Vibration-Based Condition Monitoring of Drivetrains in Large Offshore Wind Turbines in a Digital Twin Perspective

Abstract

According to EU 2050 plan, offshore wind farms based on large floating wind turbines are considered as a main source of power supply in the coming future. Among the turbine systems, drivetrain contributes significantly in levelized cost of energy (LCOE). In other words, the careful selection of the drivetrain configuration and the employment of properly designed condition monitoring systems can help to considerably reduce the cost of energy in floating offshore wind turbines. As the first study in this thesis, an analytical system-level drivetrain design approach supported by numerical simulations is employed to identify the most economical drivetrain configuration of large floating offshore wind turbines in a life cycle perspective. The medium-speed PMSG drivetrain technology is selected as a compro mise between design, manufacturing and installation, and operation and maintenance costs, and in the next steps of this research, the drivetrain vibration-based condition monitoring tools are designed or developed to re alize the improved availability of future offshore wind turbines by following the sequence described underneath: In the first step, the classical vibration-based condition monitoring based on the time and frequency domain analytical tools based on the translational vibration measurements captured by accelerometers which are placed on the different parts of drivetrain are reviewed as the wind turbine standard con dition monitoring solution. As the research contribution in time domain analysis of translational vibration measurements for condition monitoring of the drivetrain components, a data-driven statistical learning-based con dition monitoring approach grounded on monitoring the unusual variations of the parameters of the multi-variate distribution which fits the combined measurements of the drivetrain accelerometers is studied. The potentials of this approach in early-stage fault detection compared to the classical time domain fault detection approach based on monitoring the exceedance of the root mean square (RMS) of the axial and lateral acceleration and velocity is demonstrated. To take the vibration-based condition monitoring of the driv etrain further, the topic of innovative condition monitoring of the drivetrain based on the torsional vibration measurements captured by the drivetrain encoders is introduced and discussed, and the possibility of observing the different classes of drivetrain faults by using the torsional instead of trans lational vibration measurements is studied. The possibility of using the tor sional response to provide better insights into the drivetrain internal and external excitation sources is discussed, which can be used to improve the available condition monitoring systems for realizing an earlier stage fault detection. As the second step, the drivetrain condition monitoring by using the tor sional measurements is investigated in more detail. In this study, the possi bility of performing drivetrain modal analysis by using the torsional response is discussed. Then an analytical approach is proposed to diagnose the driv etrain faults at system-level by monitoring the variations of the drivetrain dynamic properties (i.e. natural frequencies, mode shapes and damping coef ficients) which can be estimated from the torsional measurements. In the third step, the drivetrain online fault prognosis by monitoring the residual life of the components is emphasized. For this purpose, the multi degree of freedom (DOF) linear torsional models of the drivetrain are pro posed to be used as the digital twin of drivetrain, and their capabilities in prediction of the remaining useful lifetime (RUL) of the different drivetrain components is discussed. Digital twin in this thesis context is defined as the combination of equivalent model, online measurements and RUL model. The algorithm for the near real-time estimation of the parameters of drive train equivalent reduced order model (ROM) in the cases of different degrees of model complexity by using the drivetrain torsional measurements is pre sented, and the application of proposed digital twin model for estimating the degradation of the drivetrain gears and shafts is demonstrated. Load ob servers are designed for the different components of the drivetrain, which receive the parameters of the updated drivetrain equivalent model and the online torsional response to estimate loads in the different components of the drivetrain. The employed stochastic physics-based degradation model works based on the real-time cycle counting of the equivalent stress, and is able to provide confidence interval for the estimated damage. The integration of the model with the real-time operational data in a dig ital twin platform, which provides the drivetrain updated ROM parameters and dynamic properties, can also support fault diagnosis algorithms which are discussed in this PhD thesis: In a direct way, by having access to the real-time values of system parameters, it is possible to define different fault states of the different classes of progressive faults in the drivetrain compo nents in terms of the variation of the ROM parameters. Seeing that the ROM parameters are directly connected to the physics of the system and the com ponents, defining thresholds for the different states of progressive faults is straightforward. In an indirect approach, the proposed algorithm provides the real-time values of drivetrain dynamic properties, which can support the proposed fault diagnosis approach based on monitoring the variations of dynamic properties to estimate the state of the faults.