Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article

Degree

Doctor of Philosophy

Program

Civil and Environmental Engineering

Supervisor

Bitsuamlak, Girma T.

2nd Supervisor

Dai, Kaoshan

Affiliation

Sichuan University

Joint Supervisor

3rd Supervisor

Lu, Wensheng

Affiliation

Tongji University

Co-Supervisor

Abstract

Hybrid simulation (HS) is a promising technique for studying wind turbines due to the presence of scaling errors in wind tunnel testing. However, HS of wind-loaded structures is limited by the current practice of using lower-accuracy, "pre-calculated" aerodynamic loads, which uncouple the aerodynamic loading from the structural response. This thesis presents six stand-alone studies that collectively build towards a novel HS framework that employs a computational fluid dynamics (CFD) based surrogate model to generate higher-accuracy aerodynamic loads within the HS loop. An experimentally-validated residential wind turbine model equipped with an external damping system was used to illustrate the proposed framework.

Development of the proposed HS framework occurred in the following stages: firstly, the limitations of the existing HS framework were identified and quantified through a numerical case study of an industrial wind turbine. The aerodynamic surrogate model was developed through CFD simulations of airfoils to determine optimal test parameters, followed by the identification of an optimal convolutional neural network (CNN) architecture to act as the surrogate model. The test case for the HS was chosen based on a numerical study of semi-active tuned mass damper (STMD) systems for reducing turbine tower vibrations. A dynamic numerical model was developed of a 1.1 m residential wind turbine on a rotating base validated through a set of full-scale aeroelastic wind tunnel tests.

These components were ultimately combined for a series of twelve artificial HS to compare the effectiveness of the proposed surrogate model-based HS framework to the existing "pre-calculated" technique. A number of CFD simulations were performed to generate the training data for the CNN surrogate model, which was combined with the dynamic turbine model to act as the numerical HS substructure. This was paired with the optimized STMD acting as the virtual experimental substructure, including simulated equipment delays and measurement errors. The results of these HS indicate that the proposed framework has improved aerodynamic accuracy and aeroelastic fidelity compared to the existing technique, though it faces hurdles from computationally costly CFD data generation. This framework offers a promising tool for future HS of wind turbines and other wind engineering applications.

Summary for Lay Audience

Hybrid simulation (HS) is a structural engineering research technique that tests a structure as two components: a physical component in a laboratory and a simulated component on a computer. This allows complex, hard-to-simulate structural elements to be tested without having to recreate the entire structure. Wind turbines (WT) are a promising option for this technique because the standard method for testing wind-sensitive structures -- wind tunnel testing -- struggles to accurately capture the resulting wind forces, creating the need for an alternative physical testing method.

Unfortunately, existing HS of WTs have been limited. Since the HS components run in tandem with each other, the program hosting the simulated component must run quickly -- delays in the program will cause de-syncing with the physical component and lead to inaccurate results. Thus, in an HS of a WT, the simulated component must be able to quickly predict the wind forces. Various methods exist to predict wind forces, however, they generally prioritize either speed or accuracy, not both. Therefore, HS of WTs have historically been limited to low-accuracy wind force predictions, reducing the technique's utility.

This thesis explores a method to incorporate an improved simulated component into HS, using a tool called a neural network (NN) to act as a surrogate model. NNs are artificial intelligences that are trained to predict outputs based on inputs without performing the underlying calculations, acting as a high-speed surrogate to other models. Computational fluid dynamics (CFD) is a time-consuming method to simulate high-accuracy wind forces; by training a NN on CFD data, a program for predicting wind forces that is both high-speed and high-accuracy can be created and incorporated into HS of WTs.

This thesis presents six papers that detail the investigation and development of this surrogate model, culminating in a series of HS of a WT equipped with a vibration-suppressing damping system to investigate the effectiveness of the model. Ultimately, the upgraded HS technique presented here is identified as a promising option for effectively studying wind-loaded structures which warrants further investigation.

Creative Commons License

Creative Commons Attribution 4.0 License
This work is licensed under a Creative Commons Attribution 4.0 License.

Available for download on Saturday, October 01, 2022

Share

COinS