Particle Image Velocimetry Dataset Research Articles

Experimental dataset investigation of deep recurrent optical flow learning for particle image velocimetry: flow past a circular cylinder

Current optical flow-based neural networks for particle image velocimetry (PIV) are largely trained on synthetic datasets emulating real-world scenarios. While synthetic datasets provide greater control and variation than what can be achieved using experimental datasets for supervised learning, it requires a deeper understanding of how or what factors dictate the learning behaviors of deep neural networks for PIV. In this study, we investigate the performance of the recurrent all-pairs field transforms-PIV (RAFTs-PIV) network, the current state-of-the-art deep learning architecture for PIV, by testing it on unseen experimentally generated datasets. The results from RAFT-PIV are compared with a conventional cross-correlation-based method, Adaptive PIV. The experimental PIV datasets were generated for a typical scenario of flow past a circular cylinder in a rectangular channel. These test datasets encompassed variations in particle diameters, particle seeding densities, and flow speeds, all falling within the parameter range used for training RAFT-PIV. We also explore how different image pre-processing techniques can impact and potentially enhance the performance of RAFT-PIV on real-world datasets. Thorough testing with real-world experimental PIV datasets reveals the resilience of the optical flow-based method’s variations to PIV hyperparameters, in contrast to the conventional PIV technique. The ensemble-averaged root mean squared errors between the RAFT-PIV and Adaptive PIV estimations generally range between 0.5–2 (px) and show a slight reduction as particle densities increase or Reynolds numbers decrease. Furthermore, findings indicate that employing image pre-processing techniques to enhance input particle image quality does not improve RAFT-PIV predictions; instead, it incurs higher computational costs and impacts estimations of small-scale structures.

The combined influence of spin and roughness frequency on sphere aerodynamics

AbstractThe lift and drag of spinning spheres roughened with macro-roughness elements are examined. The velocity field of these same spheres in flight is measured with particle image velocimetry (PIV). Several spheres with varying roughness are examined at various spin rates and fixed Reynolds number. Unlike previous studies, where the roughness height is varied, in the present work, the number of roughness elements is varied. The PIV datasets are used to determine the boundary layer separation points for each case. Comparing the lift and drag to the separation points reveals that (1) the separation points become more asymmetric with spin (the Magnus effect), (2) The drag increases with the size of the wake, and (3) the drag increases with the asymmetry of the separation points, meaning that lift on spheres is accompanied by increased drag. Scant evidence of this third effect has been reported previously. Additionally, it is shown that, counter to smooth spheres, the force transmitted to the surface through the roughness elements leads to significant drag. The drag is shown to increase with the number of roughness elements while the lift decreases. Results have implications for understanding aerodynamic forces on bluff bodies with roughness and passive control of aerodynamic forces through roughness element frequency rather than the traditional roughness height.

Correlations among high-order statistics and low-occurrence wind speeds within a simplified urban canopy based on particle image velocimetry datasets

Predicting infrequent and extreme wind speeds in a built environment is essential for ensuring comfortable and safe pedestrian spaces. Recent studies have employed statistical methods that assume a distribution function for wind speeds at pedestrian levels. Fundamental studies on the relationship between canopy flow and statistics are required to further develop statistical models. Therefore, this study aimed to understand the characteristics of strong and weak wind events within a simplified urban canopy and to scrutinize the relationship between high-order moments and extreme wind events. Particle image velocimetry (PIV) was employed to capture the velocities within a canopy consisting of cubes arranged in a staggered layout with a packing density of 25 %. The probability density functions (PDFs) of the velocity components classified by the mean flow patterns revealed that the PDF shapes were altered by the reverse and spanwise flows. In addition, strong correlations were verified between the gust or peak factor (PF) and high-order moments such as skewness, kurtosis, fifth-order, and sixth-order moments. Accordingly, the PF of the velocity components and wind speed were compared with the predictions by statistical methods based on the Weibull and Gram–Charlier series (GCS). These observations validate the previous statistical methods based on Weibull or GCS distributions. Although the physical interpretation of these statistics is ambiguous, the present analyses indicate that PF can be predicted by high-order moments, especially in particular, by skewness and kurtosis.

An Improved Convolutional Neural Network for Particle Image Velocimetry

With the wide application of Particle Image Velocimetry (PIV) technology in various engineering and research fields, the requirements for the accuracy, computational efficiency, and robustness of PIV algorithms are increasing. Although traditional algorithms have wide applicability, they suffer from low accuracy, large computational cost, and poor robustness. Recently, deep learning algorithms have provided new solutions, especially, convolutional neural networks with different structures, which have achieved good performance on synthetic PIV datasets. This paper proposes a structural improvement scheme for PIV convolutional neural network models. Experiments verify that the proposed method can significantly optimize the performance of the model on synthetic PIV datasets, providing a novel approach for improving other convolutional neural networks for PIV analysis.

Evidence that superstructures comprise self-similar coherent motions in high Reynolds number boundary layers

We present experimental evidence that the superstructures in turbulent boundary layers comprise smaller, geometrically self-similar coherent motions. The evidence comes from identifying and analysing instantaneous superstructures from large-scale particle image velocimetry datasets acquired at high Reynolds numbers, capable of capturing streamwise elongated motions extending up to 12 times the boundary layer thickness. Given the challenge in identifying the constituent motions of the superstructures based on streamwise velocity signatures, a new approach is adopted that analyses the wall-normal velocity fluctuations within these very long motions, which reveals the constituent motions unambiguously. The conditional streamwise energy spectra of the Reynolds shear stress and the wall-normal fluctuations, corresponding exclusively to the superstructure region, are found to exhibit the well-known distance-from-the-wall scaling in the intermediate-scale range. It suggests that geometrically self-similar motions are the constituent motions of these very-large-scale structures. Investigation of the spatial organization of the wall-normal momentum-carrying eddies, within the superstructures, also lends empirical support to the concatenation hypothesis for the formation of these structures. The association between the superstructures and self-similar motions is reaffirmed on comparing the vertical coherence of the Reynolds-shear-stress-carrying motions, by computing conditionally averaged two-point correlations, which are found to match with the mean correlations. The mean vertical coherence of these motions, investigated for the log region across three decades of Reynolds numbers, exhibits a unique distance-from-the-wall scaling invariant with Reynolds number. The findings support modelling of these dynamically significant motions via data-driven coherent structure-based models.

Numerical simulation and PIV measurements of the wave-induced flow field near semi-circular breakwaters

Careful design of the high-tech semi-circular breakwaters (SBW), which are used for protection against sea waves, is essential to avoid the failure of these large-scale projects. An improved design tool adopting numerical simulations, complemented with experimental measurements is introduced. Unstructured boundary-fitted grids are used to model the curved SBW boundary accurately and resolve the flow adjacent to it. Experimental measurements based on Particle Image Velocimetry (PIV) are introduced and the numerically simulated velocity fields are validated. For the first time, the strong free surface deformation upstream of the SBW, termed trough-breaking, is reproduced numerically. The phenomenon is investigated in detail and its impact on several SBW advantages and stability is emphasized. The improved accuracy due to unstructured boundary-fitted grids complemented by the PIV data set, offers valuable prospects for performing complex designs and exploring future applications.Article HighlightsAn efficient design tool for the high-tech semi-circular breakwater is introduced adopting numerical simulations and experimental measurements.The numerical results agree well with the experimental measurements. Hence, this experimental data set should be used extensively for validation in future studies.The important flow phenomenon of “trough-breaking” is reproduced numerically for the first time, and its impact on the semi-circular breakwater features is clarified.

Temporal super-resolution using smart sensors for turbulent separated flows

Particle image velocimetry (PIV) data of high Reynolds number unsteady turbulent flows are often undersampled in time; this leads to aliasing of important spectral content. The present work proposes a novel data-driven estimation technique that uses oversampled sparsely placed surface-mounted pressure sensors and long short-term memory neural networks to resolve the aliased transient velocity dynamics from undersampled PIV data. The method leverages the time-resolved pressure dynamics to estimate the temporal evolution of a proper orthogonal decomposition-based low-dimensional subspace of the velocity field. The proposed approach is demonstrated on a PIV dataset of a high Reynolds number turbulent separated flow over a Gaussian speed-bump benchmark geometry ( $$\text{Re}_{H}=2.26\times 10^{5}$$ , where H is the Bump height). The 15 Hz PIV data is super-resolved to 2 kHz, and spectral analysis of the flowfields is conducted to educe the originally aliased unsteady dynamics of the turbulent separation bubble. The estimator is shown to accurately reconstruct the Reynolds shear stress from unseen sensor data, demonstrating its generalizability to resolve the coherent motions. The estimated velocity spectra show distributions consistent with those of other separated flows.

