Fault Diagnosis of Gearbox Based Pitch Drives in Wind Turbines : Fault Detection by Support Vector Machine Using Motor Current Signal Analysis

Abstract

The growing dependence on wind power in recent years has increased the demand for reliantwind turbines. The pitch system of a wind turbine is one of the components with the highest failure rates. The most common way of diagnosing pitch system faults is currently through vibration analysis, which requires the installation of vibration sensors. This thesis presents a non-intrusive method for fault detection of the planetary gearbox in an electric wind tur-bine pitch system. The method is based on using the three-phase motor currents from the induction motor of the pitch system to calculate a DC offset using Extended Park’s vector approach (EPVA). Basic statistical formulas are used to extract features from both the time-and frequency-domain of the DC offset, where fast Fourier transform (FFT) is used to find the frequency-domain values. These features, along with the amplitudes of the characteristic frequencies of the planetary gearbox and its bearing, are used in the principal component analysis(PCA) to generate features that are used to train a support vector machine (SVM) classifier. This method is validated by using labeled data from the induction motor of a pitch system testbench to classify three health conditions. One of the health conditions are a healthy system, and the two other are artificially seeded faults in the system’s two-stage gearbox. These faults are a partially cracked tooth in one of the first stage planet gears, and an outer race fault inthe bearing at the input shaft. The results indicate that the proposed method is capable of classifying each of the three health conditions.