小波轉換於風力發電機葉片診斷之應用

Health diagnosis for wind turbine blades using wavelet transform

秦正宇(Jeng-Yu Chin)

指導教授 ： 王昭男

國立臺灣大學/工學院/工程科學及海洋工程學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201601551

摘要

本研究主要目的在分析風力發電機運轉時的葉片噪音特性，以進行葉片損傷之診斷。風力發電機在運轉些許時日後其各部件會有些許損壞，而不同的損壞會造成不同的訊號特徵，又以葉片破損會在運轉時產生特殊的噪音，因此葉片的損壞判別可以聽覺判斷。而傳統葉片的損壞維修是每隔數月才請人去聽葉片所發出之噪音來做判斷，陸上風機便已如此曠日廢時，未來發展出的海上離岸風力發電機難以想像多久才會發現葉片損壞。
時頻分析是訊號處理上實用的工具，可以解析出訊號時間與頻率上的對應關係，進而了解訊號所透露出的訊息。將時頻分析方法應用於風機健康診斷，本研究所採用的時頻分析方式為連續複數Morlet小波分析。研究概略流程為將一無異音正常風機實測音訊作時頻分析後的時頻對應關係，利用邊際頻譜、聲壓分貝轉換、A加權、線性迴歸等方式得到一個正常模式。並以此正常模式為基準與其他風機訊號計算殘差和並且以此得到指標，便可推估此風機之葉片損壞情況。許多的風機健康診斷研究需要將風機停機、拆機做測試才能診斷出損壞，本研究可在風機運轉情況下測量其損壞情況，可大幅降低時間、人力、金錢成本。本研究計算數台風機的實測音訊，且以風機葉片照片作為驗證。

關鍵字

小波分析 ； 邊際頻譜 ； 迴歸分析 ； 風機葉片診斷

並列摘要

The major object of this research is to analyze the noise feature of wind turbine while operating and to detect blade damage with noise feature. Many components of wind turbine will be worn after days of running. Different types of damage would have different sorts of signal feature, and the noise of worn turbine blade is extraordinary. Thus, the damage of turbine blades could be judged by hearing. Generally, it takes months for a concerned department to notice blade damage after sending employees to listen to the noise of turbine. If on land wind turbines health diagnosis is such a waste of time, the time consumption of off shore wind turbines would be unbelievable.
Time-frequency analysis is a useful method of signal processing, it shows the cor-respondence between time and frequency and analyze what the signal stands for. This research applies time-frequency analysis to wind turbine health diagnosis with contin-uous complex Morlet wavelet analysis. The approximate process is to analyze sound wave of an undamaged wind turbine with wavelet transform, marginal frequency, decibel transformation, A-weighting and polynomial regression. By this process we can build a normal model to calculate the sum of squared errors between normal mod-el and damaged wind turbine blades noise, thus we can estimate damage severity of the blades. Most of other research of wind turbine health diagnosis need to stop and take apart wind turbines to detect damage. Our results shows we can detect damage while turbines are operating, which reduce time, labour and money consumption sig-nificantly. We record several sets of sound of wind turbines and represent the results of our detection, which will be proved by blade pictures.

並列關鍵字

wavelet analysis ； marginal spectrum ； polynomial regression ； wind turbine blades health diagnosis.

參考文獻

2. 張雅茹, "翼尖小翼對風車葉片流場及空氣動力噪音之影響探討,"工程科學及海洋工程研究所.國立台灣大學, 2011.

3. 袁淑嫈,"二維風車葉片之氣動力噪音分析,"工程科學及海洋工程學研究所. 國立台灣大學, 2008

5. 蕭雲璠, "創新小型風力發電機葉片變旋角控制系統設計與發電機轉子定轉速控制之研究," 工程科學及海洋工程學研究所. 國立台灣大學,2015

7. 紀騰揚, "風力發電機關鍵零組件故障診斷之研究, "工程科學及海洋工程學研究所. 國立臺灣大學,2010

8. 熊婉贏, "研究型風機葉片之振動分析及損壞偵測, "土木工程學研究所. 國立臺灣大學,2014