Συντάχθηκε 06-07-2026 12:42
Ενημερώθηκε:
06-07-2026 12:44
Τόπος: Κ2 - Κτίριο ΧΗΜΗΠΕΡ
Σύνδεσμος τηλεδιάσκεψης
Έναρξη: 08/07/2026 10:00
Λήξη: 08/07/2026 11:00
ANNOUNCEMENT OF PRESENTATION OF POSTGRADUATE THESIS
First Name/Surname: Grigorios Spiliopoulos
Student Identification Number: 2024057019
Date: 08/07/2026
Time: 10:00 (GMT +3)
Room / Zoom Link: https://tucgr.zoom.us/j/91477145205?pwd=F3TMap3SobWa25XZHko4P8O8kla93b.1
Meeting ID: 914 7714 5205
Password: 878737
Title: “Algorithmic Synthesis and Extrapolation of Wind Tunnel Aerodynamic Coefficients for Single-Axis Tracker Systems”
Supervisor: Prof. Theocharis Tsoutsos
Three-member committee:
1. Prof. Theocharis Tsoutsos
2. Prof. Apostolos Voulgarakis
3. Assoc. Prof. Alexandros Stefanakis
Abstract:
The reliability of the structural integrity of single-axis photovoltaic (PV) tracking systems at a utilityscale, are highly dependent upon accurate aerodynamic pressure coefficients determined through
wind tunnel testing. The physical scaling limitations presented by the prototype and the economy of
testing, restrict the ability to conduct wind tunnel tests for each row pitch, ground clearance and tilt
angle. This, results in critical gaps in the existing aerodynamic pressure coefficient data. This thesis
presents a computational framework, referred to as WindFit, that provides the means to synthesize
the transient aerodynamic loads for the geometries of PV trackers that have not been tested.
The methodology employs localized polynomial regression to reconstruct the incomplete azimuth and
tilt data associated with tracker operation. Geometric synthesis will be achieved by projecting
structural parametric data onto a Cartesian coordinate system. Geometries that fall into an empirical
convex hull will be resolved using Barycentric Coordinate Mathematics, while those that fall outside
the convex hull, will utilize the Virtual Anchor Vertex protocol within the framework, to constrain
the extrapolated data. This topology-constrained boundary gradient approach prevents the numerical
divergence and reversed aerodynamic weighting that are associated with traditional unconstrained
extrapolation methods.
The computational effectiveness of the framework is verified by two benchmark approaches, which
use deterministic Inverse Distance Weighting (IDW) and stochastic Gaussian Process Regression
(GPR) models. The comparative spatial validation of the models shows that IDW operates as a low
pass filter, creating spatial blurring that greatly under-predicts the high-end boundary loads, as well
as the Venturi amplifications. The GPR model produces the same high-fidelity precision as WindFit,
but due to it's reliance on infinite mathematical smoothness, GPR removes sharp aerodynamic
discontinuities and dramatically under-predicts detached tip vortexes during extrapolation.
Finally, WindFit integrates local geometric interpolation with boundary conditions based upon
physics to accurately model complex flow separation phenomena. In addition, it uses Barycentric
algebraic weights to eliminate the extensive latency involved with performing machine learning
matrix inversions, thus making this approach highly suited for real-time parametric structural design
