基于形態(tài)成分分析的汽輪機(jī)轉(zhuǎn)子早期碰摩微弱故障診斷

王志剛,王宏超

(1.安陽(yáng)工學(xué)院 機(jī)械工程學(xué)院 河南 安陽(yáng) 455000; 2.鄭州輕工業(yè)學(xué)院 機(jī)電工程學(xué)院,河南 鄭州 450002)

基于形態(tài)成分分析的汽輪機(jī)轉(zhuǎn)子早期碰摩微弱故障診斷

王志剛1,王宏超2

(1.安陽(yáng)工學(xué)院 機(jī)械工程學(xué)院 河南 安陽(yáng) 455000; 2.鄭州輕工業(yè)學(xué)院 機(jī)電工程學(xué)院,河南 鄭州 450002)

作為電力行業(yè)的關(guān)鍵咽喉設(shè)備-汽輪機(jī),對(duì)其早期微弱故障進(jìn)行有效診斷有著重要的安全及經(jīng)濟(jì)意義.汽輪機(jī)發(fā)生早期動(dòng)靜碰摩故障時(shí)信號(hào)主要為轉(zhuǎn)子工頻成份及轉(zhuǎn)子與支撐間碰摩所致瞬態(tài)沖擊信號(hào)成份.據(jù)形態(tài)成分分析的原理,分別構(gòu)建正弦基及沖擊原子庫(kù),對(duì)故障信號(hào)進(jìn)行匹配分析,進(jìn)而實(shí)現(xiàn)轉(zhuǎn)子早期動(dòng)靜碰摩信號(hào)工頻成份與瞬態(tài)沖擊成份的有效分離,對(duì)碰摩故障進(jìn)行模式識(shí)別.通過(guò)仿真信號(hào)及轉(zhuǎn)子碰摩實(shí)驗(yàn)信號(hào)驗(yàn)證形態(tài)成分分析方法在轉(zhuǎn)子早期碰摩故障診斷中的有效性.

形態(tài)成分分析; 汽輪機(jī); 轉(zhuǎn)子碰摩; 故障診斷

0 引言

汽輪機(jī)往往是電力行業(yè)的關(guān)鍵咽喉設(shè)備,一旦發(fā)生故障不但會(huì)造成嚴(yán)重的經(jīng)濟(jì)損失,而且還有可能造成嚴(yán)重的人員傷亡事故.為提高汽輪機(jī)的轉(zhuǎn)速,汽輪機(jī)轉(zhuǎn)子與支撐軸承的間隙往往非常小,以致非常容易造成轉(zhuǎn)子動(dòng)靜碰摩故障的發(fā)生,若能對(duì)轉(zhuǎn)子的早期碰摩故障進(jìn)行有效的診斷,就能防患于未然,進(jìn)而實(shí)現(xiàn)經(jīng)濟(jì)與安全效益的最大化.對(duì)于轉(zhuǎn)子的動(dòng)靜碰摩已經(jīng)有相當(dāng)?shù)奈墨I(xiàn)報(bào)道[1-3],就針對(duì)汽輪機(jī)轉(zhuǎn)子的動(dòng)靜碰摩鮮有研究.形態(tài)成分分析(Morphological Component Analysis,MCA)[4]是近幾年剛剛發(fā)展起來(lái)的基于稀疏表征和形態(tài)差異性的信號(hào)表征方法,它是由稀疏表征基礎(chǔ)理論發(fā)展而來(lái),并在圖像紋理分離和修復(fù)、腦電信號(hào)分析等領(lǐng)域迅速得到了應(yīng)用[5,6].在機(jī)械故障診斷中,不同的振源反應(yīng)到信號(hào)中的結(jié)構(gòu)往往存在一定的差異,如旋轉(zhuǎn)機(jī)械中常見(jiàn)的諧波成分、沖擊成分及其他瞬態(tài)成分等,這些多成分的共存導(dǎo)致原始信號(hào)較為復(fù)雜,給故障特征提取帶來(lái)了難度.形態(tài)成分分析的基本思想就是基于信號(hào)內(nèi)部不同結(jié)構(gòu)的形態(tài)相異性,通過(guò)構(gòu)建不同類(lèi)型的字典稀疏表示不同的信號(hào)結(jié)構(gòu)分量,進(jìn)而實(shí)現(xiàn)不同形態(tài)成分的分離.

