Autoencoder-based anomaly root cause analysis for wind turbines

A popular method to detect anomalous behaviour or specific failures in wind turbine sensor data uses a type of neural network called an autoencoder. These models have proven be very successful detecting such deviations, yet cannot show the underlying cause failure directly. Such information is necessary for implementation these planning maintenance actions. In this paper we introduce novel method: ARCANA. We use ARCANA identify possible root causes anomalies detected by It describes process reconstruction as optimisation problem that aims remove properties from anomaly considerably. This must similar and thus only few, but highly explanatory features, sense Ockham’s razor. The proposed applied on open set data, where artificial error was added onto speed measurements acquire controlled test environment. results are compared with errors autoencoder output. points out correctly significantly higher feature importance than other whereas using non-optimised does not. Even though deviation one input large, many features large well, complicating interpretation anomaly. Additionally, apply offshore data. Two case studies discussed, demonstrating technical relevance