Anomaly Detection Using a SCADA Feature Extractor and Machine Learning to Detect Lightning Damage on Wind Turbine Blades

Abstract

Many wind turbines in Japan are damaged by lightning strikes. In particular, blades that are already damaged by lightning strikes are further damaged by continuous blade rotation. To prevent such additional damage, an emergency stop device that is triggered by lightning detection is required to be installed on wind turbines in winter lightning areas. Normally, the wind turbine restarts after soundness is confirmed by inspection. However, it is often difficult to inspect the turbine visually because of bad weather, and consequently, the wind turbine downtime is prolonged. This type of downtime is a reason for the reduced availability of wind turbines. Therefore, in this study, we consider technologies that would allow quick restart of wind turbines and improve availability, based on understanding the soundness of blades after a lightning strike, using a machine-learning model based on supervisory control and data acquisition system data.

An Enhanced Interval‐Based Approach for Anomaly Detection in Time Series

“…Time series data can be encountered in various fields such as social economics, engineering technology, and so on, and are receiving more and more attention [1][2][3]. Time series anomaly detection plays an important role in avoiding major accidents and reducing economic losses, such as to detect damage already caused by lightning strikes in wind turbines to prevent further damage and to detect bearing anomalies in the manufacturing industry, and so on [4,5]. Time series anomaly detection is required to solve two main problems: (i) defining what kinds of data are anomalies in a given set of data; (ii) finding an effective way to detect these anomalies [6].…”

“…A plethora of anomaly detection techniques have been discussed in the literature, most of which are based on pure numeric data [3][4][5][8][9][10][11]. An approach for anomaly detection on the basis of information granules is reported in this study, which has not been investigated or studied in depth.…”