Font Size: 

Reducing the cost of energy from wind is an important challenge the industry faces today. Operation and maintenance costs
constitute a large proportion of the total cost of energy from wind in large wind farms. Increasing the wind turbine availability
and therefore reducing these costs, can be achieved through the successful detection of incipient faults before they become
catastrophic failures. Consequently, condition monitoring systems are increasingly being developed and integrated in wind farms.
Modern wind turbines are equipped with vibration monitoring systems for the online active remote monitoring and control of
their components. Earlier detection, diagnosis and prognosis of faults in an accurate automated process is still a challenge. One
of the most expensive components to replace is the gearbox. The vibration signal from gears is affected greatly by load, so
distinguishing such variations from changes in condition is important. In this respect, the aim of this paper is to apply an infinite
Gaussian mixture model as a health indicator of wind turbine gearbox vibration signals. When a gear has a local defect, such as a
fatigue crack, the stiffness of the neighbouring teeth is affected and this produces changes in the vibration signal. These changes
are defined by amplitude and frequency modulation. In the frequency domain, the spectrum will comprise the fundamental and
harmonics of the meshing frequency surrounded by modulation sidebands. A Gaussian mixture model is a probabilistic model that
assumes all the data points are generated from a mixture of a finite number of Gaussian distributions with unknown parameters.
Assuming that the narrowband spectrum of the gear mesh frequency and the neighbouring sidebands is a mixture of Gaussian
distributions, then the proposed algorithm fits a number of classes of Gaussian distributions. The severity of fault is indicated
by the presence of sidebands and the results of this method shows the number of classes increases significantly in the presence
of sidebands. The infinite Gaussian mixture model method is compared with other well established methods used in
condition monitoring. All proposed methods are applied on real wind turbine gearbox vibration data. The failure mode examined
in the case study is a gear tooth crack in the intermediate speed shaft stage of a planetary gearbox. Data is collected for a healthy
gearbox and for various time instants prior to the gear failure. The advantage of the infinite Gaussian mixture model compared
to other methods is that it can detect potential presence of a fault regardless of the loading conditions, while its disadvantage is
that it needs more computational time. The above work will provide a robust framework for the early detection of wind turbine
gearbox faults.