風力發電在綠能產業具有潛力，其中風機運轉與健康狀況關係到發電的效能與安全性，為了檢測風力發電機的運轉狀況，先以接觸理論分析軸承損壞下的受力情況，軸承在不同部位損壞下會出現特定頻率的動態訊號，對此訊號使用短時距傅立葉分析(short-time fourier transform)、總體經驗模態分解法(Ensemble Empirical Mode Decomposition)與快速譜峰度法(Fast Kurtogram)做診斷。第一個案例中是已知軸承損壞的機械系統，利用上述訊號處理方法對訊號進行分析，並比較不同方法間之優劣。後再將此分析法運用在第二個分析案例，即分析風力發電機上量得之振動訊號；結果顯示即使分析元件包含齒輪與軸承，但使用快速譜峰度法仍然能找出疑似軸承損壞訊號；當使用短時距傅立葉時，會偵測到齒輪轉速訊號與軸承損壞訊號，判別上較為複雜；而使用總體經驗模態分解法軸承訊號所在之階層並不固定，必須每一階層觀察效率較差。;The goal of the research was to diagnose the bearing faults in rotary machines though examining their dynamic responses or vibration signals. A bearing has specific defect vibration frequencies which can be calculated from bearing design data. Short-time-Fourier transform, ensemble empirical mode decomposition (EEMD) and fast kurtogram were signal processing methods for detecting bearing faults. These signal processing methods were used to analyze the dynamic responses from a rotor system with a damaged bearing and were compared to each other. Then, the methods were applied to vibration signals from a small and medium wind turbine. The signals contain not only bearing signal but also gear mesh signal. The results show short-time-Fourier method could detect the bearing defect, the gear mesh signal and its harmonics, but the bearing fault signals are too complex to diagnose. Applying the fast kurtogram method which would usually filter out gear mesh signal and its harmonics could detect bearing fault signal more effective due to the transient characteristic of bearing fault signals. Using EEMD to analyze the vibration signals would decompose the original signal to several intrinsic mode functions (IMFs) possibly containing bearing fault signal in some decompositions. However, one still need to check each IMF for effectiveness.