1 形態(tài)成分分析

對(duì)于任意包含K個(gè)不同結(jié)構(gòu)分量的信號(hào)s,我們期望找到它在所有字典Φk下的最稀疏表達(dá):

(1)

由于字典Φk原子的單一性,可用其高效表達(dá)信號(hào)中的某一成分,而無(wú)法有效表達(dá)其他成分,則式(1)優(yōu)化問(wèn)題的求解保證了信號(hào)各個(gè)成分的分離.根據(jù)稀疏分解相關(guān)理論,式中的l0范數(shù)問(wèn)題是非凸的難以實(shí)際求解,將l0范數(shù)替換為l1范數(shù)則可以將式(1)優(yōu)化問(wèn)題通過(guò)線性規(guī)劃求解,且l1范數(shù)可以保證解的稀疏性:

(2)

放寬式(1)的約束條件,我們可以得到:

(3)

(4)

對(duì)于式(4)的優(yōu)化問(wèn)題存在一種快速數(shù)值求解算法Block-Coordinate Relaxation Method,具體可參看文獻(xiàn)[7],在此不再贅述.

基于Block-Coordinate Relaxation Method,式(4)的優(yōu)化問(wèn)題求解可轉(zhuǎn)化為如下數(shù)值算法:

(1) 設(shè)定迭代次數(shù)Lmax,閾值δ=λ·Lmax,初始化形態(tài)分量sk=0,k=1,…,K;

(2) 執(zhí)行K次循環(huán):更新形態(tài)分量sk,假定其他形態(tài)分量sl,l≠k確定:

—計(jì)算稀疏系數(shù)αk=Tkr

(3) 更新閾值δ=δ-λ;

(4) 如果δ>λ則返回步驟2,否則算法結(jié)束.

2 實(shí)驗(yàn)

圖1是某企業(yè)汽輪機(jī)正常運(yùn)行時(shí)所采集到振動(dòng)信號(hào)的時(shí)域波形圖,汽輪機(jī)轉(zhuǎn)子轉(zhuǎn)速為3 000 r/min.使用鄭州恩普特設(shè)備診斷科技股份有限公司研制的PEDS-F儀器對(duì)振動(dòng)信號(hào)進(jìn)行采集,配以位移傳感器.信號(hào)采樣頻率fs=2560 Hz.取前2560個(gè)點(diǎn)進(jìn)行分析.由圖1可以看出,其波形圖基本為正弦,符合轉(zhuǎn)子正常運(yùn)行的理論特征.

圖1 正常轉(zhuǎn)子時(shí)域圖Fig.1 The time waveform of normal state

轉(zhuǎn)子發(fā)生早期微弱碰摩故障時(shí)的時(shí)域波形圖見(jiàn)圖2.由圖2看出,轉(zhuǎn)子正常運(yùn)行的正弦特征已不存在,而由碰摩所致瞬態(tài)沖擊成分不明顯,由該波形圖上較難判斷轉(zhuǎn)子是否發(fā)生碰摩故障.采用本文所述方法對(duì)圖2信號(hào)進(jìn)行分離,結(jié)果見(jiàn)圖3.圖3(a)為故障信號(hào)正弦成分,主要為轉(zhuǎn)子工頻成分;圖3(b)為瞬態(tài)沖擊成分,經(jīng)計(jì)算兩沖擊峰值間隔約為0.02 s;圖3(c)為分解余項(xiàng),主要為背景噪聲等成分.

為突出所述方法的優(yōu)越性,以下將給出基于總體經(jīng)驗(yàn)?zāi)B(tài)分解方法(Ensemble Empirical Mode Decomposition,EEMD)的分析結(jié)果.EEMD是經(jīng)驗(yàn)?zāi)B(tài)分解(Empirical Mode Decomposition,EMD)方法的延伸.EEMD相對(duì)于EMD方法不僅保留了其原有的所有優(yōu)點(diǎn),而且還克服了EMD方法固有的頻混現(xiàn)象的弊端,分解精度更高.

