화학공학소재연구정보센터
학회 한국화학공학회
학술대회 2019년 봄 (04/24 ~ 04/26, 제주국제컨벤션센터)
권호 25권 1호, p.1042
발표분야 화학공정안전(Process Safety)
제목 Consequence modeling for Real-Time Analysis Using Generative Adversarial Networks
초록 We suggest a way to determine the ragne of damage from high-risk materials. There are many flammable and toxic substances following numerous pipelines in petrochemical complexes. Some consequence modeling are used rigorous equations, but they had low accuracy results. One of modeling methods is Computational Fluid Dynamices(CFD) to calculate release range of various fluids or gases by each scenario. It has high accuracy results, but long time to show result from simulation set.  In this study, we propose variational autoencoder with generative adversarial network (VAE-GAN) for real-time analysis of probability of death. The acrylonitrile is released from underground pipe is main scenario in this study, the result is shown different damage following hole size, mass flowrate and wind condition such as direction or speed. We occur damage result faster than CFD model by data memorization. The proposed model compared mean square error with other models.  
저자 계혜리1, 나종걸2, 이철진1
소속 1중앙대, 2kist
키워드 위험성평가
E-Mail
원문파일 초록 보기