2022
DOI: 10.1177/0309524x221080470
|View full text |Cite
|
Sign up to set email alerts
|

An unsupervised data-driven approach for wind turbine blade damage detection under passive acoustics-based excitation

Abstract: Existing passive acoustics-based techniques for wind turbine blade damage detection lack the robustness and adaptability necessary for an operational implementation due to their physics- and model-based dependency. In contrast, this study develops an entirely unsupervised, data-driven damage detection technique. The novelty of the technique lies in (i) the development and comparison of spectral and cepstral-domain features for the robust characterization of the cavity-internal acoustics, (ii) the use of autoen… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
3
0

Year Published

2023
2023
2024
2024

Publication Types

Select...
6
1

Relationship

1
6

Authors

Journals

citations
Cited by 7 publications
(5 citation statements)
references
References 56 publications
(63 reference statements)
0
3
0
Order By: Relevance
“…The aim of feature extraction is to summarize the salient signal characteristics in each frame to optimize the input for machine learning. In this study, linear frequency cepstral coefficients (LFCC) were extracted largely due to their previous success in characterizing WTB damage (Solimine and Inalpolat, 2022). LFCC’s are derived from cepstral coefficients, which are calculated from the inverse Fourier transform of the logarithm of a spectrum (equation (1) (Randall, 2011)),…”
Section: Methodsmentioning
confidence: 99%
“…The aim of feature extraction is to summarize the salient signal characteristics in each frame to optimize the input for machine learning. In this study, linear frequency cepstral coefficients (LFCC) were extracted largely due to their previous success in characterizing WTB damage (Solimine and Inalpolat, 2022). LFCC’s are derived from cepstral coefficients, which are calculated from the inverse Fourier transform of the logarithm of a spectrum (equation (1) (Randall, 2011)),…”
Section: Methodsmentioning
confidence: 99%
“…However, within this broader scope, a variety of more specific applications can also be delineated. Namely, this includes applications such as detection, to find issues when they occur [64,67,73,74,81], diagnosis, to classify issues [64,[68][69][70][71]74,77,82], and prediction, to determine when an issue may occur [65,66,72,[77][78][79][80]. The purpose of these tasks, as stated by J.-Y.…”
Section: Condition Monitoringmentioning
confidence: 99%
“…Man, Z. Zhang, and Q. Zhou present a framework to predict wind turbine shutdowns through data-driven predictive analytics [72]. Jaclyn Solimine and Murat Inalpolat present a data-driven, acoustics-based method to detect damage to the turbine blades, reaching between 89% and 99.8% detection accuracy depending on the type of fault [73]. D. Yu et al propose a "deep-belief network" data-driven method for fault detection and classification which is not only compared to traditional models, but also to a selection of multiple existing data-driven models as well to emphasize its accuracy over previous methods [74], a comparison illustrating just how well adapted data-driven modeling is to fault detection as an application, and how populous it has become in the space.…”
Section: Condition Monitoringmentioning
confidence: 99%
“…The experimental results show that the 1, 2, 4, 6, and 8 dimensions of the 12-dimensional MFCC features were more obvious in distinguishing the faulty samples from the normal ones. Solimine and Inalpolat [189] set up simulated damage in the form of penetrating holes and cracks at three different locations (the leading edge of the front cavity, the side of the front cavity, and the side of the back cavity) and compared the MFCC, GTCC, LFCC40, and LFCC80 feature sets, finding that the LFCC40 and LFCC80 feature sets provided the highest detection accuracy in both hole-type and crack-damage-type damage and significantly outperformed the common MFCC feature set. If a damage location is given, a hole-type damage of 0.16 cm can be detected 100% of the time, and the accuracy of detecting cracks with a minimum length of 1.27 cm can reach 97%.…”
Section: Aerodynamic Noise Feature Extractionmentioning
confidence: 99%