A clustering-based approach to vortex extraction
- PDF / 1,954,246 Bytes
- 16 Pages / 595.276 x 790.866 pts Page_size
- 100 Downloads / 226 Views
R E G UL A R P A P E R
Liang Deng • Yueqing Wang • Cheng Chen • Yang Liu • Fang Wang • Jie Liu
A clustering-based approach to vortex extraction
Received: 7 July 2019 / Revised: 21 August 2019 / Accepted: 28 October 2019 Ó The Visualization Society of Japan 2020
Abstract Since vortex is an important flow structure and has significant influence on numerous industrial applications, vortex extraction is always a research hotspot in flow visualization. This paper presents a novel vortex extraction method by employing a machine learning clustering algorithm to identify and locate vortical structures in complex flow fields. Specifically, the proposed approach firstly chooses an objective, physically based metric that describes the vortex-like behavior of intricate flow and then normalizes the metric for applying on different flow fields. After that, it performs the clustering on normalized metric to automatically determine vortex regions. Our method requires relatively few flow variables as inputs, making it suitable for vortex extraction in large-scale datasets. Moreover, this approach detects all vortices in the flow simultaneously, thereby showing great potential for automated vortex tracking. Extensive experimental results demonstrate the efficiency and accuracy of our proposed method in comparison with existing approaches. Keywords Vortex extraction Normalization Cluster analysis Unsteady flow fields
1 Introduction Vortex is one of the most important structures in fluid flows, which plays a key role in many engineering problems, such as fluid mixing, energy transport, chemical reaction, aerodynamic noise generation and emission, and drag reduction (Jeong and Hussain 1995). Due to the significant importance of vortex, diverse vortex extraction methods based on an associated implicit vortex definition have been intensively studied in recent years. These methods can be roughly divided into three categories: local methods (Chong et al. 1990; Hunt 1987; Jeong and Hussain 1995; Liu et al. 2016; Zhou et al. 1999), global methods (Serra and Haller 2016; Haller et al. 2015; Sadarjoen et al. 2002) and hybrid methods (Bin and Yi 2018; Biswas et al. 2015; Deng et al. 2019; Franz et al. 2018; Rajendran et al. 2018; Kim and Gu¨nther 2019; Lguensat et al. 2018; Stro¨fer et al. 2019; Zhang et al. 2014). The local methods are typically based on physical characteristics of fluid flows and are conceptually easy to explain, such as ‘locally high vorticity,’ ‘locally low pressure,’ and ‘rotation overtakes stretching.’ They L. Deng Y. Liu J. Liu College of Computer, National University of Defense Technology, Changsha, China E-mail: [email protected]; [email protected] J. Liu E-mail: [email protected] L. Deng (&) Y. Wang C. Chen Y. Liu F. Wang Computational Aerodynamics Institute, China Aerodynamics Research and Development Center, Mianyang, China E-mail: [email protected]
L. Deng et al.
only use the local information to compute some criterions including the Q-criterion (Hunt 1987), the Xcriterion (Liu et a
Data Loading...
