Premature failures in large offshore Wind Turbines are often attributed to bearing failure despite gearboxesbeing designed and developed using the best bearing design practices. Furthermore, as turbine size andrated power increase, bearings display an enhanced tendency to fail. Unscheduled bearing replacement atsea is a complex, costly, weather-dependent and time-consuming operation that results in high turbinedowntimes. Market trends show an increase in turbine rated capacity and a noticeable shift towards deeperwaters and far-off remote sites which further delays and complicates unscheduled maintenance activitiesand aggravates the cost penalties of idle turbines. Detecting an incipient bearing fault (diagnosis task) istherefore a major aspect to evaluate drivetrain and overall wind turbine reliability. Moreover, estimating theremaining useful life of bearings and predicting their operational state in the future (prognosis task) canachieve a breakthrough in optimizing maintenance programs, improve wind farm operation and decreasewind turbine downtime which can bring about a significant cost reduction.
The purpose of this work is to investigate the health monitoring and prognostics possibilities of drivetrainbearings in a floating spar-buoy offshore wind turbine. The drivetrain concept considered in this work isbased on DTU's 10-MW reference wind turbine. Specifically, this study targets the prognosis of four criticaldrivetrain bearings located in the main shaft and the high-speed shaft. The absence of run-to-failure dataof real wind farms, although inconvenient, is overcome by using model-generated degradation data. A high-fidelity numerical twin of a state-of-the-art drivetrain concept is used in this work and is established using amulti-body system (MBS) approach. The numerical twin models a medium-speed 10-MW gearbox thatconsists of 3 stages, 2 planetary stages and 1 parallel stage, supported in a 4-point configuration layoutwith two main bearings and two torque arms. The drivetrain concept studied in this work uses a novelselection of bearings which is currently gaining traction in large offshore wind turbines. The two mainbearings that support the main shaft are tapered roller bearings (TRB) that carry both axial and radial loadsas opposed to the main bearings used in traditional high-speed gearbox designs which typically use acylindrical roller bearing to carry radial loads and a spherical roller bearing to carry axial loads.
Faults are applied on the main bearing and on the high-speed shaft bearings of the numerical model. Themodel-generated degradation data, namely forces and acceleration measurements at several shafts andbearings, is used as input data for two independent prognosis models: a physics-based prognosis modeland a data-driven prognosis model. The physics-based approach will culminate in a prediction of theremaining useful life (RUL) of several bearings under a range of faults. The fault detection and faultprognosis capabilities of the proposed prognosis methods is evaluated and compared. This work will alsoassess the merits and limitations of using model-generated degradation data for the development ofprognosis models. Lastly, based on this study, the requirements to enable bearing prognosis from a purelydata-driven approach, as opposed to a physics-based approach, is put forth.