On the turbulent boundary layer over a flat plate at moderate Reynolds numbers

Two separate experimental campaigns of a spatially developing turbulent boundary layer under approximately zero-pressure-gradient at moderate Reynolds numbers (1700<Reθ<3400) are conducted with stereoscopic Particle Image Velocimetry (PIV) and one component Hot Wire Anemometry. This range of Reynolds numbers is found to be of particular interest for turbulent boundary layer control investigations. The motivations behind this work rely on the lack of recent studies that provide a rigorous experimental database on a flat plate turbulent boundary layer, openly available online. This is critical as, in most of the cases, the modification of the statistics resulting from turbulent boundary layer control strategies are compared with a smooth baseline reference. The statistics of the velocity fields, obtained with the two techniques, show a good match with the direct numerical simulation in literature results. We focused on the skin friction evaluation by means of Clauser's chart technique. The near wall turbulence activity and the associated coherent structures are investigated by means of the Variable Interval Time Averaging technique using the hot wire signal. The influence of the acquisition and algorithm parameters as well as the effect of the Reynolds number are reported. The logarithmic and outer structures are investigated by applying the Uniform Momentum Zones technique to the PIV dataset. The hierarchical distribution of the uniform momentum zones as a function of the wall distance as well as their variation with the Reynolds number confirm the validity of the attached eddy model even at the moderate Reynolds numbers of the current investigation.

A robust single-pixel particle image velocimetry based on fully convolutional networks with cross-correlation embedded

Particle image velocimetry (PIV) is essential in experimental fluid dynamics. In the current work, we propose a new velocity field estimation paradigm, which is a synergetic combination of cross correlation and fully convolutional network (CC-FCN). Specifically, the fully convolutional network is used to optimize and correct a coarse velocity guess to achieve a super-resolution calculation. And the traditional cross correlation method provides the initial velocity field based on a coarse correlation with a large interrogation window. As a reference, the coarse velocity guess helps with improving the robustness of the proposed algorithm. CC-FCN has two types of input layers, one is for the particle images, and the other is for the initial velocity field calculated using cross correlation with a coarse resolution. First, two pyramidal modules extract features of particle images and initial velocity field, respectively. Then the fusion module appropriately fuses these features. Finally, CC-FCN achieves the super-resolution calculation through a series of deconvolution layers to obtain the single-pixel velocity field. As the supervised learning strategy is considered, synthetic data sets including ground-truth fluid motions are generated to train the network parameters. Synthetic and real experimental PIV data sets are used to test the trained neural network in terms of accuracy, precision, spatial resolution and robustness. The test results show that these attributes of CC-FCN are further improved compared with those of other tested PIV algorithms. The proposed model could therefore provide competitive and robust estimations for PIV experiments.

Unsupervised Recurrent All-Pairs Field Transforms for Particle Image Velocimetry

Convolutional neural networks have been successfully used in a variety of tasks and recently have been adapted to improve processing steps in Particle-Image Velocimetry (PIV). Recurrent All-Pairs Fields Transforms (RAFT) as an optical flow estimation backbone achieve a new state-of-the-art accuracy on public synthetic PIV datasets, generalize well to unknown real-world experimental data, and allow a significantly higher spatial resolution compared to state-of-the-art PIV algorithms based on cross-correlation methods. However, the huge diversity in dynamic flows and varying particle image conditions require PIV processing schemes to have high generalization capabilities to unseen flow and lighting conditions. If these conditions vary strongly compared to the synthetic training data, the performance of fully supervised learning based PIV tools might degrade. To tackle these issues, our training procedure is augmented by an unsupervised learning paradigm which remedy the need of a general synthetic dataset and theoretically boosts the inference capability of a deep learning model in a way being more relevant to challenging real-world experimental data. Therefore, we propose URAFT-PIV, an unsupervised deep neural network architecture for optical flow estimation in PIV applications and show that our combination of state-of-the-art deep learning pipelines and unsupervised learning achieves a new state-of-the-art accuracy for unsupervised PIV networks while performing similar to supervisedly trained LiteFlowNet based competitors. Furthermore, we show that URAFT-PIV also performs well under more challenging flow field and image conditions such as low particle density and changing light conditions and demonstrate its generalization capability based on an outof-the-box application to real-world experimental data. Our tests also suggest that current state-of-the-art loss functions might be a limiting factor for the performance of unsupervised optical flow estimation.

