購買設(shè)計請充值后下載,,資源目錄下的文件所見即所得,都可以點開預覽,,資料完整,充值下載可得到資源目錄里的所有文件。。?!咀ⅰ浚篸wg后綴為CAD圖紙,doc,docx為WORD文檔,原稿無水印,可編輯。。。具體請見文件預覽,有不明白之處,可咨詢QQ:12401814
Evaluation of Honed Cylinder Bores
F.Puente Leon
Design of Systems on Silicon(DS2),Parque Tecnologico de Valencia,
C./Charles Robert Darwin 2,E-46980 Paterna(Valencia),Spain
Submitted by G..Spur(1),Berlin,Germany
Abstract
The quality of the honing texture on cylinder bores of combustion engines plays an important role with respect to oil consumption,noxious emissions,and running performance.To evaluate honed surfaces objectively,features describing the surface texture are extracted from 2-D data of the surface.The paper focuses on two crucial stages of the data analysis:the preprocessing,which aims at suppressing irrele-vant components and enhancing the information of interest,and the feature extraction,which yieldsreliable numerical estimates of the surface characteristics of interest,like the honing angle,groove pa-rameters,surface defects etc.The assessment results can easily be adapted to user-specific ratings.
Keywords:Honing,Surface texture,Automated visual inspection
1 INTRODUCTION
Cylinder bores of combustion engines are finished by honing.The resulting surface texture mainly consists of two bands of helical grooves placed stochastically and appearing at different angles to the cylinder axis.The texture quality is highly important for dry operation properties,oil consumption,noxious emissions,and running performance.Up to now,experts are still rating honed surfaces visually based on microscopic images.This method is tedious,subjective,and time consuming.To get objective and reproducible results,an automated method of inspection is necessary.
2 INSPECTION APPROACH
2.1 Surface data
There are basically different ways to measure the texture of a honed surface;see Table 1.Typically,a mechanical stylus only performs a I-D measurement of the surface profile.In contrast to this,grey level images and optical profilometers provide 2-D data in a reasonable amount of time.
Because the lateral-geometric features of honing textures can only be analysed with 2-D data,in the following we will concentrate on such kinds of data.Other characteris-tics related to the different measurement principles investigated are also included in this table.
A signal model describing the essential characteristics of a honing texture constitutes the basis of the evaluation approach presented in this paper.Based on this model,clear and mathematically well-defined features are introduced,which enable a reproducible and objective assessment of the texture.This strategy differs from many popular methods-such as those relying on neural
networks-,which are often treated as a'black box'[I].The features chosen are inspired by the Honing Atlas[2],by many opinions of experts,and have also been ex-tended by adding new volumetric parameters for the case of analysing profile data.This results in an extensive set of features that can be customized to match the needs of individual users.
2.2 Properties of honing textures
Figure 1 shows some of the properties of honing tex-tures,based upon which features are to be defined.The most popular ones are the roughness parameters,such as those based on the Bearing Ratio Curve(Abbott Curve)[3],and R,,R,and R,,[4].However,dealing with honed surfaces,it is important to define features that take the lateral geometry into account.This way,most relevant texture peculiarities can be described,such as the honing angle,material smearings,groove interrupts, stray grooves,holes,foreign bodies,and flakes,as shown in Figure 1.In addition,features describing the balance of grooves,presence of plateaus,shape of grooves,cracks,residual turning grooves,and chatter marks are also needed.
2.3 Automated inspection
Figure 2 shows an overview of the abilities and aims of automated inspection in quality control applied to the honing process.A 2-D or 3-D sensor provides data g(x) of the honed surface,where x=(~,yE )R~2 denotes the lateral spatial coordinates.The grey coloured blocks of the diagram are part of the sensor data processing.The outputs of the system can be used simply as a statement
about surface quality,to give alarms causing an interrupt of the machining process,or it can be fed back via a controller to regulate the honing process,because the honing texture contains information about both functional-ity and also machining procesdndependently of the fact whether a post-honing brushing is performed or not.In the following sections,we will focus on two crucial
stages of the automated inspection:the preprocessing of the sensor data and the feature extraction,and we will give some examples to these steps.
