This paper provides an estimation model for noise signal characteristic diagnosis based on the Time-Frequency analysis technique. The marginal spectrum and statistical regression analysis is used as an estimation method for feature extraction. In order to approach the actual conditions, with the assistance of Department of Renewable Energy, Taiwan Power Company, the noise signals of a blade-damaged wind turbine and normal wind turbine are measured. In the case of low wind speed and noise, the time-frequency spectra are compared, and the feature magnification indicator is analyzed. The results show that the blade crack is caused by high frequency noise, mainly above 4000Hz. The time-varying analysis of the indicator shows that the index value is apparently enlarged when the damaged blade rotation is measured by microphone, and the number of damaged blades can be obtained.
以時頻率與邊際頻譜為基礎,提供噪音訊號特徵診斷預估模式。使用統計學迴歸分析為特徵抽取預估方法。在台灣電力公司再生能源處的協助下,進行葉片損傷及正常風機噪音測量。比較時頻譜差異,進行隨頻率分析之特徵放大指標判斷葉片是否損壞,也利用放大指標隨時間變化分析顯示葉片損傷數量,與實際狀況比較有相當理想之結果。