Deep particle image velocimetry supervised learning under light conditions

Particle image velocimetry (PIV) is an important fluid visualization technology which extracts the velocity field from two successive particle images. Recently, some researchers have begun to use convolutional neural network (CNN) to tackle the PIV problem successfully. Some supervised learning methods make use of the PIV dataset with ground truth for network training. However, the existing dataset is composed of pairs of particle images under ideal light conditions and does not take into account the changes in actual experimental conditions. In this paper, we firstly generated a new and more challenging dataset called Light-PIV which fully simulates the change of the brightness of particle images in the real PIV experiment. Secondly, we present here a novel approach for fluid motion estimation which is based on an optical flow network LiteFlowNet. The proposed approach is verified by the application to a diversity of synthetic and experimental PIV images. We not only improve the structure, but also combine the traditional prior assumptions knowledge with the loss function to better guide the network training. The proposed approach is verified by the application to a diversity of synthetic and experimental PIV images. The experimental results show that our proposed method has advantages of high accuracy, obtaining detailed information and strong robustness in our PIV dataset compared with classical PIV methods such as HS optical flow and WIDIM, and even outperforms these existing approaches in some flow cases.

Wavepacket modelling of broadband shock-associated noise in supersonic jets

Abstract

Predicting the near-wall velocity of wall turbulence using a neural network for particle image velocimetry

Near-wall velocity prediction for wall-bounded turbulence is useful for constructing a wall model and estimating dissipation and wall shear stress. A convolutional neural network is developed to improve the near-wall velocity prediction and spatial resolution for wall-bounded turbulent velocity fields obtained using particle image velocimetry (PIV). To establish the relationship between the low-resolution and high-resolution fields, this machine learning model is trained on a synthetic PIV dataset generated based on velocity fields obtained from the direct numerical simulation of turbulent channel flows at Reτ = 1000. Using a test dataset with a higher Reynolds number of Reτ = 5200, the performance of this model is assessed in terms of instantaneous fields, error analysis, velocity statistics, and energy spectra. The influences of the interrogation window, image resolution, and particle concentration on the performance of this network are also considered. We further apply this network to practical PIV data from a turbulent boundary layer at Reτ = 2200 to assess the network performance under real experimental conditions. The results indicate that the proposed machine-learning-based model can predict missing near-wall velocity fields and enhance the spatial resolution of PIV fields, but the accuracy for Reynolds shear stress prediction needs to be further improved. The presented approach shows the potential ability to predict the near-wall instantaneous velocity of high-Reynolds-number turbulence from low-Reynolds-number flow fields.

Prediction of the dynamics of a backward-facing step flow using focused time-delay neural networks and particle image velocimetry data-sets

The objective of this experimental work was to evaluate the potential of an artificial Neural Network (NN) to predict the full-field dynamics of a standard separated, noise-amplifier flow: the Backward-Facing Step (BFS) flow at Reh=1385. Different upstream local visual sensors, based on the velocity fields measured by time-resolved Particle Image Velocimetry, were tested as inputs for the Neural Network. The dynamic coefficients of a Proper Orthogonal Decomposition (POD) were defined as goals-outputs for this non-linear mapping. The coefficients time-series were predicted and the instantaneous velocity fields were reconstructed with satisfying accuracy with a Focused Time-Delay Neural Network (FTDNN). Using a time-delay appears like a crucial choice to ensure an accurate prediction of the dynamics of the BFS flow. A shallow FDTNN is sufficient to obtain good accuracy with low computational time. The influence of the choices of inputs-sensors, the size of the training data-set, the number of neurons in the hidden layer as well as the sensor delay on the accuracy of the predicted flow are discussed for this experimental fluid system.

Spurious PIV Vector Correction Using Linear Stochastic Estimation

Techniques for the experimental determination of velocity fields such as particle image velocimetry (PIV) can often be hampered by spurious vectors or sparse regions of measurement which may occur due to a number of reasons. Commonly used methods for detecting and replacing erroneous values are often based on statistical measures of the surrounding vectors and may be influenced by further poor data quality in the region. A new method is presented in this paper using Linear Stochastic Estimation for vector replacement (LSEVR) which allows for increased flexibility in situations with regions of spurious vectors. LSEVR is applied to PIV dataset to demonstrate and assess its performance relative to commonly used bilinear and bicubic interpolation methods. For replacement of a single vector, all methods performed well, with LSEVR having an average error of 11% in comparison to 14% and 18% for bilinear and bicubic interpolation respectively. A more significant difference was found in replacement of clusters of vectors which showed average vector angle errors of 10°, 9° and 6° for bilinear, bicubic and LSEVR respectively. Error in magnitude was 3% for both interpolation techniques and 1% for LSEVR showing a clear benefit to using LSEVR for conditions that require multiple clustered vectors to be replaced.