3 PREPROCESSING
The goal of the preprocessing is to suppress irrelevant components,namely the inhomogeneities i(x)and the disturbances b(x),while enhancing the information of interest,i.e.the texture t(x).In the case of acquisition of image data,the inhomogeneities i(x)could be due to
Table 1:Comparison between mechanical stylus devices,grey level images,and optical profilometers grey iron cylinder
Figure 2:Automated inspection of honed surfaces.
Figure 1:Honing textures showing lateral features and defects:(a)material smearings,groove interrupts; (b)stray grooves;(c)holes or foreign bodies;(d)flakes.
signal of
interest
inverse
transform separation transform
Components
Figure 3:Principle of the preprocessing.spatial variations of surface illumination.Other compo-nents assigned to the disturbances b(x)include e.g.deviations from the ideal course of the grooves and defects,such as material smearings,flakes etc.We use a signal model that describes the sensor data g(x)as a combination of the texture t(x)and the irrele-vant components i(x)and b(x):
To be able to recover the information of interest t(x),an assumption is necessary:the different components have to be mathematically distinguishable.
As shown in Figure 3,a transform maps the raw data g(x)such that a strict separation of their components is obtained.Then,the undesired components are sup-pressed,and finally an inverse transform is performed that yields the results of the preprocessing.
The benefits of this procedure include a simplification of the feature extraction,and a more robust image process-ing,as shown in the following examples.
3.1 Homogenization
When a groove texture is degraded by an intensity inhomogeneity i(x)due to the data acquisition process,e.g.due to an inhomogeneous lighting,a homogenization can be performed to suppress this unwanted component [6].Figure 4 shows an example of this operation for a planing texture.On the left side of the figure,the original texture is shown.The central image represents the result of a standard homogenization method-the homomorphic
Figure 4:Homogenization:(left)planning texture;(centre) homomorphic filtering;(right)homogenization result.
Figure 5:Texture decomposition:(left)honing texture;
(centre)groove texture;(right)background texture.
Figure 6:Reference surface:problems with conventional
low-pass filters.
filtering,which assumes a multiplicative combination of texture and inhomogeneity.Especially in the upper left corner,this image shows a very poor contrast.The image on the right results from the model-based approach according to Figure 3.In this case,a homogenization of the local mean value and the local contrast has been performed based on a model that considers a mixed additive and multiplicative combination of both signal of interest and disturbing inhomogeneity[6].The result is clearly more homogeneous than the former one and enables a more robust analysis of the texture.
3.2 Texture decomposition
The next example concerns the decomposition of the honing texture to ease the feature extraction.Due to the complexity of the honing texture,the extraction of rele-vant features needed for the inspection task could be simplified considerably,if the partial textures constituting the signal g(x)according to Eq.(1)were available.Thus, it would be advantageous to develop a method to sepa-rate the texture g(x)into a component t(x)containing the straight structures(i.e.the grooves)and another one b(x) showing the isotropic components(i.e.the background, including defects and objects).In this case,a homogene-ous texture will be assumed.
Fortunately,a very efficient algorithm to perform this separation already exists[7].The left side of Figure 5 shows an original honing texture;the other two images represent the results of the adaptive texture decomposi-tion computed with this algorithm.In the groove texture,only the ideal grooves can be seen,whereas the back-ground image contains all deviations from the ideal groove course as well as defects and other objects.For a more comprehensive discussion of the separation algo-rithm,interested readers are referred to[7]
Figure 7:Original honed surface and reference surface.
3.3 Reference surface
Finally,the definition of a reference surface to eliminate the shape component will be presented.The graph in Figure 6 represents a trace through the profile of a honed surface.The smooth line describes the shape component to be suppressed.However,conventional low-pass filters lead to distortions in the area of the grooves,as shown in the case of the dashed line.We have faced this problem by developing an iterative 2-D filter-a modified Gaus-sian filter-hich behaves robustly even in case of deep grooves[8].The 3-D plot depicted in Figure 7 shows a section of a honed surface as well as the resulting reference surface computed with this method.
4 FEATURE EXTRACTION
4.1 Honing angle
The first example of the feature extraction is the estima-tion of the honing angle.To this end,the periodogram (PG)is computed,which is proportional to the squared magnitude of the Fourier transform of the texture g(x):
The PG is an estimator of the power spectral density(PSD)function,which specifies the spectral properties of the stochastic process generating the texture[9].Then, the PG is projected radially;see Figure 8.
Since honing textures consist of two bands of grooves, the projection function also shows two pronounced maxima.The estimate of the honing angle results as the difference between the locations of both maxima:
Despite the variance of the PG,due to the averaging performed,the radial projection is a very smooth curve. Thus,this procedure yields a fast and statistically reliable estimate for the honing angle.
Figure 9:Illustration of the Radon transform.
Figure 10:Detection of defects:(a)groove image;
(b)Radon transform of(a);(c)multiplication of(b)and
(e);(d)background image;(e)Radon transform of(d);
(f)defective grooves detected.
Figure 11:Algorithm to detect defective grooves.
4.2 Groove parameters
The next example concerns the extraction of the groove parameters.This is accomplished based on the Radon transform,which maps each line of a 2-D image onto a point of the transformation domain;as demonstrated in Figure 9[lo].Following,all distinct peaks of the Radon transform,which correspond with grooves,are detected by means of morphological filters.Finally,for each
detected groove,the corresponding parameters(ampli-tude,width,location,and angle)are estimated based on the output of the morphological filters[9].
4.3 Detection of defects
In Section 3.2,an algorithm enabling a decomposition of honing textures has been presented.This section fo-cuses on the background texture obtained,which con-tains the main information concerning defects and objects,and discusses a robust approach allowing a detection of defects based on this image.It represents a refinement of the detection of grooves presented in the last subsection;see Figure 11[11].In this case,a Radon transform of the groove image Figure l0(a)obtained after decomposition is performed to concentrate the information concerning grooves;see Figure 10(b). Furthermore,collinear defects distributed along grooves are also concentrated by means of a Radon transform of the background image onto peaks in the Radon domain; see Figures 10(d)and(e).By combining both groove texture and background texture in the Radon domain multiplicatively,only the peaks representing defective grooves remain;see Figure l0(c).The most pronounced peaks in this image correspond with the locations of the three grooves sketched in Figure IO(f),which are indeed the most salient defective grooves of the original image;
see Figure 5(a).
5 SUMMARY AND CONCLUSIONS
This paper has shown how signal processing methods can be used to automatically evaluate relevant properties of the honing texture of grey iron cylinders with regard to different quality aspects.A preprocessing strategy has been presented that enables a robust automated as-sessment.Moreover,a feature-oriented approach has been proposed,in which the features are clear and mathematically well defined.By incorporating depth data, new function-relevant parameters can be computed. In previous approaches,only roughness parameters from first-order statistics have been used to quantify the features of interest.The presented strategy,however,is based on an analysis of the essential lateral-geometric characteristics of the texture,including those related to higher-order statistics.This enables to automate and objectify the assessment proposed by experts and standards used in different companies.
[I]Malburg,M.C.,Raja,J.,1993,Characterization of surface texture generated by plateau honing proc-ess,ClRP Annals,42/1:637840.
[2]AE Goetze GmbH,Burscheid,Germany,1993,AE Goetze Honing Guide-Rating Criteria for the Hon-ing of Cylinder Running Surfaces.
[3]DIN EN I S 0 13565-2,1996,Geometrical Product Specification(GPS)-Surface texture:Profile method;Surfaces having stratified functional prop-erties-Part 2:Height characterization using the
linear material ratio curve.
[4]DIN 4768,1990,Determination of roughness parameters R,,R,,,R,by means of stylus instru- ments;terminology;measuring conditions.
[5]Pfeifer,T.,Wiegers,L.,1998,Adaptive control for the optimized adjustment of imaging parameters for surface inspection using machine vision,ClRP An-nals,47/1:487490.
[6]Beyerer,J.,Puente Leon,F.,1997,Suppression of Inhomogeneities in Images of Textured Surfaces,Optical Engineering,36/1:85-93.
[7]Beyerer,J.,Puente Leon,F.,1998,Adaptive Separation of Random Lines and Background,Op-
tical Engineering,37/10:2733-2741.
[8]Krahe,D.,2000,Zerstorungsfreie Prufung der Tex-tur gehonter und geschliffener Gegenlaufflachen, VDI Verlag,Dusseldorf.
[9]Beyerer,J.,Krahe,D.,Puente Leon,F.,2001,Characterization of Cylinder Bores,In:Metrology and Properties of Engineering Surfaces,E.Main-sah,J.A.Greenwood,and D.G.Chetwynd(eds.),Kluwer Academic Publishers,Boston,MA.
[10]Deans,S.R.,1983,The Radon transform and some of its applications,John Wiley 8,Sons,New York.
[11] Beyerer,J.,Puente Leon,F.,1997,Detection of Defects in Groove Textures of Honed Surfaces,Int. J.of Machine Tools 8,Manufacture,37/3:371-389.
珩磨汔缸孔徑的評價
摘要
內(nèi)燃機汽缸孔的珩磨組織在潤滑油的消耗量,有害氣體排放,以及運轉(zhuǎn)特性方面發(fā)揮了重要作用.為了客觀評價珩磨表面,描述表面織構(gòu)的特征被量化成二維數(shù)據(jù).文章著重于兩個關(guān)鍵步驟的數(shù)據(jù)分析:預處理,其目的是去除不相干的成分和提取感興趣的信息,和提取特征以保證感興趣的表面特征能夠得到可靠的數(shù)值估計,如珩磨角,溝槽參數(shù),表面缺陷等,評估結(jié)果可以很容易的應用于用戶的評價。
關(guān)鍵詞:珩磨,表面紋理,自動視覺檢測
1、簡介:
內(nèi)燃機氣缸孔是用珩磨的方法加工的,經(jīng)過該加工的表面主要由兩個隨機在氣缸對稱軸不同角度出現(xiàn)的螺旋槽帶組成。紋理質(zhì)量對于氣缸的干燥作業(yè)性能,石油消費量,有害氣體排放,和運行性能是非常重要的。直到目前,專家們?nèi)匀灰揽炕谖⒂^圖像的視覺觀察來評價珩磨組織。這種方法枯燥,具有很大的主觀性,并且耗時。為了得到客觀和可重復性的結(jié)果,一個自動化的方法檢查是必要的。
2 、檢查方法
2.1表面數(shù)據(jù)
有一些不同的方法來衡量的珩磨表面。從表1中可以看出,傳統(tǒng)的方法,機械筆只執(zhí)行表面輪廓的一維測量。與此相反,灰度圖和光學簡圖提供二維數(shù)據(jù)在合理的時間。
由于珩磨紋理的橫向幾何特征只能進行分析二維數(shù)據(jù),在后面的討論中,我們將集中分析這樣的數(shù)據(jù)??疾斓呐c不同的測量原理相關(guān)的特征也被列入本表。
描述珩磨紋理重要特征的信號模型是本文所討論的評價方法的基礎(chǔ)?;谶@個模型,可以展示明確的和數(shù)學上完整定義的特性,使得組織評估具有重現(xiàn)性和客觀性。這種方法不同于許多廣泛應用的方法—如依靠神經(jīng)網(wǎng)絡(luò),它往往被視為一個“暗箱”[1] 。 特征的選擇是基于珩磨圖 [2] ,和許多專家的意見,并且也在分析輪廓數(shù)據(jù)的實例中通過增加新的體積參數(shù)來拓展該方法。這得出了一系列的可以滿足個人用戶需求的特征。
2.2珩磨組織性能
圖1顯示一些珩磨組織的性能,在這個基礎(chǔ)上來定義特征。最常用的是粗糙度參數(shù),例如那些基于承載比曲線(雅培曲線)的參數(shù) [ 3 ] ,以及Ra,Rz和Rmax。 [ 4 ] 。然而,處理珩磨表面,重要的是要確定一些將橫向幾何形狀量化的特征。通過這種方式,最相關(guān)的紋理特性可以被描述,如珩磨角度,材料涂片,斷溝, 雜散溝槽 ,洞,外構(gòu)和薄片,如圖1所示。此外,描述溝槽平衡,穩(wěn)態(tài)的存在,凹槽形狀,裂縫,轉(zhuǎn)折溝槽,零散標記的特征也需要。
機械鐵筆
灰度圖像
光學輪廓
測量區(qū)域
1-D
2-D
2-D
深度信息
是
不
是
橫向幾何信息
不
是
是
覆蓋整個表面
非常耗時
盡可能合理努力
非常耗時
計算處理費用
低
高
高
非接觸測量
不
是
是
標準化參數(shù)
是
不
是
