- 學位論文
短時傅立葉轉換於風力發電機葉片表層損傷即時診斷之應用
Application of Short-Time Fourier Transform for Real-Time Surface Damage Detection of Wind Turbine Blades
摘要
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。風力發電機長時間暴露於外頭，嚴峻的外在環境，常常造成風機各部件損壞，其中又以葉片損傷最為常見，需要定期維修檢測，避免損壞造成的安全問題及經濟損失。傳統的葉片檢測方法是由專業人士做定期性維修檢測，以個人經驗做聽覺辨別，必要時，需將風力發電機停止運轉，由擁有國外原廠認證的技師，搭上高空作業車進行維修，操作上的危險性非常高。另外，許多風力發電機興建於地勢險峻的環境中，使用此種傳統檢測方法並不方便，近年來發展的海上離岸風力發電機更是如此。隨著各國風力發電場的數量快速增加，擁有一套完整的風力發電機葉片表層損傷診斷系統是必要的。 本論文利用風力發電機運轉時，葉片產生之噪音，人員在風機下方架設麥克風，量測音訊，並以時頻分析中的短時傅立葉轉換做計算。利用MATLAB軟體進行分析，首先將一台無異音正常風機，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套正常模式，並以此正常模式作為基準，往後測量之風機皆與此正常模式比較，再透過本文定義之公式計算指標，作為葉片是否損傷的判斷依據，最後利用接收者操作特徵曲線與其曲線下面積，界定最佳損傷閾值。此檢測方法能夠在風力發電機不停止運轉的情況下，由電腦自動診斷，得知風機葉片的損壞狀況，分析出葉片損壞數量。本文中的檢測結果會以風機葉片實際維修照片作為驗證，期許未來可以應用在風力發電機健康檢測系統上。
並列摘要
The main idea of this research is to establish a complete system for surface damage detection of wind turbine blades. Because of long-term exposure in an extreme environment, the damage on wind turbine’s components is inevitable, and mostly on the blades. Thus, wind turbine requires regular detection and maintenance to avoid safety issues and financial loses. The traditional assessment for surface damage detection of wind turbine blades is to evaluate the operation noise by a professional using his ears to listen. Sometimes, judging by the professional’s personal experience, he has to stop the wind turbine and execute the detection from the above to the blades. Not only does it demand special skills, it is also accompanied with high risk caused by the extreme environment wind turbine was built at. So, it is very dangerous and inconvenient for us if we must count on ears or climbing skills to conduct the assessment, not to mention the recent arising of offshore wind turbine. With the number of global wind turbine rapidly increases and develops, establishing a complete system for surface damage detection of wind turbine blades becomes more and more essential. This research using time-frequency analysis short-time Fourier transform from MATLAB to analyze the blades’ noise while operation, by placing microphone under the wind turbine and recording. First, to build a normal model as foundation for comparison, we put recordings of a wind turbine without abnormal sound into time-frequency analysis and applying it to methods like marginal spectrum, decibel transformation and polynomial regression. Then, based on the normal model, we defined the formula this research shows, as an indicator to determine if the blades are damaged. Finally, we can also define the optimal damage threshold by utilizing the receiver operating characteristic curve and calculating the area under the curve. This new way for detection allows us to receive blades’ damage reports by computer’s automatically diagnosis without affecting or stopping the wind turbine’s operation. This research’s detection results are testified by actual photo from wind turbine’s blades. Hopefully, it can be applied as wind turbine’s health detection systems in the future.