Insights into the dynamics of conical breakdown modes in coaxial swirling flow field

The main idea of this paper is to understand the fundamental vortex breakdown mechanisms in the coaxial swirling flow field. In particular, the interaction dynamics of the flow field is meticulously addressed with the help of high fidelity laser diagnostic tools. Time-resolved particle image velocimetry (PIV) (${\sim}1500~\text{frames}~\text{s}^{-1}$) is employed in $y{-}r$ and multiple $r{-}\unicode[STIX]{x1D703}$ planes to precisely delineate the flow dynamics. Experiments are carried out for three sets of co-annular flow Reynolds number $Re_{a}=4896$, 10 545, 17 546. Furthermore, for each $Re_{a}$ condition, the swirl number ‘$S_{G}$’ is varied independently from $0\leqslant S_{G}\leqslant 3$. The global evolution of flow field across various swirl numbers is presented using the time-averaged PIV data. Three distinct forms of vortex breakdown namely, pre-vortex breakdown (PVB), central toroidal recirculation zone (CTRZ; axisymmetric toroidal bubble type breakdown) and sudden conical breakdown are witnessed. Among these, the conical form of vortex breakdown is less explored in the literature. In this paper, much attention is therefore focused on exploring the governing mechanism of conical breakdown. It is should be interesting to note that, unlike other vortex breakdown modes, conical breakdown persists only for a very short band of $S_{G}$. For any small increase/decrease in $S_{G}$ beyond a certain threshold, the flow spontaneously reverts back to the CTRZ state. Energy ranked and frequency-resolved/ranked robust structure identification methods – proper orthogonal decomposition (POD) and dynamic mode decomposition (DMD) respectively – are implemented over instantaneous time-resolved PIV data sets to extract the dynamics of the coherent structures associated with each vortex breakdown mode. The dominant structures obtained from POD analysis suggest the dominance of the Kelvin–Helmholtz (KH) instability (axial $+$ azimuthal; accounts for ${\sim}80\,\%$ of total turbulent kinetic energy, TKE) for both PVB and CTRZ while the remaining energy is contributed by shedding modes. On the other hand, shedding modes contribute the majority of the TKE in conical breakdown. The frequency signatures quantified from POD temporal modes and DMD analysis reveal the occurrence of multiple dominant frequencies in the range of ${\sim}10{-}400~\text{Hz}$ with conical breakdown. This phenomenon may be a manifestation of high energy contribution by shedding eddies in the shear layer. Contrarily, with PVB and CTRZ, the dominant frequencies are observed in the range of ${\sim}20{-}40~\text{Hz}$ only. We have provided a detailed exposition of the mechanism through which conical breakdown occurs. In addition, the current work explores the hysteresis (path dependence) phenomena of conical breakdown as functions of the Reynolds and Rossby numbers. It has been observed that the conical mode is not reversible and highly dependent on the initial conditions.

Quantification and adjustment of pixel-locking in particle image velocimetry

A quantification metric is provided to determine the degree to which a particle image velocimetry data set is pixel-locked. The metric is calculated by integrating the histogram equalization transfer function and normalizing by the worst-case scenario to return the percentage pixel-locked. When this metric is calculated for each position in the vector field, it is shown that pixel-locking is non-uniform across the field. Hence, pixel-locking adjustments should be made on a vector-by-vector basis rather than uniformly across a field, although the latter is the common practice. A methodology is provided to compensate for the effects of pixel-locking on a vector-by-vector basis. This includes applying a Gaussian filter directly to the images, processing the images with window deformation, ensuring the vector fields are in pixel displacements, applying histogram equalization calculated at each vector coordinate, and mapping the adjusted vector fields to physical space.

