外文翻譯---旋轉機械軸心軌跡識別方法研究

1、<p><b>　　附錄</b></p><p>　　Research of identification of shaft orbit for rotating machine</p><p>　　XIAO Sheng-guang</p><p>　　（Test Center of Chongqing University，Chon

2、gqing 400044，China）</p><p>　　Abstract：A novel approach for the identification of shaft orbit is presented. The vibration displacement signalsacquired in two mutually vertical directions were treated through

3、noise suppression and fitted to form a shaft orbit.Then the direction changing character was extracted and all shaft orbits were classified and identified with thefunction discriminated method according to the pattern re

4、cognition theory. Each type of shaft orbit was described indetail with one character，which can help to</p><p>　　Key words：Shaft orbits；Fault diagnosis；Geometric features；Pattern recognition；Thinning classifi

5、cation</p><p>　　1 The introduction</p><p>　　With the development of science and technology and modern industry, to rotating machines the large-scale, high-speed and automation direction, the sha

6、pe of rotating machinery state monitoring and fault diagnosis is put forward higher request, the axis trajectory for rotating machinery is an important state characteristic parameters, can be simple and straight view, vi

7、vidly reflect the running status of equipment. Through to the axis of track observation, can determine some of the common faults, </p><p>　　Axis path at present, already have several identification methods,

8、including [1-2] invariant moment method, a two-dimensional image gray level matrix [3]. literature [1-2] axis path with seven moment invariants as feature vectors, recognition by the distance between the characteristics

9、of axial trajectory shape, literature [3] the axis trajectory image coding, using neural network for identification. Both methods can better identify axis path, but the method is complex, relatively large amount o</p&

10、gt;<p>　　2 Axis locus corresponding fault mechanism analysis</p><p>　　Axis path refers to the axis on a bit relative to the trajectory of the bearing, the trajectory is in a plane perpendicular to the

11、 axis, so it requires the setting sensors in both directions in the plane. Axis path clearly describes the fault characteristics of implication in the unit, the axis trajectory can get in on the rotor bending, imbalance,

12、 instability and dynamic-static friction bearing and other information. Through the actual operation of rotating machinery fault mechanism analysis an</p><p>　　3 Image processing axis path recognition princi

13、ple [4]</p><p>　　In image recognition, is the simplest method of identification for template matching. Is the unknown image compared to a standard image, see whether they are the same or similar.</p>

14、<p>　　Has M category: 1 omega, omega 1,... , M each type of feature vector by a number of omega said, such as class I class, omega has:</p><p>　　Xi = [xi1, xi2, xi3,..., xin] T</p><p>　　For

15、 any identified trajectory image X:</p><p>　　X = [x1, x2, x3,..., xn] T</p><p>　　Calculate distance d (Xi, X), if there is one, I made:</p><p>　　d (Xi, X) < d (Xj, X), j = 1, 2,.

16、.., M, I indicates j ∈ X omega I.</p><p>　　Specific discriminant, X, Y distance two points can be used | X, Y | 2</p><p>　　Said, namely:</p><p>　　d (X, Xi) - Xi | = | X 2 = (X - Xi)

17、 T = (X - Xi)</p><p>　　XTX XTXT - XiTXi =XTX - (XTXT + XiTX - XiTXi)</p><p>　　Type of XTXT + XiTX - characterized XiTXi linear function, can be used as discriminant function:</p><p&g

18、t;　　di (X) = XTXT + XiTX – XiTXi.</p><p>　　If d (X, Xi) = min {di (X)}, then X ∈ omega I. This is the kind of problem, the minimum distance criterion. In this paper, the axis path identification in this way.

19、</p><p>　　4 Axis trajectory image feature extraction and recognition</p><p>　　Axis path can be used to identify the image feature has a lot of, now use more features are: invariant moment[1], th

20、e cross points, circle number, center of mass position, curvature, length, etc. Based on the direction of the axis trajectory change as the main characteristics, and some other features are realized tracing above.</p&

21、gt;<p>　　4.1 Axis trajectory image preprocessing [5]</p><p>　　Acquisition of two way data before the synthesis has been underway for filtering de-noising treatment, eliminate a lot of burr. </p>

22、<p>　　Figure1 The conditions of deleting</p><p>　　As shown in figure 1, axis path line is at an Angle, was on the way to draw black spots position should be in the path, but considering that in order

