华佗养生网
您的当前位置:首页改进奇异值分解方法在高维动力学系统降维中的应用

改进奇异值分解方法在高维动力学系统降维中的应用

来源:华佗养生网
东 北 石 油 大 学 学 报

第43卷 第2期 2019年4月JOURNAL OF NORTHEAST PETROLEUM UNIVERSITY Vol. 43 No. 2 Apr. 2019DOI 10. 3969/j. issn. 2095-4107. 2019. 02. 013改进奇异值分解方法在高维动力学

系统降维中的应用唐莉(大庆师范学院教师教育学院,黑龙江大庆163712 )摘要:基于奇异值分解(SVD)方法和高维动力学系统的瞬态响应特性,提出改进奇异值分解方法;应用牛顿第二

定律,建立自由度为16的两端带有滑动轴承支承和发生基础松动故障的转子系统模型;采用改进奇异值分解方法,将原 始系统降维为自由度为3的简化系统模型,对比原始系统与筒化系统模型动力学特性。结果表明:降维后的简化系统模

型保留原始系统的主要分岔特性及均方差幅值特性,验证改进奇异值分解方法对高维动力学系统降维的有效性。该结 果为高维非线性动力学系统定性分析提供指导。关 键 词:改进奇异值分解方法;动力学系统;转子;降维;分岔;均方差幅值中图分类号:V231. 96 文猷标识码:A 文章编号:2095 -4107(2019)02 -0119 -060引言很多实际工程系统是高维复杂的动力学系统,如航空发动机和燃气轮机等。大型旋转机械结构复杂,

非线性因素多,虽然可以建立高维等效的动力学模型,但维数巨大,很难进行非线性动力学特性分析 当转子系统发生故障时,故障的强非线性与转子运动耦合在一起,很难对故障转子系统进行动力学分析。 因此,需要研究高效的大型旋转机械转子系统的降维方法,大幅降低原始系统的维数,用简化系统模型代

替原始系统进行动力学分析。常见的降维方法为中心流形方法図、李雅普诺夫斯密特约化方法间、近似惯性流形方法⑷、Galerkin

方法旳、奇异值分解(SVD)方法和本征正交分解方法⑷等。REGA G等卬对非线性动力学应用进行研 究,综述动力学系统模型中常用的降维方法。奇异值分解方法和本征正交分解方法被广泛应用于高维复 杂系统降维⑷。YU Hai等⑷将本征正交分解方法应用于两端带有滑动轴承支承的转子系统模型,并验

证方法的有效性。LU Kuan等「遡对本征正交分解方法进行综述,给出分类方式,并展望在工程系统中的 应用。常见的转子系统故障包括碰摩故障、裂纹故障、基础松动故障等刃o人们研究基础松动故障,MA Hui等[⑹建立基础松动转子系统模型,研究松动参数对系统动力学特性的影响。陈果匚⑷建立含不平衡碰 摩基础松动耦合故障的滚动轴承的转子系统模型,分析滚动轴承间隙、碰摩刚度、转子偏心及轴承座质量 对系统动力学特性的影响。目前,未见基于瞬态响应的SVD方法应用于转子系统降维。笔者提取转子系统的瞬态响应信息并提

出改进SVD方法;应用牛顿第二定律,建立两端带有滑动轴承支承和发生基础松动故障的转子系统模型; 采用改进SVD方法对原始系统降维得到简化系统模型,对比原始系统与简化系统模型的分岔与幅频响应

特性,验证改进SVD方法有效性。1奇异值分解方法1.1传统方法任意mXm阶实对称矩阵A形式为收稿日期:2018-12-11;编辑:蔡田田基金项目:国家自然科学基金项目(11802235)作者简介:唐莉(1981-),女,硕士,讲师,主要从事动力学系统混沌与分岔方面的研究。• 119 •东北 石油大 学学报第43卷2019年A = UAUt , (1)式中:U为标准正交矩阵;A为对角阵。U的列是矩阵A的特征向量,A的对角元素是A的特征值。如果

