Particle Image Velocimetry Based on a Deep Learning Motion Estimator

Unsupervised deep learning for super-resolution reconstruction of turbulence

“…Furthermore, our model can be utilized to support high-resolution reconstruction of measurement data, such as PIV (Rabault, Kolaas & Jensen 2017; Cai et al. 2020), synchronization of different experiments, removal of experimental noise, semantic inpainting (Buzzicotti et al. 2020) and data assimilation (Leoni, Mazzino & Biferale 2020).…”

“…We expect that the proposed network will be of great assistance to LES modelling, including the production of pair data for the development of subgrid-scale models and synchronizations for model evaluation. Furthermore, our model can be utilized to support high-resolution reconstruction of measurement data, such as PIV (Rabault, Kolaas & Jensen 2017;Cai et al 2020), synchronization of different experiments, removal of experimental noise, semantic inpainting (Buzzicotti et al 2020) and data assimilation (Leoni, Mazzino & Biferale 2020). Our code will be released as an open-source code upon publication.…”

Flow over an espresso cup: inferring 3-D velocity and pressure fields from tomographic background oriented Schlieren via physics-informed neural networks

“…The PINN framework (Raissi et al 2019) has also been applied to perform complex flow simulations (Sun et al 2020; Jin et al 2021). On the other hand, data-driven algorithms have also been investigated for dealing with real experimental data (Rabault, Kolaas & Jensen 2017; Cai et al 2019; Jin et al 2020). For instance, a convolutional neural network (CNN) was applied to infer a dense velocity field from PIV images by Cai et al (2019).…”

“…On the other hand, data-driven algorithms have also been investigated for dealing with real experimental data (Rabault, Kolaas & Jensen 2017; Cai et al 2019; Jin et al 2020). For instance, a convolutional neural network (CNN) was applied to infer a dense velocity field from PIV images by Cai et al (2019). Moreover, Raissi, Yazdani & Karniadakis (2020) proposed the ‘hidden fluid mechanics’ (HFM) method based on PINNs to integrate the Navier–Stokes equations with visualization data, which enables quantification of the velocity and pressure from 3-D concentration fields.…”

Analysis of large visualization datasets for thermographic studies in fluid dynamics

“…A large amount of information and the development of technologies for storing and transmitting it has contributed to the introduction of machine learning into the analysis and processing. However, when getting new knowledge about the object of research in unique experimental conditions, the connection of machine learning is still problematic [1]. This is especially true for problems for which a numerical solution is not yet available, and it is difficult to obtain the initial synthesized data necessary for training.…”