透過您的圖書館登入
IP:54.221.110.87
  • 學位論文

結合經驗模態分解(EMD)能量轉換與支援向量機 於冷氣機失效模式辨識

Synthesis of EMD Energy Transformation and SVM for Identifying Fault Models of Air Condition

指導教授 : 許俊欽
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


小波轉換分析(Wavelet Transform Analysis)是最常用於研究各個領域的訊號分析之工具。小波轉換同時擁有時間域(time domain)與尺度(scale)的解析資訊,具有強大的解析能力。然而小波轉換之主要缺點為採用固定的小波基底(Wavelet-based)來分解訊號,如果基底無法符合訊號之特性,那將失去訊號分析之意義。 近年來,經驗模式分解(Empirical Mode Decomposition, EMD)被發展出來用於非穩態(Non-stationary)和非線性(Non-linear)訊號處理上,EMD能夠將信號內部變化的時間尺度作為能量與頻率析出,將信號分解成數個本質模態函數(Intrinsic Mode Function, IMF),其IMF包含信號中不同尺度的特性,能夠充分表達信號中之物理特性。目前廣泛被應用在故障診斷、邊緣偵測、訊號處理、影像處理、資料壓縮、地震波分析等工程領域。 本研究主要目的在於發展「智慧型聲音檢測系統」,用來做為各種聲音異常失效之診斷。所提之方法主要分為兩個子系統: 1.聲音收集子系統 聲音收集子系統主要目的建立收集風扇馬達聲音訊號之實驗平台。透過陣列式麥克風對於馬達聲音異常收集,並藉由PXI控制器將聲音資料做儲存。 2.聲音檢測子系統 此系統結合(EMD)與支援向量機(Support Vector Machine ,SVM)去進行失效模式診斷與分類。主要是藉由EMD將聲音訊號分解成數個本質模態函數,經由能量轉換後,以SVM做失效模式辨識與分類。 本研究以窗型冷氣機作為研究案例,使用智慧型聲音檢測系統對於冷氣機風扇馬達系統異常失效診斷。所提方法之有效性,將與傳統故障診斷方法進行驗證比較。比較方法為小波轉換結合倒傳遞類神經網路(BPNN)、小波轉換結合SVM、EMD結合BPNN和EMD結合SVM。實驗結果顯示EMD結合SVM能有效提升冷氣機風扇馬達系統異常失效模式故障診斷之能力。

並列摘要


Wavelet Transform Analysis is the most commonly used tool to study various fields of signal analysis. Although Wavelet Transform Analysis can resolve the information of time domain and scale, the disadvantage of the Wavelet Transform Analysis is that it uses the fixed base to analyze signals. Consequently, if the basement can not meet the characteristics of the signal, it will lose the meaning of signal analysis. In recent year, Empirical Mode Decomposition (EMD) has been developed to use in the signal analysis of Non-stationary and Non-linear. EMD can change the time scale of the signal’s internal changing as energy and frequency and separate out, and decompose the signal into several Intrinsic Mode Function (IMF). IMF contains the characteristics of different scale in signal; it is able to fully express the physical characteristics of the signal. EMD is now widely used in fault diagnosis, image edge detection, signal processing, data compression, seismic waves’ analysis and other engineering fields. The purpose of this study is to develop the Intelligent Sound Detection System and using it as the diagnosis of abnormal sound. The proposed method is mainly divided into two subsystems: 1.Sound Collection Subsystem: Mainly composed of PXI controller and a microphone array. This system is mainly used microphone array to make the sound anomaly detection, and then by the PXI controller to do the data storage of detected voices. 2.Abnormal Voice Recognition System: This system is composed of EMD and Support Vector Machine (SVM), and then diagnosis and classify failure modes. It is mainly by the sound detection system to collect unusual sound signal, by using EMD to decompose the sound signal into several IMF, then the SVM to classify the failure modes. In this study, window-type air-conditioners as a case study, using an Intelligent Voice Detection System for air-conditioner fan motor system to do unusual failure diagnosis. The effectiveness of the proposed method was compared to traditional methods. Comparison methods are combined the Wavelet Transform with Backpropagation Neural Network (BPNN), combined the Wavelet Transform with SVM, combined the EMD with BPNN and combined the EMD with SVM. Experimental result shows that, EMD combined with SVM is the best type for the air-conditioner fan motor system’s failure modes’ diagnosis and classification. The method in this study can effectively enhance the ability of fault diagnosis.

參考文獻


[31] 陳文杰,「雜訊下經驗模態分解法於語音辨識之應用」碩士論文,國立中央大學電機工程研究所(2006)。
[7] C. W. Hsu, C. C. Chang & C. J. Lin, 2003, A practical guide to support vector classification.
[1] B.A. Paya, I.I. Esat, M.N.M. Badi, Artificial neural network based fault diagnosis of rotating machinery using wavelet transforms as processor, Mechanical Systems and Signal Processing ,11 (5) (1997) 751–765
[3] Chih-Chung Chang and Chih-Jen Lin*,2001, a Library for Support Vector Machines. vailable:http://www.csie.ntu.edu.tw/~cjlin/libsvm.pdf
[4] C. Cortes and V. Vapnik. Support-vector network. Machine Learning 20: (1995) 273-297.

延伸閱讀