A是正定的,特征根就是正数。定义U的第i列为冷,特征值为为A的对角元素),则Alii =

一个nXm阶实矩阵B分解形式为o (2)(3)B = urvT,

式中:U为\"Xm阶有标准正交列的矩阵(UtU = E) ;V为mXm阶标准正交矩阵(卩丁卩=£) ;F为mXm阶

有非负元素的对角阵,对角元素为奇异值。构造两个正定对称阵BBt和分解为bbt = urv'rvruT = urzuT,

btb = vr2vT,

⑷(5)其中 BBt 是\"X\"阶,BtB 是m Xm 阶,n^-ma根据式(5),确定BTb的特征向量和特征根分别为V的列和的对角元素。设bTb的一个特征向

量为的对角元素为7,得到BtBv=/v,两边左乘B得BBtBv = fBv,

则BBt的特征向量“ = Bv,特征值为/ o(6)1.2改进方法传统SVD方法「旧是对动力学系统的稳态过程信息进行提取,改进SVD方法考虑系统瞬态响应信 息。在给出的初始条件下,系统的瞬态过程复杂,包含系统的自由振动信息和强迫振动信息。因此,基于 提取的瞬态过程信息,采用改进SVD方法得到的简化系统模型更完整地保留原始系统的动力学特 性E6ao] o从系统瞬态过程得到一组模态,取所需的前几阶模态作投影空间,将原始系统投影到前几阶模态

张成的子空间,得到低自由度的近似等价模型。多自由度系统动力学方程表示为Z=—C2—KZ + F,

式中:Z为位移向量;C为阻尼矩阵;K为刚度矩阵;F为外激励向量。(7)改进SVD方法数学构造过程:给出动力学系统初始条件,提取每个自由度的位移,记为Z! (t) ,z2 (¢), …,况(t),其中L为系统的自由度个数,将各自由度等时间间隔的位移序列记为= 九),召仏),…,N

(切))丁,: = …,L。每个自由度时间间隔个数N相等,时间序列构成的矩阵为X=[zj,Z2,…,zj,其 阶数为NXL,得到阶数为LXL的自相关矩阵T = XTXa令…,仇为相应特征向量,特征根为

A!>A2> - >Alo U为T前”阶特征向量组成的矩阵,包含T前”个最大特征根,矩阵UTu的阶数为

n Xno进行坐标变换,得到新向量P,记为Z=UP,将Z代入式(7)得UP =-CUP-KUP + F,

p =一 (uTu'-'iTcuP 一 (.^Uy^KUP + (旳厂“尸。

(8)(9)令 CR = (UTU)T[JTCU,KR = (UTU)TUTKU,FR = (UTU)TUTF,则p=-crp-krp + frq

(10)改进SVD方法将原始高维动力学系统降维到自由度为”的简化系统模型。简化系统模型的能量(简 化系统模型”个特征根的和与原始系统L个特征根的和的比例)占原始系统的95. 0%以上。2转子系统建模建立两端带有滑动轴承支承和发生基础松动故障的转子系统模型(见图1),列出转子系统动力学方・120・第2期唐莉:改进奇异值分解方法在高维动力学系统降维中的应用程。其中,6、。7分别为左右轴承的几何中心;0”(\" = 2,3,・・・,6)为各圆盘的几何中心;0®为各圆盘的重心;

^(/ = 1,2,-,7)为等效的集中质量;叫、观9分别为轴承两端松动块质量;^(Z = l,2,-,6)为相应各盘的

等效刚度;和、怎2分别为轴承两端的松动刚度;°为集中质量处的等效阻尼;如、°2分别为轴承两端基础松

动端的阻尼;&为松动端和基础之间最大间隙。松动刚度和阻尼是分段线性的,表达式参考文献口门。盘1Fig. 1 Rotor system model采用短轴承假设,油膜力表达形式参考文献[20]。根据牛顿第二定律,得到自由度为16的转子系统动力学方程为泌+ &+KZ = F,式中:M为质量矩阵,ClCl(IDM= m7 ;C=m7观8观9—居16X16ki—b\\—尿—ki~k2k\\ + kz—层K=~k5

k5 +k6—尿—层居+上6k.—上6—爲爲bs2 ” 16X16・121・东北石油大学学报第43卷 2019年F’(Xi,Yi—丫8, 乂i,E—久) 、Fy (Xi, Yi -y8,禹,E — % )—阻 g

m2B2w2 cos(et)m2B2&/sin(®t) ~m2gF=msB6a)2 cos(F.(X7,y7,X7,Y7)F/x7,y7,x7,y7)-m7g—Fy(Xi,Yi—丫8, 乂1,V1—£)—gg-F/x7,y7-Y9,x7,y7-Y8)-m9g>向的油膜力;X,、x分别为为sy方向位移;丫8、丫9分别为左右轴承基础松动位移分别为偏心

