Impact of perturbed eddy-viscosity modeling on stability and shape sensitivity of the hydro-turbine vortex rope using linearized Reynolds-averaged Navier-Stokes equations

This study demonstrates that consistently linearizing the eddy-viscosity turbulence model is essential for accurately capturing shape sensitivities of hydro-turbine vortex ropes, as neglecting these perturbations leads to incorrect sensitivity trends despite having minimal impact on the global mode's eigenvalues and eigenmodes.

The Gist

Imagine a hydroelectric power plant as a giant, high-speed water slide. When the water flows perfectly, everything is smooth. But when the plant is running at "part load" (not at full power), the water doesn't just slide; it starts to twist and turn into a giant, spiraling rope of turbulence called a vortex rope.

This vortex rope is a nuisance. It vibrates the turbine, makes loud noises, and can even damage the machinery. Engineers want to stop it, but they need to know exactly where to tweak the shape of the pipe to calm the water down.

This paper is about a new way of figuring out those tweaks, and it reveals a surprising truth about how we model the invisible "friction" of water.

The Problem: The "Frozen" vs. "Living" Model

To predict how to fix the vortex rope, scientists use computer simulations. These simulations try to guess how the water will behave if you change the shape of the pipe slightly.

There are two ways to do this math:

1. The "Frozen" Model (The Old Way): Imagine the water's internal friction (called eddy viscosity) is like a block of ice. It's hard, solid, and doesn't change. In this model, when you change the shape of the pipe, the computer assumes the water's internal friction stays exactly the same, frozen in place. It's a bit like trying to predict how a car handles a turn by assuming the tires are made of concrete.
2. The "Perturbed" Model (The New Way): Imagine the water's internal friction is like a living, breathing organism. When you change the shape of the pipe, the water reacts. The friction changes, swirls, and adapts. This model lets the "friction" wiggle and change along with the water flow.

The Big Surprise: The "Ghost" Effect

The researchers ran both models to see which one gave the right answer for fixing the vortex rope. Here is the twist:

• On the "What" (The Vortex Itself): Both models predicted almost the exact same thing about the vortex rope itself. They agreed on how fast it spins and how unstable it is. It was as if both models were looking at the same storm cloud and agreeing on its size and speed.
• On the "How" (The Fix): This is where they completely disagreed.
  • The Frozen Model said: "To stop the vibration, you need to make the pipe thinner at this specific spot."
  • The Perturbed Model said: "No, to stop the vibration, you need to make the pipe thicker at that exact same spot."

They were giving opposite directions!

The Detective Work: Why the Difference?

The authors dug deeper to find out why. They broke the problem down into two parts:

1. The Direct Effect: How the shape change immediately hits the water. (Both models agreed on this).
2. The Indirect Effect (The Base Flow): How the shape change alters the background turbulence, which then changes the water's behavior.

The Analogy:
Think of the water flow like a crowded dance floor.

• The Frozen Model assumes that if you move a pillar (the pipe shape), the dancers (the water) just bump into it and keep dancing exactly as they were before.
• The Perturbed Model realizes that if you move the pillar, the dancers get jostled, they change their rhythm, and the whole crowd's energy shifts.

The researchers found that the Indirect Effect is the real boss. The "Frozen" model ignored the fact that moving the pipe changes the background turbulence, which is actually the most important part of the solution. Because it ignored this "living" reaction, it gave the wrong advice on how to fix the pipe.

The Proof: Real Life vs. Computer

To settle the argument, they compared their computer models with real-world experiments.

• When they actually built a pipe with a thicker bump (the "Perturbed" prediction), the vibration stopped.
• When they tried the "Frozen" prediction (making it thinner), it didn't work as well.

The "Living" model was right. The "Frozen" model was wrong.

The Takeaway

This paper teaches us a valuable lesson about engineering and physics: You can't treat turbulence like a static object.

Even though the "frozen" model looks good enough for predicting what is happening (the speed of the vortex), it fails miserably when you need to know how to fix it. If you want to design better, safer hydro-turbines, you have to use the model that acknowledges that the water's internal friction is alive, reactive, and constantly changing.

In short: To fix a wobbly water slide, you have to understand that the water itself is reacting to your changes, not just sitting there like a frozen statue.

Technical Summary

Here is a detailed technical summary of the paper "Impact of perturbed eddy-viscosity modeling on stability and shape sensitivity of the hydro-turbine vortex rope using linearized Reynolds-averaged Navier–Stokes equations."

1. Problem Statement

The study addresses the challenge of controlling global instabilities in fully turbulent flows, specifically the vortex rope phenomenon in Francis turbine draft tubes operating at part-load conditions. The vortex rope is a precessing vortex core that causes severe pressure pulsations and structural vibrations.

While Linear Stability Analysis (LSA) and adjoint-based sensitivity analysis are established tools for flow control and shape optimization, their application to turbulent flows remains controversial. A critical unresolved issue is the treatment of turbulence modeling during linearization:

• Frozen Eddy-Viscosity Model: Assumes turbulent quantities (eddy viscosity, $k$, $\varepsilon$) remain constant during perturbations ($\tilde{\nu}_t = 0$). This is the standard approach when turbulence model equations are not explicitly linearized.
• Perturbed Eddy-Viscosity Model: Explicitly accounts for fluctuations in turbulence model quantities ($\tilde{\nu}_t \neq 0$) by linearizing the turbulence closure equations (e.g., $k-\varepsilon$).