23、not to make the direction changing, change to figure this is on the corner points, should be deleted (corresponding to the four kinds of situations), delete the conditions are:</p><p>　　| x [I + 2] - [I] x |

24、 | = 1 and y [I + 2] - [I] y | = 1</p><p>　　If meet the above conditions, the delete (x + 1], [I, y [I + 1].</p><p>　　4.2 Feature extraction and quantification of [5]</p><p>　　To qu

25、antify characteristics, specifies the following four directions: to the right, down, left, up (in the program can be expressed in Numbers or corresponding bits, this paper use Numbers 1, 2, 3, 4), contains the scope of t

26、he direction as shown in figure 2.</p><p>　　Figure2 Stability in the direction of the range</p><p>　　Was three scope are included in each direction, is to avoid a small perturbation to change di

27、rection, you can see from the above four, four direction on the diagonal lines, each containing in two directions, to determine the direction of, have the following rules:</p><p>　　(1) for each starting poin

28、t, when the shaft rotates clockwise, to choose direction priority sequence is to the right, down, left, up, and the corresponding number is 1, 2, 3, 4; When the shaft rotates counterclockwise, to choose direction priorit

29、y sequence is to the left, down, right, up, the corresponding number is 1, 2, 3, 4. Such axis path is to work in the same state, which is formed by the different direction of rotation of the characteristic value.</p&g

30、t;<p>　　(2) if you have in one direction, so in one direction, then should keep and original in the same direction as far as possible, so that the direction fluctuation in a small scope, can be aligned, unless hav

31、e jumped from the direction of scope, which is to avoid the characteristics of the two adjacent to the opposite direction.</p><p>　　After got the direction sequence, to assist in the description, also can ca

32、lculate some feature such as number of intersection point, end point, the distance to the intersection first point from the intersection point of distance, etc., these features also use numerals, this paper selects the n

33、ode number to describe.</p><p>　　4.3 The classification of axis path description</p><p>　　Using the above methods can be classified on the axis trajectory graphics recognition, but belong to the

34、 same kind of classification of the two graphics, shape may also have very big difference. In order to understand the severity of the failure, and other characteristics to measure the size of the track deformation.</p

35、><p>　　4.3.1 Unbalanced fault</p><p>　　Axis trajectory for the oval, graphic long axis and short axis L L, the ratio of their C = L/L is fine length [6], C axis path can represent the size of the d

36、eformation degree. Due to the direction of the circle and ellipse feature vector is the same, in C can also be used to distinguish whether there is a fault. 0 C or less or less1, C is smaller, the elliptical deformation

37、degree, the greater the failure, C = 1 indicates no fault.</p><p>　　Figure3 Length of the thin</p><p>　　4.3.2 Imbalance and comprehensive fault in the wrong</p><p>　　Axis trajectory

38、 graphics for banana fan, its deformation characteristics can be expressed in its bending degree. To take the first axis trajectory of the center of mass. According to the physical concept of center of gravity, define th

39、e two-dimensional gray-scale image centroid is as follows:</p><p>　　Find two corner point axis path, become card axis of connections between them with the center of mass. , finding the Angle between the two

40、card axis AArg define AArg for bending. 0 or less AArg PI or less, the smaller AArg, said graphics completely, the greater the degree of the failure is more serious.</p><p>　　4.3.3 Misalignment and oil film

41、vortex breakdown</p><p>　　Axis trajectory is figure 8 and figure 8, respectively the distinction of the two tracks is have a intersection point. Find trajectory intersection to intersection point as segmenta

42、tion point, the original sequence is divided into two parts. Respectively in the area of the two parts of S1, S2. The area ratio of two ring is:</p><p><b>　　C1=</b></p><p>　　Where 0

43、< C1 is 1 or less, the size of C1 unstable factors in the reaction the rotation axis of strong or weak, C1, said the greater the role played by the unstable factors.</p><p>　　5 The simulation research<

44、/p><p>　　For each categories of axis path, select a representative which can identify four kinds of computing. The result is shown in figure 4.</p><p>　　Figure4 The axis trajectory simulation</p

45、><p>　　Axis of the calculation result shows that each categories of trajectory eigenvalues were extracted by different, use criterion can easily draw categories to which they belong, to judge fault in rotating

