Data-driven and Model-based Bearing Fault Analysis - Wind Turbine Application
Abstract
This thesis is primarily about fault analysis for wind turbine generator bearings including vibration data analysis, and modeling of bearing outer race faults to understand how the vibration signal develops as the faults progresses. Case studies from large wind turbines are described where amplitude and phase demodulation are used to identify the outer race fault on the generator bearing.
In many applications of roller bearing fault diagnosis, such as in wind turbines, shaft rotational speed varies with respect to time, and consequently the normal frequency domain analysis, which is valid for constant speed condition, cannot be used. Alternatively, order tracking in conjunction with phase demodulation has been used to resample the time domain signal into angular domain, and estimate the shaft speed, respectively. Speed estimation using phase demodulation has been mainly used for gearbox applications where a reference signal or a shaft harmonic is well defined, which is not the case for bearings. In this thesis, Empirical Mode Decomposition (EMD) was used for decomposing the vibration signals, and shaft speed was estimated by phase demodulation. The estimated instantaneous shaft speed was then used for order tracking to capture the bearing fault, under variable speed profile. The method was applied on a simulated bearing signal having an outer race fault, and then on operational vibration data of a 1.5 MW wind turbine which had a low signal-to-noise ratio. In order to de-noise the vibration signals and select the most suitable decomposed mode for further analysis, an EMD-based de-noising algorithm including an indicator (combination of Kurtosis factor and Root Mean Square value) was proposed to eliminate the effect of unwanted components from other drive-train rotating elements as well as noise.
The idea of indicative fault diagnosis scheme based on the wind turbine tower vibration was demonstrated. It had been reported that a major fault on the generator bearing causes shock and noise to be heard from the bottom of the wind turbine tower. Thus, two accelerometers were attached orthogonally inside of the wind turbine tower. Tower vibration signals were analyzed using EMD and the outcomes were correlated with the vibration signal acquired directly from the generator bearings. It was shown that the generator bearing fault signatures are present in the vibration from the tower. The results suggest that useful condition monitoring of nacelle components, such as generator bearings can be done even when there is no condition monitoring system installed on the generator bearings, as is often the case for older wind turbines.
A dynamic model of a bearing with an outer race fault was developed including load zone, contact force and traction/friction forces. The model was validated both analytically, and using the generator bearing vibration data. Faults were modeled as simple surface step-like profile changes (to simulate small cracks), and as random sinusoidal (to simulate distributed defects). Different fault sizes were then simulated, and their effects on the vibration signal were analyzed. The results were compared with the historical generator bearing vibration data. This model can be ultimately be used to track the status of the fault size.
Description
Keywords
Applied Mechanics, Energy, Engineering--Environmental, Engineering--Industrial, Engineering--Mechanical
Citation
Mollasalehi, E. (2017). Data-driven and Model-based Bearing Fault Analysis - Wind Turbine Application (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. doi:10.11575/PRISM/25517