Assessment of RANS Modeling of Jet Interaction in Fan-Array Wind Generator Flows

This study evaluates the capability of Reynolds-Averaged Navier-Stokes (RANS) modeling with a pressure-jump boundary condition to simulate jet interactions in a 10x10 fan-array wind generator, finding that while the approach effectively predicts global mean-flow structures and downstream velocity decay, it exhibits systematic limitations in accurately capturing near-field magnitude discrepancies and turbulence intensity due to eddy-viscosity closure constraints.

The Gist

The Big Picture: Building a "Weather Machine" in a Box

Imagine you are a scientist trying to test how a drone or a wind turbine handles a wild, stormy day. The problem is that traditional wind tunnels are like a calm, perfectly straight highway; the air flows smoothly and evenly. Real life, however, is more like a chaotic city street with gusts, swirls, and turbulence coming from every direction.

To fix this, engineers built a Fan-Array Wind Generator (FAWG). Think of this as a giant wall made of 100 small, independent computer fans (a 10x10 grid). By turning different fans on and off at different speeds, they can create a "storm" inside a lab that mimics the messy, turbulent air of the real world.

The Problem: While these machines are great for creating the wind, nobody knew how to simulate them on a computer. If you want to design a new drone, you don't want to build a physical fan wall every time; you want to run a computer simulation. But simulating 100 individual fans interacting with each other is incredibly hard for a computer.

The Goal: The authors of this paper asked, "Can we use a standard, fast computer model (called RANS) to predict how these 100 fans will mix their air streams together, and will it be accurate enough to be useful?"

---

The Method: The "Actuator" Trick

To simulate 100 fans without turning the computer into a toaster, the researchers used a clever shortcut.

Instead of modeling the tiny blades spinning inside every single fan (which would take forever), they treated each fan like a magical invisible door.

• The Analogy: Imagine a door that doesn't have a handle or hinges. When air tries to push through it, the door magically adds a "kick" of energy to the air, pushing it forward.
• They programmed this "kick" based on the fan's manual (how much push it gives at different speeds). This allowed them to simulate the whole 10x10 wall in a reasonable amount of time.

They tested two ways to draw these "doors":

1. The Flat Sheet: Just a thin line where the air gets kicked.
2. The Ducted Tube: A more realistic version where the fan is inside a little plastic tube (the housing), accounting for the friction of the walls.

---

The Findings: What Worked and What Didn't

The researchers compared their computer predictions against real-world measurements taken from the actual fan wall. Here is what they found:

1. The "Big Picture" was Good

The computer model was very good at predicting the overall shape of the wind. It correctly showed where the fast air streams (jets) would merge together and how the wind would slow down as it traveled further away from the fans.

• Analogy: If you threw 100 water hoses at a wall, the computer could accurately predict where the big puddle would form and how far the water would spray.

2. The "Fine Details" were Messy

While the big picture was right, the model struggled with the turbulence (the chaotic, swirling bits).

• The Issue: The computer tended to smooth things out too much. In the real world, the air near the fans is a violent, churning mess. The computer model made it look a bit too calm and uniform.
• Why? The math they used (RANS) is like looking at a crowd from a helicopter. You can see the crowd moving, but you can't see the individual people bumping into each other. The model averages out the chaos, missing the sharp spikes in turbulence intensity.

3. The "Speed" vs. "Chaos" Surprise

They tested what happened if they turned the fans faster or slower.

• Velocity: As expected, turning the fans faster made the wind blow harder.
• Turbulence: Surprisingly, changing the fan speed didn't change the turbulence levels much in the simulation.
• The Analogy: Imagine a blender. If you turn the speed up, the water spins faster, but the amount of bubbles (turbulence) created by the blades hitting the water stays roughly the same relative to the speed. The model showed that the turbulence is created by the fans crashing into each other (jet interaction), not just by how fast the fans are spinning.