46、machinery. By detailed description of parameters of the calculation result shows that belong to the same categories of axis trajectory, the shape also has the very big difference, refinement parameters can well said this

47、 kind of difference, help us to judge the severity of the fault.</p><p>　　6 conclusion</p><p>　　Axis path based on a number of engaged in automatic identification research results, the scholars

48、in the direction of the direction of quantitative change characteristics, combined with the other characteristics, to build into a template, then use the theory of pattern recognition to identify, for the axis trajectory

49、 automatic identification provides a new method.</p><p>　　References</p><p>　　1 Thousands of xiuzhou district, Li Yonggang Li Heming. Based on moment invariant features and the new automatic axi

50、s trajectory shape correlation recognition [J]. Journal of engineering for thermal energy and power, 2005, 20 (3) : 239-241.</p><p>　　2 NiChuanKun Zhou Jianzhong, FuBo. Based on the improved moment invariant

51、 generator axis trajectory recognition [J]. Electric power science and engineering, 2004 (3) : 16-19.</p><p>　　3 Professor. Axis locus and automatic recognition for the purification of research [J]. Journal

52、of wuhan university of technology, transportation science and engineering edition, 2003, 27 (6) : 878-881.</p><p>　　4 Yang Shuying. Image pattern recognition [M]. Beijing: tsinghua university press, 2005.<

53、;/p><p>　　5 Zhang Honglin. Visual c + + digital image pattern recognition technology and engineering practice [M]. Beijing: people's posts and telecommunications press, 2003.</p><p>　　6 Jiang Z

54、hinong Li Yanni. Rotating machinery axis trajectory feature extraction technology research [J]. Journal of vibration and the test and diagnosis, 2007, 27 (2) : 98-102.</p><p>　　旋轉機械軸心軌跡識別方法研究</p><

55、p><b>　　肖圣光</b></p><p>　　(重慶大學測試中心，重慶400044)</p><p>　　摘要：提出了一種識別軸心軌跡的新方法。采集方向相互垂直的兩路振動位移信號，經消噪處理后擬合為軸心軌跡，提取軸心軌跡的方向變化特征，利用模式識別理論中的函數判別法進行分類識別。并對每種類別的軸心軌跡，用一個特征參量來進行細化描述，不僅可以判斷機械的運行狀

56、態(tài)，在發(fā)生故障的時候還能對故障嚴重程度進行評估。通過對仿真分析，取得了良好效果。</p><p>　　關鍵詞：軸心軌跡；故障診斷；幾何特征；模式識別；細化分類</p><p><b>　　1 引言</b></p><p>　　隨著科學技術和現代工業(yè)的發(fā)展，旋轉機械向著大型、高速和自動化方向發(fā)展，這對旋轉機械狀態(tài)監(jiān)測和故障診斷提出了更高的要求，軸

57、心軌跡作為旋轉機械的一個重要的狀態(tài)特征參量，能簡單、直觀、形象地反映設備的運行狀況。通過對軸心軌跡的觀察，可以判斷出一些常見的故障，例如油膜渦動、油膜振蕩、軸不對中等。傳統的軸心軌跡形狀和動態(tài)特性的識別是基于人機對話模式實現的，嚴重影響了故障診斷的智能化水平。為了提高故障診斷的智能化程度，需要深入研究旋轉機械的軸心軌跡自動識別技術。</p><p>　　目前，已經有了幾種軸心軌跡識別方法，其中包括不變矩法[1-2

58、]，二維圖像灰度矩陣[3]。文獻[1-2]用軸心軌跡的7 個不變矩作為特征向量，通過特征量之間的距離來識別軸心軌跡形狀，文獻[3]將軸心軌跡圖象進行編碼，利用神經網絡進行識別。這兩種方法都能較好的識別軸心軌跡，但是方法復雜，計算量比較大。在總結前人工作的基礎上，針對軸心軌跡本身的變化特點，提出了一種新的識別方法，通過提取軸心軌跡一個周期的方向變化特征來進行分類識別，并對每種類別的軸心軌跡，提出一種能，來細化描述其變形程度的參量，進一步了

59、解故障的嚴重程度，而且特征提取速度快，效率高。</p><p>　　2 軸心軌跡對應的軸承故障機理分析</p><p>　　軸心軌跡是指軸心上一點相對于軸承座的運動軌跡，這一軌跡是在與軸線垂直的平面內，因此它要求在該平面內兩個方向上設置傳感器。軸心軌跡清晰地描述了蘊涵在機組內的故障特征，軸心軌跡中可以獲取有關轉子彎曲、不平衡、軸瓦失穩(wěn)和動靜摩擦等信息。通過對實際運行的旋轉機械故障機理的分析

