Digital Twins for Fault Prognosis in Offshore Wind Turbine Drivetrains

This thesis investigates the relatively novel concept of Digital Twin (DT) in the context of offshore wind turbine drivetrains. DT can be described as a virtual representation of a system or asset that calculates system states and makes system information available, through integrated models and data, with the purpose of providing decision support over its life cycle. Applying these ideas to the use case of drivetrains, there remain many research questions on the specifics of drivetrain DTs such as its model architecture, its capability and value to the user, its limitations and remaining development obstacles, and the errors and uncertainty in its output, which are addressed in this thesis. The scope is narrowed to the fault diagnosis and prognosis, i.e. the remaining useful life (RUL) estimation of mechanical drivetrain components such as bearings and gears. Through this capability, DTs can provide wind farm operators with a form of decision support and thereby contribute to a higher reliability and availability. The investigation is divided into three sections that cover each of the main elements of DT: the data, the models and the decision support. Real-time data streams from physical sensors are crucial to inform the DT on the current state of the drivetrain. The methodology of data acquisition, processing and analysis from different sources such as SCADA and CM systems is presented. DT models differ from conventional, standalone simulation models, in that they are connected to the physical drivetrain and must be updated continuously to reflect its behaviour. Reduced-order models are investigated, which are computationally efficient to allow real-time simulation while maintaining a sufficient model fidelity to accurately capture the complex drivetrain dynamics. In addition, state estimation and system identification methods are presented to synchronize the DT and the physical drivetrain in the kinematic states and model parameters. Decision support are services provided by the DT that assist stakeholders in making key decisions. The focus of this thesis is the RUL estimation based on fatigue assessment, which is facilitated through online load monitoring techniques, referred to as virtual sensing. The methodology for virtual load sensing at the main bearings and at the gearbox components is developed and evaluated in numerical and experimental case studies. This thesis contributes to a better understanding of the DT concept by developing and showcasin