Studying propagating turbulent structures in the near wake of a sphere using Hilbert proper orthogonal decomposition

This study demonstrates that applying the Hilbert transform directly to standard Proper Orthogonal Decomposition (POD) modes offers a computationally efficient and artifact-free method for identifying propagating turbulent structures in a sphere's near wake, effectively replicating the results of the more complex Hilbert POD technique.

The Gist

Imagine you are standing by a river, watching a large, smooth boulder sitting in the middle of the current. The water rushing past it doesn't just flow smoothly; it churns, swirls, and creates chaotic patterns of eddies and vortices. This is turbulence. To the naked eye, it looks like random noise. But scientists know that hidden inside this chaos are "coherent structures"—organized, repeating patterns, like a dance routine that the water is performing over and over again.

The goal of this paper is to figure out how to find and describe these hidden dance routines in the wake of a sphere (a ball) moving through water.

The Problem: Finding the Needle in the Haystack

The researchers used a powerful mathematical tool called POD (Proper Orthogonal Decomposition). Think of POD as a super-smart camera that takes thousands of snapshots of the water and tries to sort them into "modes" or "themes."

• How it works: It's like listening to a chaotic orchestra and trying to separate the sound of the violins from the drums. POD separates the water's motion into different layers of importance.
• The Catch: In simple flows (like a cylinder), the "dance" is a perfect back-and-forth wobble. POD sees this easily as a pair of modes (one for the left swing, one for the right). But in complex flows (like a sphere), the dance is messy. The "left swing" and "right swing" might get mixed up with other random noises, making it hard to tell which modes belong to the same moving wave.

The Solution: The "Hilbert" Magic Trick

To solve this, the authors introduced a new method called HPOD (Hilbert POD).

• The Analogy: Imagine you are watching a runner on a track. If you just take a photo, you see them at one spot. If you take a video, you see them moving. The Hilbert Transform is like a special lens that turns a single photo into a "ghostly" video. It predicts where the runner was a split second ago and where they will be a split second later.
• The Result: By applying this lens, the math can "see" the wave traveling downstream. It groups the "left swing" and "right swing" together perfectly, even if they are buried in noise.
• The Downside: This magic lens isn't perfect. Because it assumes the wave goes on forever in a loop, it sometimes creates "ghosts" or artifacts at the edges of the picture (called spectral leakage). It's like a projector that assumes the movie is a loop; if the movie actually ends, the projector might smear the last frame into the first one.

The Breakthrough: A Shortcut

The researchers realized they didn't need to run the heavy, complex HPOD calculation on the entire messy water data.

• The Shortcut: They first used the standard POD to get the basic "themes" (the modes). Then, they applied the "Hilbert lens" only to those themes.
• Why it's great: It's like sorting your laundry by color first (POD), and then only ironing the shirts (Hilbert Transform) instead of ironing the whole pile of clothes. It saves a massive amount of computer power and time, but it still finds the exact same "traveling waves" as the heavy method.

What They Found in the Sphere's Wake

Using these methods on a sphere moving through water (at a specific speed), they discovered the sphere's wake isn't just one big mess. It has distinct "dance moves":

1. The Flap (Modes 1 & 2): The biggest structure is the wake "flapping" side-to-side, like a flag in the wind. The water swings left, then right, propagating downstream.
2. The Pulse (Modes 4 & 5): There is also a "breathing" motion where the wake expands and contracts, pulsing forward and backward.
3. The Smaller Flap (Modes 6 & 8): A faster, shorter version of the side-to-side flapping.
4. The Small Pulse (Modes 9 & 10): A smaller, faster version of the breathing motion.

The Takeaway

The paper proves that you can find these traveling waves in complex turbulence without needing the most computationally expensive tools.

• Old Way: Use the heavy "Hilbert lens" on the whole messy data. (Slow, but finds the waves).
• New Way: Use the standard "sorting" first, then apply the "Hilbert lens" only to the sorted results. (Fast, accurate, and finds the same waves).

In simple terms: The authors found a way to quickly identify the organized "waves" hiding inside the chaotic "noise" of water flowing past a ball. They showed that by looking at the patterns in a clever way, we can understand how turbulence moves and evolves, which helps engineers design better cars, planes, and ships that cut through the air and water more efficiently.

Technical Summary

1. Problem Statement

Turbulent flows, while appearing random, contain coherent structures (eddies, vortices) that drive their dynamics. Proper Orthogonal Decomposition (POD) is a standard tool for extracting these structures from experimental data. However, in complex flows like the wake of a sphere, identifying propagating structures is challenging.

• The Limitation of Standard POD: In strongly periodic flows, POD often yields pairs of modes representing the same structure at different phases. In complex, less periodic flows, these pairs are not inherently linked by the decomposition, making it difficult to isolate propagating waves.
• The Limitation of Standard HPOD: Hilbert Proper Orthogonal Decomposition (HPOD) addresses this by applying POD to the analytic signal (derived via the Hilbert transform), yielding complex modes with a $\pi/2$ phase shift that effectively capture propagating structures. However, HPOD introduces spectral leakage due to the inherent periodicity assumption of the Hilbert transform. Furthermore, performing HPOD on the entire turbulent fluctuation dataset is computationally expensive.
• The Gap: There is a need for a method that can efficiently identify POD mode pairs corresponding to propagating structures without the computational cost of full HPOD or the spectral leakage artifacts inherent in the standard HPOD approach.

2. Methodology

The study investigates the flow in the wake of a sphere at a Reynolds number of $Re_D = 7780$.

Experimental Setup:

• Facility: Vertical water tunnel with a 1.5 m test section.
• Object: A 40 mm diameter sphere, 3D printed with a smooth surface, mounted on a sting to minimize flow disturbance.
• Measurement: 2C-2D Multi-grid/Multi-pass Cross-correlation Digital (MCCD) Particle Image Velocimetry (PIV).
• Data: 12,600 instantaneous velocity fields ($u, v$) captured with a spatial resolution of 37 $\mu$m/px. The field of view extended from just behind the sphere to 5.2 diameters downstream.

Analytical Framework:

1. Standard POD: Applied to the turbulent velocity fluctuations ($u', v'$) using the method of snapshots to extract spatial modes ($\psi_k$) and temporal coefficients ($a_k$).
2. Hilbert POD (HPOD): Applied to the analytic signal of the velocity fluctuations. Since the data was not time-resolved, the Hilbert transform was applied in the streamwise direction ($x$) to treat spatial propagation as temporal evolution. A Hann window was applied to reduce spectral leakage.
3. Novel Approach (Hilbert Transform on POD Modes): Instead of performing HPOD on the raw data, the authors applied the streamwise Hilbert transform directly to the spatial POD modes ($\psi_k$).
   • Theoretical Basis: If a POD mode represents a propagating structure, its Hilbert transform should approximate the POD mode representing the same structure at a $\pi/2$ phase shift.
   • Goal: To identify pairs of POD modes ($k, j$) that are mutual best matches when one is transformed, thereby replicating the pairing found in HPOD but with lower computational cost.

Validation:

• Correlation coefficients ($R$) and $R^2$ values were calculated between HPOD modes and POD modes to identify pairs.
• Phase-averaging was performed on the identified POD pairs and HPOD modes to visualize the propagating structures (velocity, vorticity, Reynolds stress, TKE).

3. Key Contributions

• Methodological Innovation: The paper introduces a computationally efficient technique to identify propagating POD mode pairs by applying the Hilbert transform directly to the POD modes, bypassing the need to compute the analytic signal of the entire turbulent dataset.
• Validation of Equivalence: The study demonstrates that the mode pairs identified via the direct Hilbert transform on POD modes are identical to those identified via full HPOD.
• Characterization of Sphere Wake: The work provides a detailed breakdown of the coherent structures in a sphere wake at $Re_D = 7780$, distinguishing between different types of motion (flapping vs. pulsating) and their spatial scales.
• Analysis of Spectral Leakage: The paper quantifies the trade-off between the clarity of propagating structures in HPOD and the introduction of non-physical artifacts (spectral leakage) near boundaries and in the near-wake region.

