Transfer learning applications for autoencoder-based anomaly detection in wind turbines

Anomaly detection in wind turbines typically involves using normal behaviour models to detect faults early.Normal behaviour models are often implemented through the use of neural networks,of which autoencoders are particularly popular in this field.However,training autoencoder models for each turbine is time-consuming and resource intensive.Thus,transfer learning becomes essential for wind turbines with limited data or applications with limited computational resources.This study examines how cross-turbine transfer learning can be applied to autoencoder-based anomaly detection.Here,autoencoders are combined with constant thresholds for the reconstruction error to determine if input data contains an anomaly.The models are initially trained on one year’s worth of data from one or more source wind turbines.They are then fine-tuned using small amounts of data from the target wind turbine.Three methods for fine-tuning are investigated:adjusting the entire autoencoder,only the decoder,or only the threshold of the model.The performance of the transfer learning models is compared to baseline models that were trained on one year’s worth of data from the target wind turbine.The results of the tests conducted in this study indicate that models trained on data of multiple wind turbines do not improve the anomaly detection capability compared to models trained on data of one source wind turbine.In addition,modifying the model’s threshold can lead to comparable or even superior performance compared to the baseline,whereas fine-tuning the decoder or autoencoder further enhances the models’performance.

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