本研究旨在建立一套完整的風力發電機葉片表層損傷診斷系統。利用風力發電機運轉時，葉片所產生之噪音特 性，以時頻分析中的短時傅立葉轉換進行分析，首先將一無異音正常風力機，實測音訊並進行時頻分析，接著利用 邊際頻譜、聲壓分貝轉換、多項式迴歸等方法建立一個正常模式。並以此正常模式做為基準，往後測量之風力機皆 和此正常模式比較，計算出指標，利用指標判斷風力機葉片表層損傷與否。最後利用ROC 曲線理論，界定出最佳閾 值，在不停止運轉的情況下，便可得知風力機葉片表層損壞情況，本文中的檢測結果以風力機葉片實際照片作為驗 證，期許未來可以應用在風力發電機健康檢測系統上。

The main purpose of this study is to establish a complete wind turbine blade surface damage diagnosis system. The noise characteristics of the blades are generated by the operation of the wind turbines. In this article, the method of time-frequency signal is analyzed by a short-time Fourier transform. In order to establish the normal module, first, we measure the sound signal of a normal wind turbine, and analyze sound signal by a short-time Fourier transform. Second, we use marginal frequency, decibel transformation polynomial regression, and so on. Regarding the normal module as a reference, we can compare it with other wind turbines to calculate the index. After calculating the index, we use the ROC curve theory to determine the optimal threshold. The optimal threshold can estimate the situation of surface damage on the wind turbine blade. We can detect damage while turbines are operating. The final result of the paper is verified by photo, and we look forward to applying it on the health detection system of wind turbine.