Condition monitoring of wind turbines using machine learning based anomaly detection and statistical techniques for the extraction of 'healthy data'

Abstract

Premature failures caused by excessive wear are responsible for a large fraction of the maintenance costs of wind turbines. Therefore, it is crucial to be able to identify the propagation of these failures as early as possible. To this end, a novel condition monitoring method is proposed that uses statistical data analysis techniques and machine learning to construct a multivariate anomaly detection framework, based on high-frequency temperature SCADA data from wind turbines. This framework contains several steps. First, there is a preprocessing step in which relevant wind turbine states are extracted from the data. These states are the operating conditions and whether or not the turbine exhibits transient behavior. The second step entails anomaly detection on the temperature time series data. Fleet information is used to filter out exogenous (environmental) factors. Furthermore, multiple models are combined to get more stable and robust anomaly detections. By combining them the weaknesses of the individual models are alleviated resulting in a better overall performance. A limitation of machine learning-based anomaly detection on temperature data is the requirement that at least one year of “healthy” (meaning without anomalies) training data is available to account for seasonal effects. The lack of verified “healthy” data spread out evenly over the seasons generally means that the anomaly detection accuracy is severely compromised for the unrepresented seasons. This research uses smart retraining to reduce this limitation. Statistical techniques that leverage the information of the fleet are used to extract “healthy” data from at least one year, but preferably multiple years, of unverified data. This can then be used as training data for the machine learning-based models. To validate the pipeline, temperature and failure data of a real operational wind farm is used. Although the methodology is general in its scope, the validation case focusses specifically on generator bearing failures.

Overview of normal behavior modeling approaches for SCADA-based wind turbine condition monitoring demonstrated ondata from operational wind farms

“…This can be done in several ways, e.g. Hermite interpolation (see (Bermúdez et al, 2022) and (Campoverde et al, 2022)), linear interpolation (see (Chesterman et al, 2022), (Chesterman et al, 2021) and (Miele et al, 2022)), .... Another technique is called carry forward and/or backward. In this technique, the missing value is replaced by the last known value before the missing value (carry forward) or the first known value after the missing value (carry backward).…”

“…In this technique, the missing value is replaced by the last known value before the missing value (carry forward) or the first known value after the missing value (carry backward). This is used in (Bermúdez et al, 2022), (Campoverde et al, 2022), (Chesterman et al, 2022), (Chesterman et al, 2021) and (Mazidi et al, 2017). The missing values can also be handled by filling them with the values of similar non-missing observations.…”

“…CC BY 4.0 License. can be done for example by using the interquartile range (see (Campoverde et al, 2022)), or a custom dynamic or user-defined threshold (see (Chesterman et al, 2022), (Chesterman et al, 2021) and (Catellani et al, 2021)), the 5-sigma rule (see (Miele et al, 2022)) or clustering (see (Cui et al, 2018) and (Bangalore et al, 2017)). This operation needs to be done carefully to avoid that abnormal values associated with the failure of interest are also removed.…”

“…If it is possible to clean the signal this should be considered, since it will improve the performance of the NBM model. This can be done for example by aggregating the data to a lower resolution (see (Chesterman et al, 2022), (Chesterman et al, 2021) and (Turnbull et al, 2021)), or cleaning or filtering the data using expert knowledge (see (Peter et al, 2022), (Takanashi et al, 2022), (Verma et al, 2022), (Turnbull et al, 2021), (Udo and Yar, 2021), (Beretta M. and J., 2020) and (Kusiak and Li, 2011)).…”

“…-10-minutes: (Bermúdez et al, 2022), (Black et al, 2022), (Campoverde et al, 2022), (Chesterman et al, 2022), (Maron et al, 2022), (Mazidi et al, 2017), (Miele et al, 2022), (Peter et al, 2022), (Takanashi et al, 2022), (Beretta et al, 2021), (Beretta M. and J., 2020), (Catellani et al, 2021), (Chen et al, 2021), (Chesterman et al, 2021), (Meyer, 2021), (Turnbull et al, 2021), (Udo and Yar, 2021), (Beretta M. and J., 2020), (Liu et al, 2020), (McKinnon et al, 2020, (Renström et al, 2020), (Zhao et al, 2018), (Bangalore et al, 2017), (Dienst and Beseler, 2016), (Bangalore and Tjernberg, 2015), https://doi.org/10.5194/wes-2022-120 Preprint. Discussion started: 21 February 2023 c Author(s) 2023.…”