Replication Data for: Detection of wind turbine rotor imbalance using unsupervised output-only vibration data analysis

This dataset supports research on rotor imbalance detection in wind turbines using output-only vibration data. It includes 65 experiments conducted on a laboratory-modified Enair E30Pro wind turbine (3 kW), adapted for indoor use. The turbine was modified to maintain the original moment of inertia a...

Descripción completa

Detalles Bibliográficos
Autores: Vidal, Yolanda, Tutivén Gálvez, Christian Javier, González, Iván Ariel, Gómez Campuzano, Abel Francisco
Tipo de recurso: conjunto de datos
Fecha de publicación:2023
País:España
Institución:Consorci de Serveis Universitaris de Catalunya (CSUC)
Repositorio:CORA.Repositori de Dades de Recerca
OAI Identifier:oai:dnet:cora.rdr____::f569a4afabb398639b3154d43e33012a
Acceso en línea:https://doi.org/10.34810/DATA2304
Access Level:acceso abierto
Palabra clave:Engineering
Wind turbines
Aerogeneradors
Condition monitoring
Rotor imbalance
Vibration
Unsupervised learning
Extended isolation forest
Descripción
Sumario:This dataset supports research on rotor imbalance detection in wind turbines using output-only vibration data. It includes 65 experiments conducted on a laboratory-modified Enair E30Pro wind turbine (3 kW), adapted for indoor use. The turbine was modified to maintain the original moment of inertia after resizing the blades and to allow for electrical operation and systematic data collection. Vibration data were collected using 8 triaxial accelerometers (PCB Piezotronics 356A17) placed on the turbine structure. This setup produced 24 measurement channels. Each experiment was recorded over 60 seconds at a sampling rate of 1706.66 Hz, resulting in a matrix with 102400 time points and 24 channels. Data are stored in individual .mat files, each containing one matrix per experiment. Of the 65 experiments, 45 correspond to healthy operating conditions and are divided into training, validation, and test subsets. The remaining 20 simulate rotor imbalance conditions across four severity levels by adding calibrated masses to the blades. The imbalance scenarios are as follows: Level 1: one blade with a 250 g mass, representing 12.5 percent of blade weight Level 2: one blade with a 500 g mass, representing 25 percent of blade weight Level 3: one blade with a 750 g mass, representing 37.5 percent of blade weight Level 4: two blades each with a 250 g mass (12.5 percent), one blade without added mass These configurations replicate realistic imbalance causes such as ice buildup or material wear. The dataset is well suited for time-series analysis, feature extraction, and the development of fault detection algorithms.