The drivetrain of a wind turbine generator is an expensive but important component of the energy generating system. It’s composed of the main bearing, main shaft, gearbox, brake, generator shaft, and generator. The multi-physical complexity of this arrangement and the expense involved in downtime and maintenance due to failure, ensures that it is both recommended and normal for turbines to have several measures in place to monitor the health of the components.
The most common remote monitoring systems are SCADA and vibration monitoring, collectively known as a condition monitoring system (CMS). SCADA offers an excellent approach for analysis of low frequency signals coming from sensors measuring temperatures, pressures, particle count, current, shaft speeds, and so on.
A good analyst will filter the data and trend the results to look for anomalies across a wind farm, or to detect differences in data across a reference, for example, wind speed. Specifically, within a gearbox sump an oil particle count and temperature events at relatively low data rates can provide invaluable information regarding the wear of components and subsequent friction that would occur as a result of contaminated oil.
A system will contain limit settings, that when breached will trigger an alarm and a service team can be deployed accordingly. However, by the time a temperature limit or particle count is reached it can be argued that this is too late. Figures 0.1 and 0.2 demonstrate analysis that uncovered an elevating particle count trend and abnormal temperature readings across a wind farm.
Vibration monitoring provides a superior method of damage detection, albeit at a higher monetary cost. Utilising spectral techniques, phenomena in data can be uncovered. Damage can be pin-pointed to a component location likely to be undergoing some structural change. Classical spectral methods, such as the fast Fourier transform (FFT), provide valuable information on frequency information, directly related to the kinematics of a gear or bearing.
The rotational nature of the drivetrain lends itself nicely to being able to mathematically determine events per cycle that allow pinpoint accuracy when detecting and locating damage. Further analysis in orders and harmonic analysis for sidebands provides useful information when making decisions on the type of damage that a signal analyst could be looking at.