4. Key Results

Mode Pairing and Correlation:

• HPOD vs. POD: The first HPOD mode (representing a propagating structure) was found to be a linear combination of the 1st and 2nd POD modes with a phase shift of $\approx \pi/2$. Similarly, the 2nd HPOD mode corresponded to the 4th and 5th POD modes, the 3rd HPOD to the 6th and 8th POD, and the 4th HPOD to the 9th and 10th POD.
• Direct Transform Success: Applying the Hilbert transform directly to the POD modes successfully identified these exact same pairs (e.g., Mode 1 matched Mode 2, Mode 4 matched Mode 5) with high correlation coefficients ($R > 0.85$).
• Efficiency: The direct transform method preserves the energy ranking of the original POD modes while identifying the propagating pairs, offering a significant computational advantage over full HPOD.

Physical Structures Identified:

1. Flapping Motion (Modes 1 & 2 / HPOD 1):
   • Represents large-scale planar-antisymmetric streamwise fluctuations and alternating transverse fluctuations.
   • Resembles "wake flapping."
   • HPOD emphasizes the downstream propagation, while POD shows stronger intensity immediately behind the sphere.
2. Streamwise Pulsation (Modes 4 & 5 / HPOD 2):
   • Characterized by V-shaped structures in the streamwise velocity and antisymmetric transverse fluctuations.
   • Represents a pulsating motion of the low-velocity region behind the sphere.
   • HPOD reveals these structures as more symmetric and uniform downstream compared to the asymmetric POD modes.
3. Shorter Wavelength Flapping (Modes 6 & 8 / HPOD 3):
   • Similar to the first pair but with a shorter wavelength.
   • Shows distinct transverse shifting in POD modes (indicating 3D motion) which is suppressed in HPOD due to the periodicity assumption.
4. Higher-Order Pulsation (Modes 9 & 10 / HPOD 4):
   • Similar to the 4th/5th pair but with smaller scales in the near wake, transitioning to V-shaped structures further downstream.

Energy Contributions:

• The first POD pair (Modes 1 & 2) contributes 8.24% of the total Turbulent Kinetic Energy (TKE).
• The corresponding HPOD mode contributes 7.70% to the TKE of the analytic signal.
• Higher-order HPOD modes generally contribute less to the analytic signal than their POD counterparts contribute to the raw data, as HPOD filters out non-propagating structures.

Comparison of Phase-Averaged Fields:

• POD Phase-Averaging: Captures both propagating and non-propagating features (e.g., strong near-wake structures, 3D asymmetries).
• HPOD Phase-Averaging: Produces cleaner, more uniform structures that emphasize propagation. However, it introduces artifacts:
  • Spectral Leakage: Causes structures near the sphere to appear coupled with downstream structures.
  • Loss of Near-Wake Detail: Non-propagating features confined to the immediate near wake are attenuated or misrepresented.

5. Significance and Conclusion

This research establishes that Hilbert Proper Orthogonal Decomposition is a powerful tool for isolating propagating coherent structures in complex turbulent wakes, but it comes with the caveat of spectral leakage and computational cost.

The primary significance of the work lies in the novel "Hilbert-on-POD" method. By applying the Hilbert transform directly to the spatial POD modes, researchers can:

1. Identify the correct pairs of POD modes representing propagating structures.
2. Avoid the computational expense of performing HPOD on the full dataset.
3. Avoid the spectral leakage artifacts introduced by transforming the raw turbulent fluctuations, as the POD modes have already been filtered by energy ranking.

The study concludes that while HPOD provides a clearer view of propagating dynamics, the direct application of the Hilbert transform to POD modes offers a robust, efficient, and less artifact-prone alternative for analyzing turbulent wakes, particularly when dealing with non-time-resolved experimental data. The identified structures (flapping and pulsating) provide a fundamental understanding of the transition and instability mechanisms in sphere wakes at moderate Reynolds numbers.