The paper investigates whether neglecting these perturbations (the frozen approach) leads to erroneous predictions regarding shape sensitivity, which is crucial for designing geometric modifications to stabilize the flow.

2. Methodology

The authors developed a physics-based framework combining Linearized Reynolds-Averaged Navier–Stokes (RANS) equations with a standard $k-\varepsilon$ turbulence closure.

• Base Flow Tuning: The base flow was computed using 2D axisymmetric RANS simulations. To ensure physical relevance, the inlet boundary conditions (specifically turbulence quantities $k, \varepsilon, \nu_t$) were tuned via a scaling factor. This tuning forced the RANS-based LSA to match the bifurcation point (onset of instability) observed in 3D Unsteady RANS (URANS) simulations and experiments.
• Linear Stability Analysis (LSA): The RANS equations were linearized around the base flow. Two models were compared:
  1. Frozen Model: The linearized operator acts only on mean velocity and pressure; turbulence variables are fixed.
  2. Perturbed Model: The linearized operator includes coupled equations for fluctuations in $k$, $\varepsilon$, and consequently $\nu_t$.
• Shape Sensitivity Analysis: Using perturbation theory, the authors derived the sensitivity of the eigenvalue (growth rate $\sigma$) to wall-normal shape deformations ($d\sigma/dn$).
  • The total sensitivity was decomposed into Feedback contributions (changes in the linear operator due to shape) and Base-flow contributions (changes in the base flow due to shape).
  • The base-flow contribution was further decomposed into physical mechanisms (production, advection, diffusion, etc.) to identify the dominant physics.
• Validation: Results were validated against experimental data from a Francis-99 turbine model and 3D URANS simulations. A stochastic amplitude model was used to extract growth rates from experimental pressure signals.

3. Key Contributions

1. Decoupling of Eigenvalues and Sensitivities: The study demonstrates a fundamental decoupling between the impact of turbulence modeling on eigenvalues (stability) versus shape sensitivities (control).
2. Decomposition of Sensitivity Mechanisms: The authors provide a rigorous decomposition of shape sensitivity, showing that while the frozen and perturbed models agree on the dominant mechanisms (velocity production and advection), the perturbed model captures additional, critical mechanisms related to eddy-viscosity fluctuations.
3. Experimental Validation of Sensitivity: This is one of the first studies to validate RANS-based shape sensitivities against experimental data in a fully turbulent global instability, proving that linearized turbulence models are essential for accurate design guidance.

4. Key Results

A. Impact on Eigenvalues and Eigenmodes

• Marginal Effect: The perturbed eddy-viscosity model has a negligible impact on the eigenvalues (growth rate and frequency) and eigenmode shapes compared to the frozen model.
• Reasoning: The linearized $k-\varepsilon$ equations appear to be quasi-decoupled from the momentum equations regarding the global mode's frequency and growth rate. The eigenvalue is determined primarily by the linear operator acting on the mean flow, which remains similar in both models.

B. Impact on Shape Sensitivity

• Significant Divergence: In stark contrast to the eigenvalues, the shape sensitivities differ drastically between the two models.
  • Frozen Model: Predicts that thinning the centerbody at the point of maximum thickness stabilizes the flow.
  • Perturbed Model: Predicts that thickening the centerbody at the same location stabilizes the flow.
• Experimental Confirmation: Experimental data confirmed the perturbed model's prediction. Increasing the centerbody thickness (bulge) reduced the growth rate, while the frozen model's prediction was entirely opposite.

C. Physical Mechanism Analysis

The discrepancy arises because the base-flow contribution dominates the total shape sensitivity.

• Frozen Model: Only captures changes in velocity production and advection. It predicts that thickening increases the wake deficit (destabilizing via advection), leading to a net destabilizing effect.
• Perturbed Model: Captures additional terms related to eddy-viscosity fluctuations.
  • Thickening the centerbody increases shear, which increases the base-flow eddy viscosity ($\nu_t$).
  • This increased $\nu_t$ enhances the diffusion of coherent fluctuations and modifies $k$ and $\varepsilon$ production.
  • These additional mechanisms provide a strong stabilizing effect (damping) that outweighs the destabilizing wake effects, reversing the net sensitivity sign.

5. Significance and Conclusion

• Critical for Control Design: The study concludes that while the frozen eddy-viscosity model may suffice for predicting whether a flow is unstable (eigenvalues), it is insufficient and potentially misleading for designing flow control strategies (shape optimization).
• Necessity of Linearized Turbulence Models: For gradient-based shape optimization in turbulent flows, it is imperative to consistently linearize the turbulence model equations to capture the feedback between shape deformation and the turbulent background field.
• Broader Implications: The findings suggest that the "frozen" assumption, often used for convenience in turbulent LSA, can lead to incorrect physical interpretations and suboptimal or counter-productive control designs. The framework presented offers a validated path for sensitivity-based design of hydro-turbines and other turbulent global instability problems.