圖表1:表1 :比較機械手寫設(shè)備,灰度圖像和光學簡圖灰鑄鐵氣缸套
圖1 :珩磨紋理顯示橫向特點和缺陷: (a)材料涂片 ,溝中斷; (b)雜散溝槽; (c)孔或外構(gòu)的合作; (d)薄片。
2.3自動檢測
圖2顯示自動檢測應用的概述和其在珩磨加工中對質(zhì)量控制的目的。一個二維或三維傳感器提供珩磨表面的數(shù)據(jù)g(x),其中x = (x,y)T∈ R 2指橫向空間坐標系。灰色塊圖是傳感器數(shù)據(jù)處理系統(tǒng)的一部分,該系統(tǒng)的輸出數(shù)據(jù)可以用來簡單地說明表面質(zhì)量,還可以在加工工藝發(fā)生中斷時發(fā)出警報,或可以通過反饋控制器調(diào)節(jié)珩磨過程,因為珩磨組織包含有關(guān)功能和加工過程的信息,不論珩磨后珩磨刷執(zhí)行或不執(zhí)行。 以下各節(jié)中,我們將集中于自動化檢測的兩個關(guān)鍵步驟:對傳感器數(shù)據(jù)的預處理和特征提取,我們將針對這些步驟舉出一些例子。
圖2 :自動檢測的磨練表面。
3、預處理
預處理的目的是要抑制無關(guān)部分,即不均勻性i(x)和外界干擾b(x) ,同時增強感興趣的信息,比如組織t(x)。在圖像數(shù)據(jù)的獲取過程中,不均勻性i(x)可能是由于表面光潔度的空間差異。其他的產(chǎn)生外界干擾的原因包括偏離理想情況下的溝槽和缺陷,比如說材料涂片和薄片等。
我們用一個信號模型g(x)來描述傳感器數(shù)據(jù),包含組織t(x)和無關(guān)成分i(x)和b(x):
為了能夠替代感興趣的信息t(x),首先要進行以下的假設(shè):不同的成分必須在數(shù)學上是可以進行區(qū)分的。
如圖3所示,經(jīng)過嚴格的分離程序,我們可以得到原始數(shù)據(jù)的成分。然后,分離出不需要的成分,經(jīng)過逆變換可以得到預處理的結(jié)果。
圖3:預處理的原理
這種處理過程的好處是簡化了特征提取的步驟,并且使圖像處理過程更穩(wěn)定,這可以通過以下的例子來說明。
3.1 均質(zhì)
當凹槽組織被數(shù)據(jù)采集過程中的非均勻性強度所降級,如非均勻照明,均質(zhì)化就可以制止這種有害的組成部分[ 6 ] 。圖4用這種方法做某一特定組織的例子。圖的左邊,是原始的形態(tài)。中間是用標準的同質(zhì)化方法——同質(zhì)濾波,假設(shè)非均質(zhì)和組織結(jié)合——得出的結(jié)果。特別是在左上角的部分,這張圖像的對比度很差。右邊的圖像是采用了圖3中的模型得出的結(jié)果。在這種情況下,對比度和臨界值都是均勻的,它基于一種綜合考慮了感興趣的信號和非均勻干擾信號的模型[6]。這的結(jié)果顯然比前者更均勻,并且使組織分析更具有說服力。
圖4 :均質(zhì): (左)規(guī)劃紋理; (中)同態(tài)濾波; (右)基于模型的同質(zhì)化
3.2紋理分解
下一個例子是對磨紋理進行,以簡化特征提取。由于珩磨組織很復雜,如果根據(jù)方程 (1)能夠得出部分組織信號g(x),那么檢測任務所需要的特征的提取就變得很容易了。這樣,開發(fā)出一種從組織信號g(x)中分離出包含直接結(jié)構(gòu)(比如溝槽)的成分函數(shù)t(x)和表現(xiàn)各向同性的組成(比如背景,包括缺陷和物體)的函數(shù)b(x)。在這種情況下,我們將利用均質(zhì)假設(shè)。
幸運的是,一個非常有效的執(zhí)行此分離的算法已經(jīng)存在[7].圖5的左邊顯示的是原始珩磨紋理,其他兩個圖像的結(jié)果代表的是應用該算法計算出的自適應紋理結(jié)果。在溝槽的組織中,只有理想的凹槽才能被觀察到,而背景圖像包含所有偏離理想槽和缺陷以及其他物質(zhì)。至于更全面分離算法的討論,有興趣的讀者可參照[ 7 ]。