4. The "Inlet" Didn't Matter

They also tried changing the "background noise" (turbulence) of the air before it even hit the fans.

• The Result: It didn't matter. The fans were so powerful and the mixing so intense that the initial conditions were washed away. The chaos was generated inside the machine, not imported from outside.

---

The Real-World Test: The Flat Plate

To prove this matters, they put a simple square plate (like a tiny wing) in the wind.

• Scenario A (Normal Wind): The plate sits in smooth, even wind.
• Scenario B (FAWG Wind): The plate sits in the chaotic, bumpy wind from the fan array.

The Shocking Result: Even though the average wind speed was the same in both cases, the plate in the FAWG wind experienced 108% more lift and 380% more drag.

• The Takeaway: It's not just about how fast the wind is blowing; it's about how messy it is. The bumpy, uneven wind hit the plate in violent, localized bursts, creating huge forces that a smooth wind tunnel would never predict. This is crucial for designing drones and turbines that won't break in real storms.

---

The Conclusion

The Good News: This computer modeling method is a fast, efficient, and "good enough" tool for engineers. It can predict the general flow of these complex fan arrays without needing a supercomputer.

The Bad News: It's not perfect. It smooths out the violent, chaotic details of the turbulence. If you need to know exactly how a tiny vibration will affect a sensor, this model might miss it.

The Bottom Line: We now have a way to simulate these "fan walls" on a computer. This helps us understand how real-world turbulence affects aircraft and wind turbines, moving us one step closer to designing machines that can survive the wildest weather nature can throw at them.

Technical Summary

1. Problem Statement

Conventional wind tunnels are limited in their ability to reproduce the highly turbulent, non-uniform, and transient flow conditions found in real atmospheric environments. Fan-Array Wind Generators (FAWGs) have emerged as a solution, using independently actuated fans to generate controlled, spatially varying turbulent inflows. While experimental characterization of FAWGs is advancing, numerical modeling (CFD) of these systems remains largely unexplored.

The primary challenges in modeling FAWGs include:

• The complexity of simulating hundreds of interacting jets and strong turbulent shear layers.
• The computational infeasibility of resolving individual blade geometries for large arrays (e.g., 100+ fans).
• The lack of validated frameworks to predict mean flow structures and turbulence intensity (TI) in these highly mixing-dominated flows.

2. Methodology

The study assesses the capability of Reynolds-Averaged Navier–Stokes (RANS) modeling to predict jet interactions in a 10 × 10 FAWG (100 axial-flow fans).

• Fan Representation: Instead of resolving blade geometry, the authors used a pressure-jump boundary condition (actuator disk concept). The pressure rise was defined by a fourth-degree polynomial derived from manufacturer performance data, allowing the operating point to be determined implicitly.
• Geometric Models: Two fan representations were compared:
  1. Surface Model: Fans modeled as zero-thickness annular surfaces.
  2. Ducted Model: Fans modeled as annular fluid regions with explicit hub and casing walls to capture confinement effects.
• Turbulence Modeling: The $k-\omega$ Shear Stress Transport (SST) model was employed. This was chosen as a computationally feasible alternative to Large Eddy Simulation (LES), which would be too expensive for the large number of jet cores.
• Validation: Numerical predictions were validated against experimental data (axial velocity and TI) from a reference study (Cao et al.) for two activation patterns (B and D).
• Sensitivity Analysis: The study examined the effects of:
  • Grid resolution (Grid Convergence Index via Richardson extrapolation).
  • Fan operating speed (scaled via affinity laws).
  • Inlet turbulence conditions (varying TI and turbulent viscosity ratio).
• Demonstrative Application: A low-aspect-ratio flat plate ($AR=1.0$) was simulated in both FAWG-generated flow and uniform freestream to assess aerodynamic impacts.