距、外激励频率和时间。3计算实例系统采用参数:观i=4 kg,m2 =25 kg,m3 =20 kg,m4 = 10 kg,m5 =20 kg,m6 =25 kg,m7 =4 kg;m8 =

50 kg,m9 = 50 kg;ci =c7 =800 N • s/m,c2 =c6 = 1 250 N • s/m,c3 = c5 = 1 050 N • s/m,C4 = 850 N • s/m;

ki — 2. 0 X 107 N/m,怠i = 7. 5 X 107 N/m, ks2 = 2. 5 X 108 N/m; csl = 350 N/m, cs2 = 500 N/m; B3

0, 01 mm,_B” = 0(九工3) ;& =0. 22 mm;轴承间隙 c = 0. 11 mm;润滑油黏度戶=0. 018 Pa • so初始条件:工4=0. 5,j/4 = 0. 5,)8 =$9 =0,乜GH4),积分步长为兀/256,速度为^ = ^ = 0. 001,

『8 =$9 =0. 001 分别为Z、y方向量刚一的位移,)8、旳分别为左右轴承量刚一的基础松动位移。系统前3阶模态能量分布见图2。由图2可知,转速为300-1 800 rad/s时,系统能量转迁明显;当转

速为400 rad/s时,总能量达到最大,占原始系统总能量的97.6%,得到最优简化模型。当转速为400 rad/s时,卩的时间历程曲线见图3。由图3可知,当量纲一的时间rG [0,320]时,系统

处于瞬态过程;当量刚一的时间r大于320时,系统处于稳态过程。图3转速为400 rad/s时系统y的时间历程曲线Fig. 2 The first three order mode energy distribution

diagram of systemFig. 3 Time history of yi when the rotating speed is400 rad/s由于系统前3阶模态能量占原始系统主要能量,采用改进SVD方法,将自由度为16的转子系统模型 简化成自由度为3的模型,对比降维前后转子系统模型分岔特性与均方差幅值特性,验证方法的有效性。降维前后原始系统与简化系统的分岔特性和均方差幅值曲线见图4和图5。由图4可以看出,简化 系统模型保留原始系统的主要分岔特性,验证改进SVD方法的有效性。由图5可以看出,降维前后均方 差幅值在分岔点位置有一定的误差,降维后整体的幅值大小、均方差幅值走势与原始系统的吻合良好。降

• 122 •第2期唐莉:改进奇异值分解方法在高维动力学系统降维中的应用维后简化系统模型的均方差幅值曲线保留原始系统的均方差幅值特性,再次验证改进SVD方法的有

效性。1. 0 ---------------------------------------------------------------1

1. o0. 50. 50-0. 50-0. 5-1. 0—1 5-1. 0-'•50500

1 000 w/(rad-s_1)(a)原始系统1 500 2 000 0500

1 000 w/(rad-s_1)(b)简化系统1 5002 000图4计算实例原始系统与简化系统的分岔特性Fig. 4 Bifurcation diagram of the original system and the reduced system in calculation example1. 51. 51. 01. 00. 50. 50500

1 000 w/(rad-s_1)(a)原始系统1 5002 0000500

1 000 w/(rad-s_1)(b)简化系统1 500 2 000图5计算实例原始系统与简化系统的均方差幅值曲线Fig. 5 Mean square error amplitude curves diagram of the original system and the reduced system in calculation example4结论(1) 考虑动力学系统的瞬态响应特性,提出适用于高维非线性转子系统模型的改进奇异值分解方法; 应用牛顿第二定律,建立自由度为16的两端带有滑动轴承支承和发生基础松动故障的转子系统模型。(2) 根据转子系统的前3阶模态能量曲线,判断转速在400 rad/s时,前3阶模态能量占原始系统主要 能量;应用改进SVD方法,将自由度为16的转子系统模型降维为自由度为3的简化系统模型。(3) 简化系统模型能够保留原始系统的分岔及均方差幅值特性,验证改进奇异值分解方法的有效性,

