Determining the speed of a variable speed wind turbine from the vibration response

Condition monitoring of wind turbines is very important to minimise maintenance costs (virtually the only costs once they are installed), and to maximise production. New diagnostic techniques have had to be developed to deal with the varying speed and load of the most efficient turbines, but these are greatly aided by having a measure of the speed. This paper shows how the speed of a wind turbine was extracted very accurately from the response vibration signal, making use of a phase demodulation method for determination of a rotational angle vs time map, as used for order tracking. The angle vs time relationship can be determined by the phase demodulation method to any degree of resolution if the corresponding carrier frequency and its sidebands are isolated in the frequency domain with no encroachment from adjacent components. For the case used here for demonstration of the method, the speed range of each signal section had to be limited to about ± 15% to avoid overlap. The overall signal was divided into eight overlapping sections, in each of which a suitable candidate harmonic was isolated, and phase demodulated in two iterations, the second able to use higher harmonics for increased accuracy once separated by the first iteration. Specially designed window functions allowed the results from each section to be smoothly joined. Noise from the differentiation of these phase curves to rotational speed was removed by two methods, which gave very similar results; lowpass filtration in the frequency domain (which however gave some distortion at the two ends) and polynomial curve fitting, which was chosen in this case. The final speed vs time record over the whole record length could then be order tracked using the speed curve itself (integrated to obtain a phase vs time map) to express it as speed vs rotation angle.