Conditional variational autoencoders for probabilistic wind turbine blade fatigue estimation using Supervisory, Control, and Data Acquisition data

Abstract

Wind turbine fatigue estimation is based on time‐consuming Monte Carlo simulations for various wind conditions, followed by cycle‐counting procedures and the application of engineering damage models. The outputs of the fatigue simulations are large in volume and of high dimensionality, as they typically consist of estimates on finite‐element computational meshes. The strain and stress tensor time series, which are the primary quantities of interest when considering the problem of fatigue estimation, are dictated by complex vibration characteristics due to the coupled effect of aerodynamics, structural dynamics, geometrically non‐linear mechanics, and control. A Variational Auto‐Encoder (VAE) is trained in order to model the probability distribution of the accumulated fatigue on the root cross‐section of a simulated wind turbine blade. The VAE is conditioned on historical data that correspond to coarse wind‐field measurement statistics, such as mean hub‐height wind speed, standard deviation of hub‐height wind speed and shear exponent. In the absence of direct measurements of structural loads, the proposed technique finds applications in making long‐term probabilistic deterioration predictions from historical Supervisory, Control, and Data Acquisition (SCADA) data, while capturing the inherent aleatoric uncertainty due to the incomplete information on strain time series of the wind turbine structure, when only SCADA data statistics are available.

Improving Seasonal Forecast Using Probabilistic Deep Learning

“…For the latter, we actually use the mean of p ψ , rather than a sample of this learned Gaussian distribution as our sampling result. This widely‐applied practice yields an average estimate of all possible outcomes conditioned on a sampled latent z (Mukkavilli et al., 2020; Mylonas et al., 2021; Walker et al., 2016). These expected Y outcomes are therefore considered as probabilistic forecasts.…”

“…Recently, approximate probabilistic inference powered by deep neural networks has enabled probabilistic modeling on big data (Kingma & Welling, 2013). Existing works that leverage this idea have demonstrated imminent potential for various probabilistic modeling and forecasting tasks, such as video prediction (Walker et al., 2016), wind turbine fatigue prediction (Mylonas et al., 2021), and stochastic parameterization of unresolved processes in climate models (Mukkavilli et al., 2020). We make use of this novel technique to learn from climate simulation for probabilistic seasonal forecasting.…”

Fatigue modeling using neural networks: A comprehensive review

“…Tao et al 202 proposed a model integrating a convolutional autoencoder, an RNN, and a neural ordinary differential equation to process the guided wave information. Mylonas et al 203 used a variational autoencoder to model fatigue damage probability distribution for a wind turbine blade.…”

Data-driven farm-wide fatigue estimation on jacket-foundation OWTs for multiple SHM setups

“…Also applied to estimate the blade root flapwise damage equivalent loads (DELs), Schröder's (2020) work has emphatically demonstrated how a surrogate model based on ANNs outperforms other surrogate models, such as polynomial chaos expansion and quadratic response surface, in computational time, model accuracy and robustness, further applying it to connect wind farm loads to turbine failures (Schröder, 2020). As for Mylonas et al (2020), it used conditional variational auto-encoder neural networks to estimate the probability distribution of the accumulated fatigue on the root cross-section of a simulated wind turbine blade, making long-term probabilistic deterioration predictions based on historic SCADA data (Mylonas et al, 2020(Mylonas et al, , 2021.…”