圖2 轉(zhuǎn)子動(dòng)靜碰摩時(shí)域圖Fig.2 The time waveform of rub-impact state

圖3 故障信號(hào)分解結(jié)果Fig.3 The separation result of the rub-impact fault signal

圖4是汽輪機(jī)轉(zhuǎn)子碰摩故障信號(hào)(即圖2)基于EEMD的分解結(jié)果,相對(duì)于所述方法的分析結(jié)果,效果相差甚遠(yuǎn).說(shuō)明了所述方法的優(yōu)越性.

3 結(jié)論

利用形態(tài)成分分析方法可以實(shí)現(xiàn)復(fù)雜故障信號(hào)中不同結(jié)構(gòu)成分的分離,進(jìn)而提高故障探測(cè)的準(zhǔn)確性.與基于傳統(tǒng)方法相比,它能充分利用信號(hào)內(nèi)部的結(jié)構(gòu)特征,不存在因頻帶重疊等帶來(lái)的困擾.基于此,提出了基于MCA的汽輪機(jī)轉(zhuǎn)子早期碰摩微弱故障信號(hào)的故障診斷方法,為汽輪機(jī)轉(zhuǎn)子動(dòng)靜碰摩故障診斷方法研究提供有益參考價(jià)值.

圖4 EEMD分解結(jié)果Fig.4 The decomposition results by EEMD

[1] Chu Fulei,Lu Wenxiu.Experimental observation of nonlinear vibrations in a rub-impact rotor system [J].Journal of Sound and Vibration 283(2005)621-643.

[2] Cheng Junsheng,Yu Dejie,Tang Jiashi,et al.Local rub-impact fault diagnosis of the rotor systems based on EMD[J].Mechanism and Machine Theory,2009(44):784-791.

[3] Li Yungong,Zhang Jinping,Wang Liqiang,et al.A fault feature extraction method for rotor rubbing based on load identification and measured impact response[J].International Conference on Advance in Engineering,Procedia Engineering,2011(24):793-797.

[4] BOBIN J,MOUDDEN Y,STARCK J-L.Enhanced source separation by morphological component analysis;proceedings of the Acoustics,Speech and Signal Processing[A].2006 ICASSP 2006 Proceedings 2006 IEEE International Conference on,F,2006[C].IEEE.

[5] ELAD M,STARCK J L,QUERRE P,et al.Simultaneous cartoon and texture image inpainting using morphological component analysis(MCA)[J].Appl Comput Harmon A,2005,19(3):340-58.

[6] YONG X,WARD R K,BIRCH G E.Generalized morphological component analysis for EEG source separation and artifact removal[A].proceedings of the Neural Engineering,2009 NER′09 4th International IEEE/EMBS Conference on,F,2009 [C].IEEE.

[7] DONOHO D L,JOHNSTONE J M.Ideal spatial adaptation by wavelet shrinkage [J].Biometrika,1994,81(3):425-55.

Early rub-impact fault diagnosis of turbine rotors based on morphological component analysis

WANG Zhi-gang,WANG Hong-chao

(1.AnYang Institute of technology,School of mechanical engineering,Henan Anyang 455000,China;2.Zhengzhou light industry institute,School of mechanical electricity,Henan Zhengzhou 450002,China)

As a key equipment for electric power industry, effective diagnosis on early rub-impact faults from turbine rotors possesses safety and economic significances. When the early rub-impact faults occur in rotor system, the vibration signals are mainly composed of periodic components of rotor rotation frequency and transient impulse signal components of rub-impact between rotor and bearing. Firstly, the sine-based and impulse-based dictionaries are respectively established based on morphological component analysis (MCA) theory. Then, these dictionaries are used to match the sine and impulse components. Finally, the early rub-impact vibration signals are effectively decomposed into two major components. Therefore, the feasibility and effectiveness of this method are verified using simulation and testing.

morphological component analysis; turbine; rotor rub-impact; fault diagnosis

王志剛(1977-),男,講師,碩士.E-mail:hongchao1983@126.com

TH 212; TH 213.3

A

1672-5581(2016)06-0545-03