对高维系统降维领域具有推动作用。参考文献(References)::1]陈予恕.非线性振动:M],北京:高等教育出版社,2002:1-10.CHEN Yushu. Nonlinear Vibration [M]. Beijing: Higher Education Press, 2002:1-10.:2]陈予恕.非线性振动系统的分岔和混沌理论:M].北京:高等教育出版社,1993.208-223.CHEN Yushu. Bifurcation and chaos theory of nonlinear vibration, systems [M]. Beijing: Higher Education Press, 1993:208-223.[3] GUO Shangjiang, YAN Shuling. Hopf bifurcation in a diffusive Lotka-Volterra type system with nonlocal delay effect [J]. Journal of

Differential Equations9 2016,260 (1) :781-817.L4]张家忠,陈丽莺,梅冠华,等.基于时滞惯性流形的浅拱动力屈曲研究[J].振动与冲击,2009,28(6):100-104.ZHANG Jiazhong, CHEN Liying, MEI Guanhua, et al. Dynamic bucking analysis of shallow parabolic arch based on the method of

inertial manifolds with time delay [J]・ Journal of Vibration and Shock, 2009,28(6) : 100-104.・123・东北 石油大 学学报第43卷 2019年[5] MARTINE M. Approximate inertial manifolds for reaction-diffusion equations in high space dimension [J]. Journal of Dynamics and

Differential Equantions, 19,1 (3) :245-267.[6] LU Kuan, JIN Yulin, CHEN Yushu, et al. Stability analysis of reduced rotor pedestal looseness fault model [J]. Nonlinear Dynam­

ics, 2015,82(4):1611-1622.[7] REGAG, TROGER H. Dimension reduction of dynamical systems: methods, models, applications EJJ. Nonlinear Dynamics, 2005 ,

41(1):1-15.E8]张永强,易亮.基于奇异值分解的航空发动机转子碰摩故障特征提取方法口].应用力学学报,2019,36(2):260-266.ZHANG Yongqiang, YI Liang. Extraction method of rub-impact fault behaviors of aero-engin^s rotor based on singular value decom­position [J]. Journal o£ Applied Mechanics, 2019,36(2) :260-266.[9] YU Hai, CHEN Yushu, CAO Qingjie. Bifurcation analysis for nonlinear multi-degree-o£-£reedom rotor system with liquid-film lubri­

cated bearings [J]. Applied Mathematics and Mechanics, 2013,34(6) :777-790.[10] LU Kuan, JIN Yulin, CHEN Yushu, et al. Review for order reduction based on proper orthogonal decomposition and outlooks of ap­

plications in mechanical systems [J]. Mechanical Systems and Signal Processing, 2019,123 :2-297.[11] MA Hui, ZENG Jin, FENG Ranjiao, et al. Review on dynamics o£ cracked gear systems [J]. Engineering Failure Analysis, 2015,

55:224-245.[12] ZENG Jin, CHEN Kangkang, MA Hui, et al. Vibration response analysis o£ a cracked rotating compressor blade during run-up

process [J]. Mechanical Systems and Signal Processing, 2019,118:568-583.[13] MA Hui, ZHAO Xueyan, TENG Yunnan, et al. Analysis of dynamic characteristics for a rotor system with pedestal looseness [J].

Shock and Vibration, 2011,18(1/2) :13-27.[14] 陈果.含不平衡一碰摩一基础松动耦合故障的转子一滚动轴承系统非线性动力响应分析[J].振动与冲击,2008,27(9):100-104.

CHEN Guo. Nonlinear dynamic response analysis of rotor-ball bearing system including unbalance-looseness coupled faults [J]. Jour­

nal of Vibration and Shock, 2008,27(9) : 100-104.[15] 张尚斌•列车轴承轨边声学诊断中故障声谱识别的时变阵列分析技术研究[D].合肥:中国科学技术大学,2017:67-87.ZHANG Shangbin. Research on time-varying array analysis of fault spectrum identification in trackside acoustic diagnosis o£ train

bearing [D]. Hefei: University of Science and Technology of China, 2017:67-87.[16] LU Kuan, YU Hai, CHEN Yushu, et al. A modified nonlinear POD method for order reduction based on transient time series [J].

