Get started
Work

Virtuele meting van de belasting in tandwielkasten voor windturbines

Year: 2024
Type: dissertation
Abstract: Over the last decade, the total installed capacity of wind power generation has increased significantly, reaching 837 GW and 63 GW for onshore and offshore wind, respectively. This trend is expected to continue and even accelerate according to recent forecasts, driven by societal and political motives to avoid harmful climate change, achieve energy independence, and ensure affordable energy supply. With wind power taking up a larger share of the energy mix the pressure is on to decrease the Levelised Cost of Energy (LCoE) of wind power production. In the last decade, the LCoE of wind power has already been reduced by 69 % and 59 % for onshore and offshore turbines, respectively. If wind power is to keep growing as a leading renewable energy source, this trend of reducing LCoE needs to continue. This thesis has aimed to make a contribution to this trend, with a specific focus on the reliability of the gearbox, which makes up a key component of the wind turbine drivetrain. Historic data has shown that gearbox maintenance and failures can have significant impact on the downtime of turbine, with an associated negative effect on the LCoE. In this context, there is an increasing interest to employ condition monitoring techniques in order to keep track of the health and remaining useful life of the gearbox and its components, and to schedule maintenance when necessary. Current gearbox monitoring solutions are based on vibration data, oil particle analysis, or measurements from the turbine's control system. However, these solutions do not directly measure the main driver behind gearbox failures: the loads applied to the gearbox which consist of variable torque and non-torque loads generated by the rotor. Knowledge of the actual operational load a gearbox sees in operation is currently not available, since load sensors that can withstand the high load levels in a wind turbine are too costly and intrusive. This thesis aims to make these loads available, using model-based virtual load sensing as an alternative to direct load sensing. The proposed virtual load sensing methodology is made up of three components: the real drivetrain which is instrumented with non-intrusive indirect sensors, a physics-based model of aforementioned drivetrain which can predict the output of the indirect sensors, and an estimator which will infer the incoming load by comparing the output of the real sensors to predictions made by the model. This thesis has made contributions in the three aforementioned components of virtual load sensing, and has validated the proposed virtual load sensing approach experimentally. For this research, a full-scale wind turbine drivetrain including the main shaft, gearbox and generator was tested in a drivetrain test facility. During the tests controlled loads were applied, enabling a broad coverage of possible load conditions that may occur in drivetrains in the field. The drivetrain was extensively instrumented with encoders, strain gauges on the gearbox housing, and displacement sensors on the torque arm mountings of the gearbox. A validated model for this drivetrain tailored to virtual load sensing was constructed. Analysis of the requirements for virtual load sensing lead to the definition of two-part model consisting of a dynamic torsional drivetrain model and a static reduced-order model for the flexible gearbox housing. This model has been validated extensively and as a results allows for accurate and efficient prediction of the sensor outputs. As estimation algorithm the Augmented Extended Kalman Filter is used, which is well suited for implementation in a monitoring solution because of its efficient iterative formulation. Specific attention has been paid to optimisation of the estimator performance through sensor set optimisation and automatic covariance matrix tuning. The end result is an integrated filter tuning procedure which streamlines the implementation of virtual load sensing and ensures reproducible results. In sum, the result of this thesis is an experimentally validated virtual load sensing methodology for wind turbine gearboxes. The experimental validation results show that the approach is capable of accurately estimating the torque and non-torque loads applied to the gearbox in load cases where one load is varied independently of the others. Promising results were also shown for load cases where multiple loads are estimated simultaneously, but further research is required to bring this aspect to maturity. The influence of the amount of sensors used and the sensor type and location have been investigated, leading to the definition of several feasible sensor sets that have been compared in terms of accuracy.
Language: English
Open Access status: green
OpenAlex ID: W7071991608
Cites:
Cited by:
Related to: