Using Cepstrum and Historical Data to Detect Planetary Stage Faults

Introduction

Condition monitoring system (CMS) plays an important role in wind turbines operation and maintenance. It helps detect incipient faults early and plan maintenance so that downtime is minimized. Currently Brüel & Kjær Vibro is monitoring around 8000 turbines all over the world from different wind turbine manufacturers. Not only are scalar data for trending collected regularly, but also time waveform data from different components of the wind turbines. Collecting time waveform data regularly proves to be very useful because new monitoring methods can be developed and tested based on the collected time waveforms. The current work focuses on using cepstrum to detect gear related faults in the planetary stage of wind turbines. The method is mainly data-driven due to large number of data available.

Approach

Accelerometers are installed at different components of wind turbines, such as at various gearbox stages and generator bearings, to monitor their vibrations. Various vibration measurements are collected at a short interval period and trended over time. Additionally, vibration time waveform is collected from each component at a regular interval, normally every other day. The time waveform length is normally around 10 seconds. Due to the traditionally challenging nature of detecting gear related faults at the planetary stage, this work focuses on using cepstrum to detect faults at this stage. Cepstrum data are generated from time waveforms collected from many turbines over a period of one year. Both turbines with known faults and no known faults are compared.

Main body of abstract

By compiling all data available, quefrencies related to faults usually stand out, and their ranges can be identified. The ranges are usually limited for turbines with similar gearbox ratios. Information regarding these quefrency ranges can then be used for trending the quefrency magnitudes in those ranges for faults detection in many turbines. In this way, it is not necessary to know the kinematical data of each turbine, such as specific gear mesh frequency, a particular shaft rotational speed, etc. The method was tested on turbines with known faults. Faults can be detected earlier than using other traditional vibration measurements, such as mesh frequencies, residual values, etc.

Conclusion

Collecting time waveform data is very useful to develop data-driven monitoring methods. The methods can be tested on the historical data collected. This work demonstrates the development of such a method based on cepstrum to detect faults in wind turbine planetary stage. The method shows to be effective and does not necessitate knowing specific kinematical data of each turbine monitored.


Learning objectives
This work highlights the importance of collecting and storing time waveform data from wind turbines. It demonstrates how such data can be used to develop new monitoring methods. The work also demonstrates how to use cepstrum to detect faults in traditionally challenging part of a wind turbine, namely the planetary stage.