Dirt and Mud Detection and Diagnosis on a Wind Turbine Blade employing Guided Waves and Supervised Learning Classifiers
Dirt and mud on wind turbine blades (WTB) reduce productivity and can generate stops and downtimes. A condition monitoring system based on non-destructive tests by ultrasonic waves was used to analyse it. This paper employs an approach that considers advanced signal processing and machine learning t...
| Autores: | , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2018 |
| País: | España |
| Institución: | Universidad de Castilla-La Mancha |
| Repositorio: | RUIdeRA. Repositorio Institucional de la UCLM |
| OAI Identifier: | oai:ruidera.uclm.es:10578/18937 |
| Acceso en línea: | http://hdl.handle.net/10578/18937 |
| Access Level: | acceso abierto |
| Palabra clave: | Macro Fiber Composite Wavelet Transforms Machine Learning Guided Waves Wind Turbine Blade |
| Sumario: | Dirt and mud on wind turbine blades (WTB) reduce productivity and can generate stops and downtimes. A condition monitoring system based on non-destructive tests by ultrasonic waves was used to analyse it. This paper employs an approach that considers advanced signal processing and machine learning to determine the thickness of the dirt and mud in a WTB. Firstly, the signal is filtered by Wavelet transform. FE and Feature selection (FS) are employed to remove non-useful data and redundant features. FS selects the number of the most significant terms of the model for fault detection and identification, reducing the dimension of the dataset. Pattern recognition is carried out by the following supervised learning classifiers based on statistical models to calculate and classify the signal depending on the fault: Ensemble Subspace Discriminant; k-Nearest Neighbours; Linear Support Vector Machine; Linear Discriminant Analysis; Decision Trees. Receiver Operating Characteristic analysis is used to evaluate the classifiers. Neighbourhood Component Analysis has been employed in feature selection. Several case studies of mud on the WTB surface have been considered to test and validate the approach. Autoregressive (AR) model and Principal Component Analysis (PCA) have been employed to FE. The results provided by PCA show an improvement on the AR results. The novelty of this work is focused on applying this approach to detect and diagnose mud and dirt in WTB. |
|---|