A machine-learning-based approach for active monitoring of blade pitch misalignment in wind turbines

Abstract

In recent years, timely anomaly detection in wind turbine operations, especially offshore, has become critical. Yet, promptly identifying faults and damage remains a significant challenge, leading to costly maintenance and consumption of resources. Rotor blade pitch misalignment constitutes an essential issue, causing downtime and reduced energy production. Traditional inspection methods are resource-intensive, time-consuming, and also struggle to identify the specific misaligned blades. In addition, their accuracy degrades in the case of small misalignments and strongly depends on the wind regimes, as they are less reliable in turbulence. The absence of an effective automatic solution persists, requiring costly on-site verification. To tackle this challenge, this paper introduces a novel machine-learning-based approach that relies on the combination of random forest classifier instances and linear regression for automatic pitch misalignment detection and localization. This approach not only localizes the affected blades but also detects small misalignments as low as 0.1°. Validation using virtual data coming from a state-of-the-art simulator shows the approach's ability to detect and localize misalignment accurately, even with multiple misaligned blades and in different turbulence conditions, achieving an F1 score exceeding 93 %. Additionally, regression analysis proves the capability of the framework to detect misalignments as low as 0.1° with a root mean square error of 5.48 %. The methodology relies on features extracted from a limited set of sensors already integrated into modern wind turbine systems. Specifically, the extracted indicators are designed to effectively integrate frequency and time domain information on turbine operating conditions, enabling high detection performance even in turbulent wind regimes. The approach is validated across an extended operational envelope using data gathered from a state-of-the-art simulation model commonly used for designing and certifying commercial wind turbine systems.