Acoustic Emission Monitoring of Wind Turbine Bearings

Climate change, national and international targets and a possible impending fuel scarcity is driving the need for a clean, cheap and sustainable energy source. Although onshore wind is currently the most economically viable source of renewable energy, high failure rates often occurring many years prior to their design life, are
increasing the cost through additional maintenance and downtime. The gearbox, and in particular the bearings within the gearbox are the components responsible for the
largest proportion of downtime. Field studies have shown damage to the inner race of the planetary support bearings found in the epicyclic stage of the gearbox is restricted to an arc of approximately 40° centered on the point of maximum load. Engineers at Ricardo have designed an actuation system to overcome this problem which allows the raceway to be rotated periodically thus distributing damage and increasing bearing life.
The monitoring of planetary support bearings typically found in the epicyclic stage of wind turbine gearboxes has been investigated in this thesis using acoustic emission technology due to its reported increase in sensitivity in detecting damage at low speeds compared to vibration analysis in addition to its ability to locate damage. Primarily, accelerated life tests were performed on a rolling element bearing seeded with a defect
mounted in a bespoke full scale test rig designed to mimic loading conditions experienced by the planetary support bearings. In addition, data was recorded and analysed from a split bearing test rig and the high speed shaft bearing of a recommissioned 600kW wind turbine gearbox.
Initial experiments considered the influence of the lubrication regime on the measured acoustic emission signal. It was found that as the oil film reduces, asperity contact, typical of mixed or boundary lubrication, manifests itself as high amplitude
transient events. Typical measures of bearing health in the time domain, such as peak values or kurtosis, become unreliable and demonstrates a need for a novel approach.
Previous investigations into the use of acoustic emission for the purpose of bearing condition monitoring has focused on instances where full separation between bearing components occurred whereas this work considers a mixed lubrication regime. To overcome the drawbacks of the traditional measures, this work has investigated a process employing wavelet packet decomposition, autocorrelation and cepstrum to reduce the noise and boost the periodicity of a signal from a defected bearing. Outlier analysis was shown to be able to determine the presence of a seeded defect and indicate which bearing component is defected. Such approach was shown to provide a more robust measure than time domain methods. In contrast, this approach was compared to one
employing time domain measures for a fully lubricated split bearing. In this case, a time domain approach was more successful at determining the presence of damage
than the approach taken for the partially lubricated bearing.
An attempt was made to improve the localisation of defects on a bearing which had, until now, relied on analytical time-of-flight methods. In this work artificial sources, rather than those resulting from a rolling element impinging on a defect, were generated
by a standardised pencil lead break and used as training data for two methods namely Delta T mapping and neural networks. The neural networks in particular were shown
to reduced the average error from 42mm to 17mm however given the time consuming nature of generating the training data a decision must be made regarding the relative importance of accuracy and ease of implementation.