In fluid dynamics, no object is an island. Whether it is a cluster of wind turbines, a swarm of drones, or a row of skyscrapers, the presence of one structure fundamentally alters the environment for its neighbors. This phenomenon, known as wake interference, occurs when the turbulent, low-velocity "shadow" cast by an upstream body degrades the performance and structural integrity of everything downstream.
Recent research has shifted from analyzing isolated components to treating multi-unit arrays as coupled systems. By combining high-fidelity simulations with graph-based machine learning, researchers are finding ways to predict, model, and even exploit these complex interactions.
The Shadow of the Upstream
When a wind turbine or propeller extracts energy from the air, it leaves behind a wake characterized by two primary features: a velocity deficit (slower air) and increased turbulence. For a downstream unit, this means there is less energy available to harvest and a higher risk of mechanical fatigue.
In tandem propeller configurations—common in emerging eVTOL aircraft—the rear rotor often operates entirely within the slipstream of the front rotor. Alex Zanotti, Francesco Caccia, and colleagues have shown in Acoustic and aerodynamic evaluation of POLIMI tandem propellers in hover and cruise flight conditions within GARTEUR AG26 that this interaction can lead to a 30% reduction in thrust and a 20% loss in power efficiency. Similarly, in coaxial drone rotors, Jianping Yin and Karl-Stéphane Rossignol observed that the lower rotor suffers substantial thrust loss because it cannot be treated as an isolated component; its performance is inextricably linked to the wake of the upper rotor.
The Physics of Interference
The interference is not merely a reduction in speed. Masayuki Norita and Kenichi Oyama highlight that standard models often neglect wake rotation or "swirl." As a turbine spins, it imparts angular momentum to the air. Failing to account for this swirl leads to significant overestimations of efficiency, as the rotation represents a kinetic energy loss that bridges the gap between theoretical limits and real-world performance.
Beyond energy loss, wakes create periodic fluctuations. Chun ming Qu and De sheng Miao demonstrated that the wake from large-scale turbines creates "3P" (three-per-revolution) frequency oscillations. These oscillations can sync with the structural frequencies of downstream turbines, causing excessive vibration and unplanned hardware failure. This same principle applies to civil engineering: Gang Hu and Lingbo Liu have used machine learning to show how the proximity of tall buildings can dangerously amplify wind-induced loads on neighboring structures.
The Multi-Fidelity Modeling Gap
The central challenge in this field is the "reality gap." High-fidelity tools like Large-Eddy Simulations (LES) can capture these intricate viscous interactions but are computationally ruinous, often taking 24 hours to simulate a single building or turbine layout. Conversely, fast "mid-fidelity" codes often struggle with the non-linearities of overlapping wakes.
To solve this, researchers are adopting multi-fidelity frameworks. Gianluigi Rozza and others utilize reduced-order modeling to bridge the gap, while George L. Barnett and Simon W. Funke proposed a hybrid approach for tidal turbine arrays: using fast analytical models for global searches and high-fidelity models for local refinement. This allows for optimization that accounts for "flow blockage"—where the cumulative resistance of a farm actually diverts the incoming fluid flow.
Graph Neural Networks and Attention
A breakthrough in predicting these interactions has come from treating multi-unit systems as graphs. In a wind farm, turbines are nodes, and the wind flow represents the edges.
Lars Ødegaard Bentsen and Paal E. Engelstad pioneered the use of Graph Neural Networks (GNNs) with attention mechanisms to capture wake losses. By using "attention weights," the model can learn which upstream turbines are most responsible for the power deficit at a specific downstream node. Siyi Li and Aldo A. Faisal extended this with Graph Transformers, achieving 99.8% accuracy in power prediction while running orders of magnitude faster than traditional simulations.
From Prediction to Active Management
If we can predict the wake, we can steer it. Wake steering is the practice of intentionally misaligning an upstream turbine (yawing) to deflect its wake away from downstream units. While this reduces the efficiency of the individual upstream turbine, it increases the aggregate output of the farm.
Kaixuan Chen and Yonghua Song demonstrated that deep learning-aided Model Predictive Control (MPC) can coordinate these deflections in real-time. Similarly, Damien Ernst and Florence Carton showed that reinforcement learning policies can increase total energy production by up to 14% by managing these interdependent wake patterns across all wind directions.
This shift—from viewing interference as an unavoidable nuisance to a manageable system variable—is the defining trend of modern aerodynamic research. Whether in the design of Frank Neumann’s optimized wind farm layouts or Jolan Wauters’s "first-time-right" drone designs, the goal remains the same: mastering the complex dialogue between multiple bodies and the air they share.