期刊
RELIABILITY ENGINEERING & SYSTEM SAFETY
卷 241, 期 -, 页码 -出版社
ELSEVIER SCI LTD
DOI: 10.1016/j.ress.2023.109639
关键词
Blast loading; BLEVE; Pressure-time history; Machine learning; Graph neural networks
This paper proposes an innovative machine learning approach using Graph Neural Networks (GNNs) to predict overpressure wave propagation generated by Boiling Liquid Expanding Vapor Explosion (BLEVE). The results show that GNN can accurately predict BLEVE overpressure wave propagations with significantly less computational effort compared to traditional methods. The GNN model demonstrates higher temporal resolution and superior generalizability to unseen data.
In this paper, we propose an innovative machine learning approach for predicting overpressure wave propagation generated by Boiling Liquid Expanding Vapor Explosion (BLEVE) using Graph Neural Networks (GNNs). The accurate prediction of BLEVE overpressure wave propagation is critical for effective risk assessment, mitigation, and emergency response planning. Traditional simulation methods, such as Computational Fluid Dynamics (CFD), provide comprehensive insights into BLEVE phenomena but often pose significant computational demands, thus challenging for real-time or large-scale applications. While existing machine learning models have demonstrated efficiency and accuracy in overpressure prediction, they fall short in providing full-field spatiotemporal predictions of pressure wave propagations, essential for comprehensive blast simulations. Our GNNbased approach addresses these limitations by leveraging the micro-level representation learning capabilities of GNNs with an autoregressive prediction scheme. The results from numerical data show that GNN can predict BLEVE overpressure wave propagations accurately and with significantly less computational effort compared to traditional CFD simulations. Compared with existing machine learning models, GNN attains much higher temporal resolution in pressure-time history prediction, while maintaining comparable accuracy. Moreover, the GNN model demonstrates superior generalizability to unseen data when input parameters are extrapolated from the training range. This research highlights the potential of GNNs as a promising advancement in blast loading prediction, providing a more efficient and effective risk management strategy to enhance the reliability and safety of blast-related systems.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据