Normal behaviour models for wind turbine vibrations: An alternative approach

TL;DR

This paper compares a stochastic model and a neural network for reconstructing wind turbine vibration signals, finding the stochastic approach better preserves signal statistics and frequency components, aiding abnormal behaviour detection.

Contribution

It introduces a stochastic reconstruction method for wind turbine vibration signals and demonstrates its superiority over neural networks in preserving signal characteristics.

Findings

Stochastic approach outperforms neural networks in high-frequency signal reconstruction.

Stochastic model better preserves signal statistics and frequency components.

Open source implementation available for broader application.

Abstract

The identification of abnormal behaviour in mechanical systems is key to anticipate and avoid their potential failure. Thus wind turbine health is commonly assessed monitoring series of $10$-minute SCADA and high frequency data from sensors. To monitor wind turbine vibrations, normal behaviour models are built to predict tower top accelerations and drive-train vibrations. Signal deviations from model prediction are labelled as anomalies and are further investigated. More efficient models are expected to help enhancing the identification of abnormal behaviour. In this paper we assess a stochastic approach to reconstruct the $1$ Hz tower top acceleration signal, which was measured in a wind turbine located at the wind farm Alpha Ventus in the German North Sea. We compare the resulting data reconstruction with that of a model based on a neural network, which has been previously reported as a data-mining algorithm suitable for reconstructing this signal. In order to focus the discussion on the similarities and differences of both approaches, we limit our evaluation to a single input-output system; in doing so, we avoid other differences, i.e. loading type, pre-processing or model complexity. Our results present evidence that the stochastic approach outperforms the neural network in the high frequency domain ($1$ Hz). Although neural network retrieve accurate step-forward predictions, with low mean square errors, the stochastic approach predictions better preserve the statistics and the frequency components of the original signal, remaining high accuracy levels. The implementation of our stochastic approach is available as open source code and can easily be adapted for other situations involving stochastic data reconstruction. Based on our findings we argue that such an approach could be implemented in signal reconstruction for monitoring purposes or for abnormal behaviour detection.