Virtual sensor based on machine learning for estimating gearbox input loads in wind turbines

RWTH Publications (RWTH Aachen)

Abstract

This research presents the development and validation of a virtual sensor system for estimating gearbox input loads in wind turbines using a data-centric machine learning approach. The system provides a cost-effective alternative to direct measurement systems, improving the state-of-the-art of load monitoring technology and, in turn, operational efficiency of wind turbines. The research demonstrates the potential of machine learning algorithms (artificial neural networks and ensemble methods) to accurately estimate gearbox loads under varied operational conditions. Notably, the mean relative error for torque estimation was below 1% of the rated torque. The capability of the developed system was demonstrated in estimating all degrees of freedom loads, with the exception of thrust, for a drivetrain with a 3-point bearing suspension configuration, and in estimating torque for a 4-point drivetrain configuration. Contributions of this work include: • The introduction of innovative virtual load sensing techniques as a viable alternative for direct measurement equipment. • Significant advancements in the application of machine learning to wind turbine load monitoring, addressing variable load conditions. The research also identifies limitations, such as data dependency and challenges in sensor placement and model generalizability across different turbine types. Future research directions are proposed to extend the system’s capabilities, including integration with IoT devices, application across different energy systems, and exploration of advanced machine learning models to improve robustness and accuracy. This work establishes a foundational framework for next-generation load monitoring systems in the wind energy sector, poised to significantly enhance operational efficiencies and sustainability. By integrating innovative virtual sensor technologies with advanced machine learning algorithms, this research offers the potential to drastically reduce maintenance frequency and extend turbine lifespans, thereby promoting cost savings and environmental benefits over time.

Authors 1

1. Baher Azzam corresponding Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen