Bulletin of the American Physical Society
75th Annual Meeting of the Division of Fluid Dynamics
Volume 67, Number 19
Sunday–Tuesday, November 20–22, 2022; Indiana Convention Center, Indianapolis, Indiana.
Session T16: Experimental Techniques: Data Assimilation, Bias and Uncertainty |
Hide Abstracts |
Chair: Georgios Rigas, Imperial College London Room: 143 |
Monday, November 21, 2022 4:10PM - 4:23PM |
T16.00001: Data Assimilation for Turbulent Channel Flow and Pressure Computation using Omnidirectional Integration Mohamed Amine Abassi, Qi Wang, Xiaofeng Liu Numerical data assimilation incorporates the theoretical model-based prediction with experimental observations and seeks their optimal reconciliation to achieve the optimal forecast of a dynamical process. In this study, we present an application of data assimilation to improve the spatial data availability of time-resolved tomographic Particle Image Velocimetry (PIV) measurements. The PIV spatial resolution is limited by the particle spatial concentration, which often leaves void spots in the fluid domain. Simple interpolations can fill the gap at the price of ignoring the physics of the problem, governed by the Navier-Stokes equations. In contrast, by using data assimilation, the discrepancy between the computed velocity field and the measured velocity values, as quantified by a cost function, is minimized through the adjoint method. We use the Johns Hopkins turbulence database as test cases. To compute the pressure field, instead of solving the Poisson equation, the parallel ray omnidirectional integration technique (Liu et al. 2016) is adopted at each time step. The ultimate objective of this project is to establish a noise-tolerant data assimilation framework that can be used as an augmentation tool for PIV measurements. |
Monday, November 21, 2022 4:23PM - 4:36PM |
T16.00002: Unsteady Measurements with a Five-Hole Probe Rhett Cook, Pourya Nikoueeyan, Jonathan W Naughton Multi-hole probes have provided reliable flow-field measurements for many years. However, the use of these probes has been limited to mean measurements due to the attenuation and lag that is experienced by the pressure signals passing from the ports on the probe tips to the transducer making the pressure measurement. To address this, some groups have tried to minimize the distance between the ports and the transducer, but this can cause other issues such as resonance in the tubing system. To address this issue, a novel approach for correcting the measured pressures for distortion has been applied to a five-hole probe. In this approach, each port/tube/tranducer is characterized by applying a known signal to the port and recording the measured pressure at the transducer. These measurements are used to tune a model of the tubing system, which can then be used to correct future measurements. The characterization of the probe is discussed, and its use in a jet flow is presented. Unsteady measurements from multi-hole probes make possible a wide range of measurements such as the pressure-velocity correlation. |
Monday, November 21, 2022 4:36PM - 4:49PM |
T16.00003: Classifying Turbulent Environments via Machine Learning Michele Buzzicotti, Fabio Bonaccorso, Luca Biferale The problem of classifying turbulent environments from partial observation is key for some theoretical and applied fields, from engineering to earth observation and astrophysics, e.g. it is important for the selection of optimal control policies trained in different turbulent backgrounds, to predict the probability of extreme events and/or to infer physical parameters labeling the different turbulent set-ups. To achieve such a goal one can use different tools depending on the system's knowledge and on the quality and quantity of the accessible data. In this context, we assume to work in a model-free setup completely blind to all dynamical laws, but with a large quantity of (good quality) data for training. As a prototype of complex flows with different attractors and different multi-scale statistical properties, we selected 10 turbulent 'ensembles' by changing the rotation frequency of the frame of reference of a 3d domain. We are supposed to have access to a set of partial observations limited to the instantaneous kinetic energy distribution in a 2d plane, as is often the case in geophysics and astrophysics. We compare results obtained by a Machine Learning (ML) approach consisting of a state-of-the-art Deep Convolutional Neural Network (DCNN) against Bayesian inference which exploits the information on velocity and enstrophy moments. First, we discuss the supremacy of the ML approach, presenting also results changing the number of training data and of the hyper-parameters. Second, we present an ablation study on the input data aimed to perform a ranking on the importance of the flow features used by the DCNN, helping to identify the main physical contents used by the classifier. Finally, we discuss the main limitations of such data-driven methods and potential interesting applications. |
Monday, November 21, 2022 4:49PM - 5:02PM |
