ÀÇ È帧À» ¿¹ÃøÇÏ´Â ¹æ¹ý·ÐÀ» Á¦¾ÈÇÏ°í Å×½ºÆ®ÇÑ´Ù. Àå¾Ö¹°À» ÀνÄÇÏ¿© ºü¸£°Ô 7ÃÊ ÈÄ ¹Ì·¡ÀÇ À¯Ã¼ È帧À» ¿¹ÃøÇÏ´Â ¹æ¹ýÀ¸·Î Å×½ºÆ®¸¦ ÁøÇàÇÏ¿´°í, ±× °á°ú ±âÁ¸ ½Ã¹Ä·¹ÀÌ¼Ç ¼ÒÇÁÆ®¿þ¾î ¹× CNN¸ðµ¨À» »ç¿ëÇÏ¿´À» ¶§ ´ëºñ Á¤È®µµ´Â ÃÖ´ëÇÑ º¸Á¸ÇÏ¸é¼ ½ÇÇà¼Óµµ´Â ¾à 85¹è ºü¸¥ °á°ú¸¦ ¾òÀ» ¼ö ÀÖ¾ú´Ù. |
¿µ¹®³»¿ë (English Abstract) |
Computational Fluid Dynamics (CFD) simulation is used in various fluid-related fields such as aircrafts, buildings, or automobiles design and it consumes a lot of development time due to the use of Navier Stokes equation, which incur a long execution time with high cost. In addition, it is difficult to use it in such environments as the hazardous substance or epidemic diffusion where the fluid flow must be predicted immediately. In the situations, it is important and critical to predict as quickly as possible while preserving accuracy. Hence in this study, to address the issues, we suggest and test a method that uses Attention U-Net neural network to predict the future state of the fluid in a terrain with obstacles. As a result, compared to the existing simulation software or simple CNN method, the suggested approach shows the execution speed is 85 times faster while preserving the competitive accuracy.
|
Å°¿öµå(Keyword) |
µö·¯´×
ÇÕ¼º°ö ½Å°æ¸Á
Àü»ê À¯Ã¼ ¿ªÇÐ
³ªºñ¿¡-½ºÅäÅ©½º ¹æÁ¤½Ä
À¯Ã¼ ½Ã¹Ä·¹À̼Ç
Computational fluid dynamics
Navier Stokes equation
deep learning
convolutional neural network
fluid simulation
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|