11/08/2021
By Sokny Long

The Francis College of Engineering, Department of Mechanical Engineering, invites you to attend a Doctoral Defense by Jaclyn Solimine on “Development of Robust, Data-Driven Damage Identification Techniques for the Passive Acoustics-Based Structural Health Monitoring of Wind Turbine Blades.”

Doctoral Candidate: Jaclyn Solimine
Date: Monday, Nov. 22, 2021
Time: 2 to 3:30 p.m. EST
Location: This will be a virtual defense via Zoom. Those interested in attending should contact Jaclyn_Solimine@student.uml.edu (Candidate) and Murat_Inalpolat@uml.edu (Advisor) at least 24 hours prior to the defense to request access to the meeting.

Committee Chair (Advisor):
Murat Inalpolat, Ph.D., Associate Professor, Department of Mechanical Engineering, University of Massachusetts Lowell

Committee Members:

  • Christopher Niezrecki, Ph.D., Professor, Department of Mechanical Engineering, University of Massachusetts Lowell
  • David J. Willis, Ph.D., Associate Professor, Department of Mechanical Engineering, University of Massachusetts Lowell
  • Yi Guo, Ph.D., Senior Research Scientist, National Renewable Energy Laboratory

Abstract:
Reliable structural health monitoring (SHM) systems for wind turbine blades are becoming an increasingly emergent need as installed wind power capacity increases around the world. Recently, airborne acoustics-based SHM methods have emerged as a promising contender for a viable wind turbine blade SHM system due to their low cost, ease of implementation, and sensitivity to a variety of damage modes. In passive acoustic excitation-based approaches, damage is detected by analyzing changes in the naturally occurring sound within the blade cavity. Although effective at detecting a wide range of damage modes, passive excitation based approaches are susceptible to fluctuations in uncontrolled acoustic excitation that can alter the damage signal and instigate noise. For this reason, existing passive damage detection techniques have resorted to specialized, damage-mode specific features or algorithms designed under the context of stationary healthy-baseline conditions. These techniques inherently lack the robustness and adaptability necessary for an operational implementation. With the aim of overcoming these deficiencies, this dissertation details the development of a novel, data-driven technique for the passive acoustic excitation-based SHM of wind turbine blades.

The technique introduces two complementary approaches: the stationary approach, which leverages the airflow-induced passive acoustic excitation of the blade cavity for the detection of cavity-penetrating damage modes, and the dynamic approach, which leverages non-stationary patterns in the structurally-induced passive acoustic excitation of the blade cavity for the detection of fatigue-related damage modes. In contrast to existing approaches, this technique relies solely on a data-driven approach for damage detection in both the stationary and dynamic contexts. By limiting assumptions made concerning the properties of the system excitation and damage signal throughout model development, the adaptability and operational feasibility of the approach is increased.

The stationary approach is demonstrated on a representative wind turbine blade section and subjected to airflow-induced passive acoustic excitation in a wind tunnel. Autoencoder networks are developed and proven to enhance the differentiability of the healthy and damage-case samples regardless of fluctuations in airflow-induced excitation. Using this technique, hole-type damage on the structure ranging between 0.16cm and 1.27cm in diameter is detected with 89% accuracy, while crack-type damage between 1.27 and 5.08cm in length is detected with 99% accuracy. This is accomplished (i) without the use of labeled or damage-case samples during model training and (ii) without any knowledge of the acoustic excitation for any given sample.

The dynamic approach is demonstrated on a 63m wind turbine blade undergoing edgewise fatigue testing. Cepstral-domain features and unsupervised LSTM-autoencoder networks are used to enhance the differences between healthy and damage-case acoustic sequences within the blade cavity. Autocorrelation-based features are then used to characterize the anomalies in the acoustic sequences and subsequently detect damage. Using this novel approach, the onset of fatigue-related blade damage is detected up to 120,000 cycles before a conventional strain-based SHM system. In addition, the ability of the autocorrelation-based features to distinguish between different damage-related acoustic classes via an unsupervised clustering technique is demonstrated. This investigation is the first to detect wind turbine blade damage via the analysis of structurally-induced passive acoustic excitation and offer avenues for the data-driven characterization of damage.

Finally, avenues for adapting the stationary and dynamic approaches to a full-scale, operational application are presented. Data-driven and logic-based techniques for model updating, model adaptation, and environmental and operational variation mitigation are discussed. Ultimately, the entirety of this research demonstrates the performance of the novel, data-driven, passive acoustics based SHM technique for wind turbine blades and its viability for a full-scale, operational application.

All interested students and faculty members are invited to attend the online defense via remote access.