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.
In this work, a novel approach, termed GNN-tCNN, is presented for the construction and training of Remaining Useful Life (RUL) models. The method exploits Graph Neural Networks (GNNs) and deals with the problem of efficiently learning from time series with non-equidistant observations, which may span multiple temporal scales. The efficacy of the method is demonstrated on a simulated stochastic degradation dataset and on a real-world accelerated life testing dataset for ball-bearings. The proposed method learns a model that describes the evolution of the system implicitly rather than at the raw observation level and is based on message-passing neural networks, which encode the irregularly sampled causal structure. The proposed approach is compared to a recurrent network with a temporal convolutional feature extractor head (LSTM-tCNN), which forms a viable alternative for the problem considered. Finally, by taking advantage of recent advances in the computation of reparametrization gradients for learning probability distributions, a simple, yet efficient, technique is employed for representing prediction uncertainty as a gamma distribution over RUL predictions.