T16.00004: Temporally quasi-periodic data, propagating in the laboratory frame, can be rendered periodic by Galilean transformation Bill D. Caraway, Arne J Pearlstein Velocity, temperature, pressure, or concentration data are usually acquired in the laboratory frame, but when a preferred direction exists (e.g., due to mean flow), the question arises as to whether one can gain additional insight or simplify data analysis, by considering the data in a different reference frame. For a broad class of velocity, temperature, pressure, or concentration distributions propagating rectilinearly, we show that temporally quasi-periodic behavior in the laboratory frame can be rendered periodic by appropriate Galilean transformation. The approach is illustrated analytically and numerically using as an example a closed-form model distribution generated from a one-dimensional partial differential equation describing a pressure-driven diffusion process. A detailed procedure is developed to determine appropriate frame speeds for more general quasi-periodic, one-dimensional data, either continuous in space and time, or temporally- and spatially-discretized. The approach is extended to two- and three-dimensional rectilinear (and some nonrectilinear) propagation, and implications for interpreting noise-corrupted data are also discussed. |
Monday, November 21, 2022 5:02PM - 5:15PM |
T16.00005: Low-order pressure determination in the flow of a stalled airfoil from PIV Douglas Carter, Bharathram Ganapathisubramani Low-order representations of velocity fields, for example using the proper orthogonal decomposition (POD), are steadily increasing in popularity due to the attractive feature of capturing most of the energy of a dynamical system with a limited subset of modes and mode coefficients. In this talk, we explore the question of how many velocity modes are required to reach a desired accuracy in pressure reconstructions from time-resolved particle image velocimetry data in the turbulent flow of a static stalled airfoil at a chord-based Reynolds number Rec=7.1x104. This is demonstrated cumulatively by retaining up to a cut-off number of modes as well as mode-by-mode through a Galerkin projection of the Poisson equation onto the POD space spanned by the modes. The number of velocity modes required to capture the pressure to a desired accuracy is quantified and discussed. |
Monday, November 21, 2022 5:15PM - 5:28PM |
T16.00006: Ultrasound Image Velocimetry (UIV) Uncertainty Quantification Rozhin Derakhshandeh, Sayantan Bhattacharya, Brett A Meyers, Pavlos P Vlachos Ultrasound image velocimetry (UIV) is a non-invasive flow measurement technique used to study flows through opaque media, especially in complex flows in clinical and industrial research applications. Sensitive decision-making based on these measurements intensifies the importance of characterizing the UIV measurement accuracy and uncertainty. However, UIV uncertainty quantification (UQ) is non-trivial due to the complexity of contributing error sources, their combination, and propagation through the measurement chain. This work presents a Generalized Moment of Correlation (GMC) method for UQ in a UIV measurement. GMC is based on the MC method for particle image velocimetry (PIV). However, the method accounts for the stretched shape of image intensities typical in an ultrasound image. Synthetic image analysis shows a 50% improvement in GMC UQ prediction compared to MC for particle image aspect ratios higher than 1.5. Furthermore, GMC is sensitive to the UIV error sources and predicts the measurement uncertainty with 90% accuracy. UQ in clinical ultrasound images also demonstrates reasonable levels of uncertainty with the consistent spatial distribution. GMC provides the first reliable UQ for a UIV measurement. |
Monday, November 21, 2022 5:28PM - 5:41PM |
T16.00007: Utilizing the physics of microfluidic sensing schemes in signal processing to lower the detection limit Henning Bonart, Florian Gebhard, Lukas Hecht, Tamal Roy, Benno Liebchen, Steffen Hardt Lowering the limit of detection in chemical or biochemical analysis is the key to extending the application scope of many detection methods. Using the example of a detection scheme based on microfluidic isotachophoresis, we present an approach for lowering the limit of detection via signal processing by utilizing knowledge about the physics of the electrophoretic sample transport and of the imaging process. By cross-correlating pairs of noisy fluorescence images of an analyte focused by isotachophoresis, the electrophoretic velocity of the sample can be extracted even at low signal-to-noise ratios in a first step. Based on this velocity, a Galilean transformation is then performed on the whole set of images to align the fluorescence distributions of the sample and create a series of quasi-replicate measurements. Averaging over the transformed data leads to a significant reduction of Gaussian White Noise superposing the raw images, where the signal-to-noise ratio after processing scales with the number N of frames considered as √N. In this way, the limit of detection is lowered by about two orders of magnitude without any additional instrumentation. Bayesian inference is used to include the uncertainties of the measurements and signal processing in the final detection decision. |
Monday, November 21, 2022 5:41PM - 5:54PM |
T16.00008: Complete flow characterization from snapshot PIV, fast probes and physics-informed neural networks Alvaro Moreno Soto, Alejandro Güemes, Stefano Discetti The use of physics-informed neural networks (PINNs), based on the incorporation of governing laws to constrain the training of machine-learning algorithms, has widened the possibilities for artificial intelligence to model and regularize experimental data. PINNs have been recently shown to improve the accuracy of time-resolved measurements, but their capabilities are remarkably reduced when time resolution is not available. In this work, we exploit PINNs to enhance velocity measurements from non-time-resolved field measurements, such as those from snapshot Particle Image Velocimetry (PIV). We use PINNs as a regularizer of time-resolved estimated fields from simultaneous measurements with fast pointwise probes and non-time-resolved PIV. We make use of a multilayer perceptron architecture to set a correspondence between probe data and the temporal coefficients of the Proper Orthogonal Decomposition of the PIV velocity profiles. The estimated time-resolved fields are then fed to the PINNs to enhance data accuracy and additionally extract derived quantities such as the pressure field. |