Nonlinear Dynamics, 2015,79(2) : 1195-1206*[17] LU Kuan, CHEN Yushu» JIN Yulin, et al. Application of the transient proper orthogonal decomposition method for order reduction

of rotor systems with faults [J]. Nonlinear Dynamics, 2016,86(3) : 1913-1926.[18] LU Kuan, CHEN Yushu, CAO Qingjie, et al. Bifurcation analysis of reduced rotor model based on nonlinear transient POD method

[J]. International Journal of Non-Linear Mechanics» 2017, :83-92.[19] JIN Yulin, LU Kuan, HOU Lei, et al. An adaptive proper orthogonal decomposition method for model order reduction of multi-disc

rotor system [J]. Journal of Sound and Vibration, 2017,411 :210-231.[20] ADILETTA G, GUIDO A R, ROSSI C. Chaotic motions of a rigid rotor in short journal bearings [J]. Nonlinear Dynamics, 1996,

10(3):251-269.(上接第58页)Chang8 members tight sandstone reservoirs in the southwestern part of Ordos Basin, China [J]. Natural Gas Geoscience, 2017,

28(6):909-919.[33] 陈能贵,郭沫贞,孟祥超.准鳴尔盆地西北缘中二叠统一下三叠统砂砾岩孔隙结构类型及其控制因素口].石油地质,2016,

37(4):401-408.CHEN Nenggui, GUO Mozhen, MENG Xiangchao. Pore structures and controlling factors of Middle Permian-Lower Triassic sandy

conglomerates in northwestern margin of Junggar Basin [J]. Xinjiang Petroleum Geology, 2016,37(4) :401-408.[34] 孙乐,王志章,于兴河,等.高精度CT成像技术在致密油储层孔隙结构研究中的应用:以准噹尔盆地玛湖西斜坡风城组为例[J].东北

石油大学学报,2016,40(6):26-34.SUN Le, WANG Zhizhang, YU Xinghe, et al. Application of high resolution CT image technology to the research of pore structure of tight oil reservoir: a case study of the Fengcheng Formation on the Mahu West Slope in Junggar Basin [J]. Journal of Northeast

Petroleum University, 2016,40(6) :26-34.・124・AbstractsJournal of Northeast Petroleum UniversityVol. 43 No. 2 Apr. 2019211167 9 China ; 2. School of Civil and Architecture Engineering , Northeast Petroleum University 9

Daqing, Heilongjiang 163318, China )Abstract: Based on the plate-beam theory, the total potential energy equation of flexural-torsional buck­

ling of simply supported beam with doubly symmetric sections and lateral torsional bracing at mid-span

under uniform loading is established・ The modal trial functions of the lateral displacement and torsional

angle are expressed by the first six terms of the trigonometric function. By introducing dimensionless pa­

rameters and according to the principle of potential energy standing, the analytical solution of the dimen­

sionless critical moment of flexural-torsional buckling of simply supported beam is obtained. The IstOpt mathematical optimization analysis software was used to regress the dimensionless critical moment of the simply supported beam. The dimensionless formula is used to compare with the ANSYS finite element

solution and the results show that the theoretical solution of the dimensionless formula agrees well with

the finite element solution and the error is within 士5 % , which can provide reference for engineering de- sign.Key words: simply supported beam; flexural-torsional buckling; plate-beam theory; torsional bracing;

energy variational method ; uniform loading ; finite element methodApplication of improved SVD method for dimension reduction in high-dimensional dynamic system/2019,

43(2):119-124TANG Li( School of Teacher Education , Daqing Normal University , Daqing, Heilongjiang 163712 , China )

Abstract: Most actual engineering systems are high-dimensional and nonlinear, the qualitative analysis is

very difficult, order reduction should be applied for the systems・ The modified singular value decompo-

sition(SVD) method is proposed based on the basic theory of SVD method and transient response behav­ior of high-dimensional dynamical system. A 16 degrees-of-freedom(DOF) rotor system model with ped­estal looseness and supported by a pair of sliding bearings is established with Newton's second law・ The

SVD method is used to reduce the original system to 3 DOF reduced model, the dynamical behaviors of

the original and reduced systems are compared. The results indicate that the reduced system model re­

serves the main bifurcation and mean square error amplitude behavior of the original system, which veri­fies the efficiency of the modified SVD method. The proposed dimension reduction method can provide

theoretical guidance to qualitative analysis of high-dimensional nonlinear dynamic systems・Key words: improved SVD method; dynamical system; rotor; dimension reduction; bifurcation; mean

square error amplitude・1・

因篇幅问题不能全部显示,请点此查看更多更全内容