Evaluating Large-Eddy Simulation (LES) and High-Speed Particle Image Velocimetry (PIV) with Phase-Invariant Proper Orthogonal Decomposition (POD)

This study is part of a program to understand the stochastic variations in IC engine flows; in particular, it is a comparison of measured (PIV) and computed (LES) velocity from multiple cycles of the same motored engine. Comparison procedures included traditional RANS (Reynolds Averaged Navier-Stokes) decomposition (ensemble-averaged and RMS (Root Mean Square) velocity), phase-dependent, and phase- invariant POD. Phase-dependent POD was performed on the PIV and LES samples separately and on the combined samples, thus creating separate or a single POD mode sets, respectively. The phase- invariant POD was performed both on normalized snapshots and on snapshots where the original energy was conserved. Initial comparisons of the mass-specific kinetic energies of the ensemble average and RMS velocities revealed that the PIV and LES data sets differed significantly during most of the intake stroke. This discrepancy was quantifiedfirst by comparing the relevance indices calculated between ensemble average velocity fields and, second, using phase-dependent POD, which quantified cycle-to-cycle flow variations of the RANS average and turbulence. Phase-dependent POD was applied separately to the PIV and LES data sets during the intake stroke (76° ATDCE, After Top Dead Center Exhaust), where the intake-valve jet is strong and the PIV and LES data were earlier found to be significantly different. The cyclic variability of the LES ensemble average was estimated to be significantly higher than that of the PIV data set. POD was also applied to the combined sample of LES and PIV snapshots for quantitative comparison creating a single set of modes, so that comparison could be made with POD coefficients alone. Example comparisons were made at again for data at 76° and also 330° ATDCE, which is a viable spark timing in a fired engine. The results at 76° ATDCE were similar to those obtained with the POD analysis of the separate samples. At 330° ATDCE, the PIV ensemble average and RMS velocities showed somewhat more cyclic variability. Phase-invariant POD was applied to the combined PIV and LES velocity data from all crank angles to study the flow evolution over the crank angle range. Conclusions derived from the two different energy transformations are contrasted. When the energy of each velocity field is normalized, the phase-invariant POD results focus on differences in flow structures and their evolution. On the other hand, when the energy of each velocity field is conserved, the phase-invariant POD results also take into account differences in energy between the PIV and LES data.

Minimization of divergence error in volumetric velocity measurements and implications for turbulence statistics

Volumetric velocity measurements taken in incompressible fluids are typically hindered by a nonzero divergence error due to experimental uncertainties. Here, we present a technique to minimize divergence error by employing continuity of mass as a constraint, with minimal change to the measured velocity field. The divergence correction scheme (DCS) is implemented using a con- straint-based nonlinear optimization. An assessment of DCS is performed using direct numerical simulations (DNS) velocity fields with random noise added to emulate experimental uncertainties, together with a Tomographic particle image velocimetry data set measured in a channel flow facility at a matched Reynolds number to the DNS data (Res & 937). Results indicate that the divergence of the corrected velocity fields is reduced to near zero, and a clear improvement is evident in flow statistics. In particu- lar, significant improvements are observed for statistics computed using spatial gradients such as the velocity gra- dient tensor, enstrophy, and dissipation, where having zero divergence is most important.

Mode competition in streamwise-only vortex induced vibrations

Time-resolved Particle-Image Velocimetry (PIV) has been used to study mode competition and transient behaviour in the wake of a cylinder experiencing Vortex-Induced Vibrations (VIV) in the streamwise direction. The cylinder response regime contained two branches, occurring above and below the onset of synchronisation between the wake and the cylinder motion (lock-in). During the first branch, the wake exhibited both the S-I mode (in which two vortices are shed simultaneously per vibration cycle) and the alternate A-II mode (similar to the well known von Kármán vortex street). An extended PIV data set acquired in this region revealed mode switching between the S-I and A-II modes. A criterion based on Proper-Orthogonal Decomposition was developed to identify which mode was dominant as a function of time. The A-II mode was found to be dominant for over 90% of the instantaneous fields examined, while the S-I mode appeared to be more unstable.Symmetrically shed vortices were found to rearrange downstream into an alternate structure in which the wake was no longer synchronised to the cylinder motion. The dominant frequency of transverse velocity fluctuations was measured throughout the wake in order to study the effects of this breakdown in more detail. For the majority of the wake, the fluctuations occurred at the Strouhal frequency, while in a region in the near wake the fluctuations occurred at the frequency of the cylinder motion. It is thought that during the first response branch vortices are formed at the cylinder response frequency, but tend to quickly rearrange downstream into an alternate structure which is no longer synchronised to the cylinder motion. As a result, the fluctuating drag will be synchronised to the structural motion, and is capable of providing positive energy transfer in the apparent absence of lock-in. Finally, the spatial dependence of the frequency of velocity fluctuations throughout the wake is used to explain some of the conflicting results in the literature regarding streamwise VIV, and the implications for the general study of VIV are discussed.