Farm-wide virtual load monitoring for offshore wind structures via Bayesian neural networks

TL;DR

This work presents a fleet-leader, Bayesian neural network framework for farm-wide virtual load monitoring of offshore wind structures, enabling load predictions for non-instrumented turbines with quantified epistemic and aleatory uncertainties. By selecting a reduced set of informative inputs (SCADA and a top accelerometer) and decomposing uncertainty, the approach informs data collection, retraining needs, and deployment reliability. Experimental validation in a Belgian offshore wind farm demonstrates that BNNs can detect prediction conflicts (e.g., MP02) and provide meaningful uncertainty metrics that align with measured loads, supporting safer and more cost-effective lifecycle decisions. The framework supports both full probabilistic outputs and epistemic-only variants, offering flexible deployment options across varying instrumentation constraints and operational lifetimes.

Abstract

Offshore wind structures are subject to deterioration mechanisms throughout their operational lifetime. Even if the deterioration evolution of structural elements can be estimated through physics-based deterioration models, the uncertainties involved in the process hurdle the selection of lifecycle management decisions. In this scenario, the collection of relevant information through an efficient monitoring system enables the reduction of uncertainties, ultimately driving more optimal lifecycle decisions. However, a full monitoring instrumentation implemented on all wind turbines in a farm might become unfeasible due to practical and economical constraints. Besides, certain load monitoring systems often become defective after a few years of marine environment exposure. Addressing the aforementioned concerns, a farm-wide virtual load monitoring scheme directed by a fleet-leader wind turbine offers an attractive solution. Fetched with data retrieved from a fully-instrumented wind turbine, a model can be trained and then deployed, thus yielding load predictions of non-fully monitored wind turbines, from which only standard data remains available. In this paper, we propose a virtual load monitoring framework formulated via Bayesian neural networks (BNNs) and we provide relevant implementation details needed for the construction, training, and deployment of BNN data-based virtual monitoring models. As opposed to their deterministic counterparts, BNNs intrinsically announce the uncertainties associated with generated load predictions and allow to detect inaccurate load estimations generated for non-fully monitored wind turbines. The proposed virtual load monitoring is thoroughly tested through an experimental campaign in an operational offshore wind farm and the results demonstrate the effectiveness of BNN models for fleet-leader-based farm-wide virtual monitoring.

Figures (15)

• Figure 1: Schematic diagrams comparing the topology and constituents of a standard deterministic neural network (DNN) and a Bayesian neural network (BNN), both mapping standard input monitoring data $\boldsymbol{x}$ to a load indicator $y$.
• Figure 2: Graphical representation of the reparametrization trick, where by reformulating stochastic network parameters $\boldsymbol{\theta}$ as a function of statistical distribution parameters and additional stochastic inputs, the back-propagation of the loss with respect to variational parameters can be effectively computed.
• Figure 3: Rationale of the proposed farm-wide virtual load monitoring framework featuring Bayesian neural networks as data-based virtual sensors. (Top left) A fleet leader BNN is trained based on available load measurement labels. (Top right) At the deployment stage (measurement labels are no longer available), the pre-trained BNN indicates whether the generated predictions might be inaccurate by reporting a high model uncertainty. (Bottom) Uncertainty decomposition is enabled by the proposed BNN approach, yielding information on: (i) the need to collect more data for improving the model’s performance, (ii) the intrinsic variability of the analyzed phenomena.
• Figure 4: Flowchart diagram illustrating the steps needed for the implementation of the proposed farm-wide virtual load monitoring framework.
• Figure 5: Illustration depicting the monitoring setup installed on an operational offshore wind turbine, from which data was continuously collected during the course of the experimental campaign. The monitoring setup includes a standard SCADA system, accelerometers at three different levels, and strain gauges installed at the lowest level.