Deep Neural Network Hard Parameter Multi-Task Learning for Condition Monitoring of an Offshore Wind Turbine

Abstract

Breaking the curse of small datasets in machine learning is but one of the major challenges that cause several real-life prediction problems. In offshore wind application, for instance, this issue presents when monitoring an asset in an attempt to reduce its infant mortality failures. Another challenge could emerge when reducing the number of sensors installed in order to limit the investment in monitoring systems. To tackle these issues, the aim of this article is to investigate the impact of small data-set on conventional machine learning methods, and to outline the improvement achievable by the implementation of transfer learning approach. It provides a solution to mitigate this issue by applying a hard parameter multi-task learning approach to the supervisory control and data acquisition data from an operational wind turbine, allowing for smaller datasets to efficiently predict the status of the gearbox’s vibration data. Two experiments are carried out in this paper. The first is to envisage the possibility of using hard parameter transfer on the operational data from two wind turbines. The second is to compare the results of this model to the findings from a conventional deep neural network model trained on the data from a single turbine.

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

“…The missing values can also be handled by filling them with the values of similar non-missing observations. This can be done by using clustering algorithms, for example, k-nearest neighbors (see (Black et al, 2022)). An alternative technique is simply removing the observations with missing data (see (Maron et al, 2022), (Miele et al, 2022), (Cui et al, 2018) and (Bangalore et al, 2017)).…”

“…-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.…”

“…The signals that are often used for condition monitoring of wind turbines can be more or less divided into three groups: 1) environmental data like wind speed, outside temperature, ... (used in for example (Bermúdez et al, 2022), (Bermúdez et al, 2022), (Black et al, 2022), (Campoverde et al, 2022), (Mazidi et al, 2017), (Miele et al, 2022), ... ), 2) operational data from the wind turbine like active power, rotor speed, ... (used in for example (Bermúdez et al, 2022), (Black et al, 2022), (Chesterman et al, 2022), (Miele et al, 2022), (Peter et al, 2022), ...) and 3) wind turbine temperature signals like the temperatures of the generator bearings, the temperature of the main shaft bearing, the temperature of the generator stator, ... (used for example (Bermúdez et al, 2022), (Black et al, 2022), (Campoverde et al, 2022), (Chesterman et al, 2022), (Mazidi et al, 2017), ...). The first two groups are often used by default, independent of the target signal that needs to be modeled or the failure that needs to be detected.…”

“…Both types can be found in the state-of-the-art literature. Deep neural networks are used in (Black et al, 2022), (Jamil et al, 2022) (transfer learning), (Mazidi et al, 2017), (Verma et al, 2022), (Meyer, 2021) (multi-target neural network), (Turnbull et al, 2021), (Cui et al, 2018), (Sun et al, 2016), (Bangalore and Tjernberg, 2015) (NARX), (Bangalore and Tjernberg, 2014), (Li et al, 2014), (Kusiak and Verma, 2012), (Kusiak and Li, 2011), (Zaher et al, 2009). Another popular type of model is the Autoencoder (AE).…”