60、和大量理論分析，人們總結出幾種軸心軌跡所對應的故障集。</p><p>　　實際采樣的信號并不是一個整周期的，所以需要將其按照最大周期分量對采樣數據進行截取，取得一個整周期的封閉曲線。將采集到的信號進行提純，合成后，存儲到一個表示x，y 坐標的坐標序列中：{x（n），y（n）：n=0，1，…，N-1}，通過分析這個序列中x、y 變化的特征來識別軸心軌跡。</p><p>　　3 圖像處理識

61、別軸心軌跡的原理[4]</p><p>　　在圖像識別中，最簡單的識別方法就是模板匹配。就是把未知圖像和一個標準圖像相比，看它們是否相同或相似。</p><p>　　設有M 個類別：ω1，ω1，…，ωM 每類特征由若干個向量表示，如類ωi 類，有：</p><p>　　Xi=[xi1，xi2，xi3，…，xin]T</p><p>　　對于任

62、意被識別的軌跡圖像X：</p><p>　　X = [x1, x2, x3,..., xn] T</p><p>　　計算距離d（Xi，X），若存在某一個i，使：</p><p>　　d（Xi，X）＜d（Xj，X），j=1，2，…，M，i≠j （3）</p><p><b>　　則X∈ωi。</b></p>

63、<p>　　具體判別的時候，X，Y 兩點距離可以用|X，Y|2表示，即：</p><p>　　d (X, Xi)= Xi- X 2 = (X - Xi) T (X - Xi)</p><p>　　XTX-XTXT - XiTXi =XTX - (XTXT + XiTX - XiTXi)</p><p>　　式中的XTXT+XiTX-XiTXi 為特征的

64、線性函數，可作為判別函數：</p><p>　　di(X)=XTXT+XiTX-XiTXi</p><p>　　若d(X，Xi)=min{di(X)}，則X∈ωi。這就是多類問題的最小距離判別法。本文就用這種方法識別軸心軌跡。</p><p>　　4 軸心軌跡圖像特征的提取和識別</p><p>　　軸心軌跡圖像特征的提取和識別可以用來識別軸

65、心軌跡圖像的特征有很多，目前利用較多的特征有：不變矩[1]，交叉點數，圓環(huán)數，質心位置，彎曲度，細長度等。本文以軸心軌跡的方向變化為主要特征，并用一些其他特征進行細化描述。</p><p>　　4.1 軸心軌跡圖像的預處理[5]</p><p>　　采集的兩路數據在合成前已經進行了濾波消噪處理，消除掉了很多毛刺。但是為了后面特征提取的方便以及減少數據量，還需要做一些預處理。</p&g

66、t;<p><b>　　圖1 刪除條件</b></p><p>　　如圖1，斜著的直線是軸心軌跡，本來途中畫黑點的位置都應該在路徑里的，但考慮到為了不使方向變來變去，對于改圖這種處于拐角上的點，都要刪除掉（相對應的有4 種情況），刪除的條件是：|x[I+2]-x[I]|=1 且|y[I+2]-y[I]|=1。 如果滿足以上條件，則刪除（x[I+1]，y[I+1]）點。</

67、p><p>　　4.2 特征的提取與量化</p><p>　　為了量化特征，規(guī)定了如下4 個方向：向右，向下，向左，向上(在程序中可以用數字或相應的比特位表示，本文用數字1，2，3，4 來表示)，各方向包含的范圍見圖2。</p><p>　　圖2 穩(wěn)定的方向范圍</p><p>　　之所以每個方向都包含3 個范圍，是為了避免一些小的擾動改變方向，

68、從上面4 個圖中可以看到，在斜線上的4 個方向，每一個都包含在兩個方向中，對于方向的確定，有如下規(guī)則：</p><p>　?。?）對于每一個起點，當軸順時針旋轉時，選擇方向的優(yōu)先順序依次是向右、向下，向左，向上，對應的數字是1，2，3，4；當軸逆時針旋轉時，選擇方向的優(yōu)先順序依次是向左，向下，向右，向上，對應數字是1，2，3，4。這樣是為了軸心軌跡在同樣的工作狀態(tài)，不同的旋轉方向所形成的特征值一樣。</p&