Monday, November 21, 2022 5:54PM - 6:07PM |
T16.00009: Mean-flow reconstruction of unsteady flows via data assimilation: PINN vs variational methods. Georgios Rigas, Yusuf Patel, Olivier Marquet, Vincent Mons Successful application of data-assimilation methods for fluid dynamics problems can lead to a significant reduction in costs associated with acquiring experimental data in wind tunnels or performing expensive fluid simulations. The aim of this work is to leverage experimental sparse mean flow measurements on the surface or in the unsteady wake of a body, in order to extract flow information that is not available in the measurements or in the under-determined governing Reynolds-Averaged Navier-Stokes (RANS) equations. This can be beneficial for super-resolving mean flow fields, extending the measurements and filling missing gaps, inferring pressure information and inferring closure terms for the RANS equations. In this work, two different techniques are employed to perform mean-flow data assimilation; a variational adjoint-based approach and Physics Informed Neural Networks (PINNs). Although the PINN and variational approaches aim to solve the data-assimilation problem by leveraging available data and governing equations to maximize a cost functional, there are key differences that affect the accuracy of the assimilation procedure. A thorough comparison of variational and PINN data assimilation methods will be presented for laminar and turbulent regimes. |
Monday, November 21, 2022 6:07PM - 6:20PM |
T16.00010: Distortion of passive scalar structure during suction-based plume sampling Aaron C True, John P Crimaldi Studies of plume dynamics often rely on photoionization detectors (PID) to quantify spatiotemporal distributions of passive scalars (gases, vapors, odors). However, the potential for PID suction to distort filaments and to modify sensed time records remains unclear. We used computational fluid dynamics to model a common PID to quantify and parameterize suction distortion by considering how sensed time records compare to those registered by an ideal probe. Models cover a range of realistic plumes, and we show that PID can significantly modify the peak concentration and pulse shape of sensed records. We quantified distortion variations in three nondimensional parameters describing PID geometry and sampling conditions: relative suction rate, relative filament size, and ambient flow Reynolds number. We used analytical models, dimensional analysis, and scaling arguments to interpret results and discuss when distortion is likely and what drives it. We built dimensionless distortion prediction regressions, and our results enable PID users to estimate distortion levels and to employ mitigation strategies through suction velocity tuning. These findings can inform distortion-mitigating design principles and best sampling practices for other suction-based passive scalar sensing schemes. |
Monday, November 21, 2022 6:20PM - 6:33PM |
T16.00011: Sequential Regression of Finite-Time Lyapunov Exponents for Fast Identification of Fluid Transport Features Tanner D Harms, Steven L Brunton, Beverley J McKeon Computing finite-time Lyapunov exponents (FTLE) from experimental data is a computationally intensive task that is further complicated by measurement noise and disappearing particles. This work addresses these challenges by proposing regression as a means of computing the flow map Jacobian required in FTLE calculations. Further, it demonstrates that the new approach can be performed sequentially on particle snapshots typical of experimental data for improved performance. The theory supporting the new approach is discussed in relation to traditional methods using finite-differences and the method is applied to a range of flow configurations. Simulated data are used to study the influence of the method's parameters on outcomes and robustness to noise is examined with respect to traditional FTLE results. |
Monday, November 21, 2022 6:33PM - 6:46PM |
T16.00012: Pressure Reconstruction from Noisy Measurements of Pressure Gradient by Gaussian Process Regression in Isotropic Turbulence Zejian You, Qi Wang, Xiaofeng Liu An accurate estimation of the pressure fields is of vital importance in various fluid dynamics applications. However, non-intrusive, direct pressure measurement techniques with high resolution are often very challenging. Consequently, many numerical tools have been established to reconstruct pressure fields from material acceleration obtained by Particle Image Velocimetry (PIV), such as the state-of-art Omni-directional Integration(ODI) method. This study introduces the framework of the Gaussian Process Regression(GPR) method to reconstruct the pressure field from material acceleration embedded with measurement noise. Similar to the testing method used in Liu and Moreto (2020), pressure gradient fields from Johns Hopkins Turbulent database are superimposed with 1000 statistically independent homogeneous error distributions to conduct a comparison of real and reconstructed pressure fields by ODI and GPR methods. The preliminary result demonstrates that GPR method with radial basis kernel function enables a versatile application in random and sparse observations with the same accuracy as the ODI method. Furthermore, this study provides an extended mathematical explanation of noise reduction by comparing different variance kernels of the Gaussian Process. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700