The study CARE to Compare: A real-world dataset for anomaly detection in wind turbine data presents a new dataset for predictive maintenance of wind turbines, designed to overcome the scarcity of domain-specific public datasets. The dataset includes detailed fault information from 36 wind turbines and proposes CARE, a new scoring method considering anomaly detection accuracy, reliability, and earliness. This dataset and scoring approach aim to identify effective anomaly detection models.

• Provides a comprehensive dataset with 44 labeled anomalies and 51 normal behavior time series.
• Introduces the CARE scoring method for evaluating anomaly detection models.
• Ensures training data quality with turbine-status-based labels.
• Encourages development of robust predictive maintenance strategies for wind turbines.
• Dataset includes 89 years worth of operational data from real-world wind turbines.

This publication is crucial for advancing anomaly detection in wind energy, providing researchers with valuable data for developing and benchmarking new algorithms. The CARE score methodology offers a holistic evaluation of model performances, paving the way for better predictive maintenance solutions Learn more.