Session: 41-03 AI and advanced methods for wind energy II

Submission Number: 178848

Data-Driven Decomposition of Horizontal-Axis Wind-Turbine Wakes 

This work presents an automatic, data-driven methodology for decomposing and characterizing CFD flow fields of horizontal-axis wind turbine wakes using clustering. 

The approach identifies and separates wake regions (near wake, high-shear layer, mixing/breakdown region, and far wake) without relying on threshold-based vortex criteria. Instead, it leverages unsupervised machine-learning techniques (clustering) that are data-adaptive and computationally efficient.

Multiple combinations of input features, feature selection, clustering algorithms and feature-scaling techniques are systematically compared to balance accuracy, robustness and computational costs.

Initially, we train and select the workflow on the isolated rotor configuration of the NREL 5-MW turbine, to assess performance in a setting comparable to the DTU 10-MW in [1]. This yields a preferred configuration (k-means with MinMax scaling) that balances accuracy and efficiency, aligning with [1].

Forwarding the isolated-rotor model on a case with an atmospheric boundary layer (ABL) setup led to degraded segmentation fidelity; accordingly, a second, ABL-calibrated model for the zero-yaw case was derived. This ABL model is subsequently forwarded to additional cases with the turbine working at different yaw conditions and operating points to investigate generalization capability of the approach.

The dataset used for the analysis is generated from high-fidelity large-eddy simulations (LES) with a Smagorinsky subgrid-scale model and the actuator line method (ALM) in OpenFOAM [2], providing a physically consistent description of wake dynamics.

Performance is quantified using standard clustering metrics, runtime and resource usage, and physics-based assessments against turbulent wake structures reported in the literature.

References

1. Tieghi, Lorenzo, Cerbarano, Davide, De Girolamo, Filippo, Barnabei, Valerio and Delibra, Giovanni. “A Novel Methodology for the Automatic Decomposition of HAWT Wakes With K–Means Clustering.” Wind Energy Vol. 28 (2025). DOI 10.1002/we.70030.

2. DeGirolamo, Filippo, Castorrini, Alessio, Morici, Vincenzo, Tieghi, Lorenzo and Rispoli, Franco. “Investigation on the Effect of Resolving Waves Motion in the Simulation of Off-shore Wind Farms.” 2024. DOI 10.1115/GT2024-124860.

Presenting Author: Giovanni Delibra Sapienza University of Rome

Presenting Author Biography: MSc in Mechanical Engineering
PhD in Theoretical and Applied Mechanics
Associate Professor at Department of Mechanical and Aerospace Engineering of Sapienza University of Rome
Research on Computational Methods, Machine Learning and Turbomachinery Design

Authors:

Andrea Carloni Sapienza University of Rome
Filippo De Girolamo Sapienza University of Rome
Valerio Barnabei Sapienza University of Rome
Lorenzo Tieghi University of Trento
Giovanni Delibra Sapienza University of Rome