Data-Driven Wind Turbine Blade Fault Detection and Condition Assessment Using Frequency-Time Transmissibility Analysis

  • Xuefei Wang

Student thesis: Phd

Abstract

Wind turbines (WT) are increasingly deployed worldwide to harvest wind power from nature, and WT blades are the most crucial components in the WT systems. WT blades are subject to nonstationary time-varying loads whose information is usually unknown or hard to obtain. This poses great challenges to detect the WT blade damages. It is therefore necessary to develop effective condition monitoring and fault detection techniques for WT blade defects, before they propagate and result in fatal damages to entire blade structure. The blade failures often trigger a significant financial loss. An extensive literature review work is conducted. First, the WT structure, mechanism and classification are reviewed, followed by the WT blade structure, materials and loading conditions. Second, the WT blade failure modes and existing damage detection techniques are reviewed. Third, the existing remaining useful lifetime prediction techniques for WT blade are reviewed. It is found that there still exists great challenges to detect the blade faults, such as unknown loading conditions, noise contamination and the complex composite materials of the blade. To address these challenges, this project proposes methods from both frequency and time domain to detect the WT blade faults. In frequency domain, transmissibility function (TF) based approaches are developed, such as Fourier TF (FTF) and wavelet energy TF (WETF) methods. One of the main contributions of this report is that it is the first time to apply the WETF method to WT blades fault detection. Compared with the existing FTF method, WETF is advantageous in removing the impacts of external varying loads, requiring no need of excitation information, and is more robust to noise. Another contribution is that an extended wavelet package energy TF (WPETF) method is proposed. This technique is able to provide detailed resolutions in the high-frequency zones which can indicate richer information due to damage. As a similar signal decomposition structure, Hilbert-Huang transform (HHT) is also used to detect the WT blade damage. An extended HHT energy TF (HHETF) method is then proposed to compare with FTF, WETF and WPETF methods. In time domain, system identification based techniques are applied to detect the WT blade damages. A series of regularized models such as autoregressive with exogenous input (ARX), autoregressive moving average with exogenous input (ARMAX), output error (OE) and nonlinear polynomial models are tested respectively. Instead of the need for excitation information, another novelty of this new method is that different measurement signals are used to identify the blade structure system. This technique removes the impact of the external loadings. Unique features due to damage can be then extracted from the identified models, which are able to indicate the existence of the blade defects. The effectiveness of both TF based methods and system identification methods for blade damage detection is investigated using data collected directly from three naturally damaged industrial-scale WT blades. By comparison with existing conventional methods, the superiority of the proposed novel methods is demonstrated and analysed both numerically and experimentally.
Date of Award1 Aug 2023
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorWilliam Heath (Supervisor) & Long Zhang (Supervisor)

Cite this

'