圖5 :紋理分解:(左)珩磨紋理;(中心)凹槽紋理; (右)背景紋理。
3.3、參考表面
最后為了消除形狀組成,我們將定義參考表面。圖6展示了珩磨表面輪廓的軌跡,光滑的曲線描述了將要消除的形狀組成。但是,對流低通濾波的方法導致溝槽區(qū)域的畸變,如用點劃線表示的。我們針對該問題已經(jīng)開發(fā)出了一種二維迭代的濾波器來取代高斯濾波,這種方法在處理深的溝槽時優(yōu)勢尤為明顯。圖7中所描繪的三維圖形展示了珩磨表面的一個部分和應用此方法算出的參考表面。
圖6:參考表面:對流低通濾波的問題
圖7 :原件磨練表面和參考面。
4、特征提取
4.1珩磨角
第一個特征提取的例子是估計珩磨角度。為此,首先計算出周期(PG),它與傅里葉變換出的紋理函數(shù)成正比。
周期是一個估算的功率譜密度( PSD )函數(shù)的量,它指出了產(chǎn)生紋理的隨機過程的譜線的性質(zhì)[ 9 ] 。然后,徑向估算周期,如圖8所示。
由于珩磨紋理包括兩個系列的凹槽,估算功能也顯示出兩個極大值。珩磨角度估算兩個極大值之間的差異有關(guān):
由于平均計算使周期值盡管存在差異,徑向投影卻是一個非常光滑曲線。因此,這一方法可以快速和可靠的估計珩磨角。
4.2、溝槽參數(shù)
下面的例子是關(guān)于溝槽參數(shù)的提取的。這是基于Radon轉(zhuǎn)換來完成的,這種轉(zhuǎn)換是將二維圖像的每一行畫作轉(zhuǎn)換區(qū)域的一個點,如圖9所示[10]。接著,每一個Radon轉(zhuǎn)換的和溝槽相對應的明顯的峰都通過形態(tài)濾波被檢測到。最后,對于每一個被檢測到的溝槽,其相應的參數(shù)(振幅,寬度,位置和角度)都可以通過濾波器的輸出來估測[9]。
圖9 :圖示Radon變換
4.3、缺陷的檢測
在3.2節(jié)中,我們介紹了一種分解珩磨組織的算法。本節(jié)將重點討論所得到的背景組織,它包含了關(guān)于缺陷和物質(zhì)的主要信息,并且將研究一種可以通過圖像來檢測缺陷的有效方法。它是上一小節(jié)溝槽檢測的改進,如圖11[11]。
在這種情況下,通過分解得到的溝槽圖像10的Radon轉(zhuǎn)換主要是收集有關(guān)溝槽的信息,如圖10(b)。
此外,分布在溝槽的缺陷也通過背景圖片的Radon變換集中于Radon域的峰上,見圖10 (d)和(e)。通過在Radon域內(nèi)結(jié)合凹槽紋理和背景紋理,只有代表缺陷溝槽的峰仍然存在,見圖10 (c)項。此圖片最明顯的高峰對應于圖10(f)中的三個溝槽,它們確實是原始圖像中最突出的凹槽缺陷如圖5(a)。
圖10:缺陷檢測:(a)溝槽圖像;(b)圖像a的Radon轉(zhuǎn)變;(c)b和e的乘積;(d)背景圖片;(e)圖像d的Radon轉(zhuǎn)變;(f)檢測到的溝槽
5、總結(jié)和結(jié)論
本文展示了如何利用信號處理方法從不同的方面來自動評估灰鑄鐵中珩磨組織的相關(guān)性能。為了進行有效地自動評估,首先需要進行預處理,并且一個明確的和數(shù)學上完整定義的特征導向的方法,通過納入深層數(shù)據(jù),新的與功能相關(guān)的參數(shù)就可以被計算出來。在以往的做法,只有粗糙度參數(shù)的一階統(tǒng)計數(shù)據(jù)被用來量化需要研究的特點。但是本文所探討的方法,,是基于紋理的橫向幾何特征的基本分析來進行的,包括那些與高階相關(guān)的統(tǒng)計情況。這使專家和標準所提出的自動評估可以適用于不同的公司。
參考文獻: