Tauro Flavia, Grimaldi Salvatore, Porfiri Maurizio
Department of Mechanical and Aerospace Engineering, New York University Polytechnic School of Engineering, Brooklyn, New York, United States of America; Dipartimento di Ingegneria Civile, Edile e Ambientale, Sapienza University of Rome, Rome, Italy.
Department of Mechanical and Aerospace Engineering, New York University Polytechnic School of Engineering, Brooklyn, New York, United States of America; Dipartimento per l'Innovazione nei Sistemi Biologici, Agroalimentari e Forestali, University of Tuscia, Viterbo, Italy.
PLoS One. 2014 Mar 10;9(3):e91131. doi: 10.1371/journal.pone.0091131. eCollection 2014.
From climatology to biofluidics, the characterization of complex flows relies on computationally expensive kinematic and kinetic measurements. In addition, such big data are difficult to handle in real time, thereby hampering advancements in the area of flow control and distributed sensing. Here, we propose a novel framework for unsupervised characterization of flow patterns through nonlinear manifold learning. Specifically, we apply the isometric feature mapping (Isomap) to experimental video data of the wake past a circular cylinder from steady to turbulent flows. Without direct velocity measurements, we show that manifold topology is intrinsically related to flow regime and that Isomap global coordinates can unravel salient flow features.
从气候学到生物流体力学，复杂流动的表征依赖于计算成本高昂的运动学和动力学测量。此外，如此庞大的数据难以实时处理，从而阻碍了流动控制和分布式传感领域的进展。在此，我们提出了一种通过非线性流形学习对流动模式进行无监督表征的新颖框架。具体而言，我们将等距特征映射（Isomap）应用于从稳定流到湍流的圆柱尾流的实验视频数据。在没有直接速度测量的情况下，我们表明流形拓扑与流动状态内在相关，并且Isomap全局坐标可以揭示显著的流动特征。