3. Key Results

A. Flow Field Prediction (Velocity & Turbulence)

• Global Topology: RANS modeling successfully captured the global jet interaction topology and the downstream velocity decay. The mean flow structure was qualitatively consistent with experiments.
• Discrepancies:
  • Near-Field: The model underpredicted peak velocity magnitudes in the injection region and produced smoother profiles compared to the sharp experimental peaks.
  • Turbulence Intensity (TI): While the general trend was captured, TI predictions showed larger deviations (Mean Relative Error $\approx$ 17.6% in the core, 15.8% in the periphery). The model failed to capture the pronounced mid-field peaks observed experimentally, indicating that the eddy-viscosity closure diffuses turbulence more strongly than reality in highly mixing-dominated flows.

B. Influence of Fan Representation

• Surface vs. Ducted: The Surface Model produced sharper velocity gradients and more energetic near-field mixing. The Ducted Model (with hub/casing) resulted in lower jet velocities due to confinement losses and a more gradual, diffused merging process. Both models captured the inward deflection of jet cores, but the Ducted model yielded a smoother, more uniform velocity field.

C. Sensitivity to Operating Conditions

• Fan Speed: Increasing fan speed proportionally increased axial velocity levels. However, the downstream decay rates remained identical, suggesting that flow development is governed by jet interaction and geometric spreading rather than absolute momentum.
• Turbulence Intensity: Surprisingly, varying the fan speed had no effect on the TI profiles.
• Inlet Turbulence: Varying inlet TI (1% to 25%) and turbulent viscosity ratios had negligible impact on the downstream TI distribution beyond $z/W \approx 4$.
• Conclusion: The turbulence characteristics of FAWG flows are governed primarily by internal mechanisms (shear layer development and jet-jet interactions) rather than inlet conditions or operating speed.

D. Aerodynamic Impact (Flat Plate Application)

Simulating a flat plate in FAWG flow revealed significant aerodynamic differences compared to uniform freestream:

• Lift Coefficient ($C_L$): Increased by 108%.
• Drag Coefficient ($C_D$): Increased by 380%.
• Mechanism: The non-uniform inflow caused localized jet impingement (especially near the leading edge), creating strong spatial variations in pressure and shear stress. This resulted in highly non-uniform surface loading, demonstrating that FAWG inflows fundamentally alter aerodynamic performance compared to standard wind tunnel conditions.

4. Key Contributions

1. First RANS Assessment: Provided the first comprehensive numerical assessment of jet interactions in a large-scale FAWG using RANS.
2. Validated Framework: Established a computationally efficient framework using pressure-jump boundary conditions to simulate large fan arrays without resolving individual blades.
3. Insight into Turbulence Generation: Demonstrated that FAWG-generated turbulence is self-sustaining through internal shear-layer interactions, making the flow largely insensitive to inlet turbulence specifications or fan speed scaling (regarding TI).
4. Aerodynamic Implications: Quantified the dramatic increase in lift and drag forces on lifting surfaces when exposed to FAWG-generated turbulent inflow, highlighting the limitations of using uniform freestream data for such applications.

5. Significance and Limitations

• Significance: The study confirms that RANS is a viable tool for predicting the mean-flow structure of FAWG systems, offering a practical alternative to expensive LES or fully resolved blade simulations. It provides a critical foundation for designing windshapers and testing aerospace components in realistic turbulent environments.
• Limitations:
  • Turbulence Modeling: The eddy-viscosity assumption (RANS) struggles to accurately predict localized turbulence intensity peaks and anisotropic mixing in shear layers.
  • Steady-State Assumption: The study focused on steady activation patterns; it does not capture transient effects associated with duty-cycle modulation or gust generation, which would require unsteady simulations (e.g., URANS or LES).
  • Near-Field Accuracy: Discrepancies in the immediate near-field suggest that the simplified pressure-jump model may not fully capture the complex wake dynamics of individual fan units.

Future Work: The authors suggest extending this framework to scale-resolving unsteady simulations (LES/DES) to capture transient effects and prescribed swirl, which are crucial for dynamic gust modeling.