69、gt;<p>　　（2）如果已經處在一個方向，那么對于接著的一個方向，應盡量保持和原來的方向一致，這樣方向在一個小范圍內波動，可以保持一致，除非已經跳離了這個方向所在的范圍，也就是盡量避免兩個相鄰的特征為相反方向。</p><p>　　在得到了方向序列后，還可以計算一些特征量來輔助描述，如交點個數、尾點到交點的距離、首點距交點的距離等，這些特征也用數字來表示，本文選用交點個數來描述。</p&g

70、t;<p>　　4.3 軸心軌跡的分類描述</p><p>　　用上面的方法可以對軸心軌跡圖形進行分類識別，但是屬于同一種分類的兩個圖形，形狀也可能有很大的差別。為了清楚的了解故障的嚴重程度，還要用其他特征量來衡量軌跡變形量的大小。</p><p>　　4.3.1 不平衡故障</p><p>　　軸心軌跡為橢圓形，求得圖形的長軸L 和短軸l，它們的比值

71、C=l/L 為細長度[6]，用C 可以表示軸心軌跡變形程度的大小。由于圓和橢圓的特征向量相同，用C 還可以用來判別是否有故障。0≤C≤1，C 越小，橢圓變形程度越大，故障越嚴重，C=1 時表示沒有故障。</p><p><b>　　圖3 細長度</b></p><p>　　4.3.2 不平衡與不對中綜合故障</p><p>　　曲程度來表示。先

72、求出軸心軌跡的質心。根據物理上重心的概念，定義二維灰度圖像的質心如下：</p><p>　　找到軸心軌跡的兩個拐角點，它們與質心之間的連線成為卡軸。求得兩個卡軸之間的夾角AArg，定義AArg 為彎曲度。0≤AArg≤pi，AArg 越小，表示圖形的完全程度越大，故障越嚴重。</p><p>　　4.4.3 不對中和油膜渦動故障</p><p>　　軸心軌跡分別為外

73、8 字型和內8 字形，這兩種軌跡的特點就是有一個交點。找到軌跡的交點，以交點為分割點，將原序列分成兩部分。分別求處兩部分的面積S1，S2。兩個環(huán)形的面積比:</p><p><b>　　C1=</b></p><p>　　其中0＜C1≤1，C1 的大小反應了軸系旋轉中不穩(wěn)定因素的強勢或弱勢，C1 越大，表示不穩(wěn)定因素所起作用越大。</p><p&g

74、t;<b>　　5 仿真研究</b></p><p>　　對每種類別的軸心軌跡，選取有代表性的4 種進行識別計算。得到的結果如圖4。</p><p><b>　　圖4 軸心軌跡仿真</b></p><p>　　由計算結果可知，每種類別的軸心軌跡所提取出來的特征值都不同，用判別法很容易就可以得出它們所屬的類別，從而判斷旋轉機

75、械所發(fā)生的故障。由細化描述參量的計算結果可知，屬于同種類別的軸心軌跡，其形狀也有很大的差別，細化參量可以很好的表示這種差別，幫助我們判斷故障的嚴重程度。</p><p><b>　　6 結束語</b></p><p>　　向的學者的成果，將方向變化特征量化，結合其它特征，構建成模板，然后用模式識別的理論進行識別，為軸心軌跡的自動識別提供了新的方法。</p>

76、<p><b>　　參考文獻</b></p><p>　　[1] 萬書亭，李永剛，李和明.基于不變矩特征和新型關聯度的軸心軌跡形狀自動識別[J].熱能動力工程，2005，20（3）：239-241.</p><p>　　[2] 倪傳坤，周建中，付波.基于改進不變矩的發(fā)電機組軸心軌跡識別[J].電力科學與工程，2004，16（3）：16-19.</p

77、><p>　　[3] 陳豫.軸心軌跡提純與自動識別的研究[J].武漢理工大學學報：交通科學與工程版，2003，27（6）：878-881.</p><p>　　[4] 楊淑瑩.圖像模式識別[M].北京：清華大學出版社，2005.</p><p>　　[5] 張宏林.Visual C++數字圖像模式識別技術及工程實踐[M].北京：人民郵電出版社，2003.</p&g