An image-based measurement system for the characterisation of automotive gaskets

Measurement 25(1999)169–181An image-based measurement system for the characterisation ofautomotive gasketsa a ,a b*L.Angrisani ,P.Daponte ,A.Pietrosanto ,C.Liguori a `Dip .di Ing .dell ’Informazione ed Ing .Elettrica ,Universitadi Salerno ,Via Ponte don Melillo 1,84084Fisciano (SA ),Italy b `Dip .di Ing .Industriale ,Universitadi Cassino ,Via G .Di Biasio 43,03043Cassino (FR ),Italy Received 15June 1998;received in revised form 23November 1998;accepted 30November 1998AbstractThe paper describes an automatic measurement system for the estimation both of geometrical and physical parameters of automotive gaskets.The hardware of the system consists of a general purpose computer and a camera;the software implements an original measurement procedure able to characterise the gasket under test from its digitised image.The whole system is first described focusing the attention on the measurement procedure.Then,the results of the metrological characterisation of the proposed system are given along with those obtained from experiments on different automotive gaskets.©1999Elsevier Science Ltd.All rights reserved.Keywords :Image processing;Contour detection and extraction;Automatic measurement system;Automotive gasket1.Introductionmainly,of the scraps on the global cost is remarkable for this type of production [1–4].The increasing internationalisation of economicalThe in-process quality control of such production exchanges forces automotive component manufactur-cycles is generally carried out according to the ers to be more and more competitive in a more andfollowing procedure,still based on manual inspec-more selective market.In this context,the survival oftions:(i)accurate cutting of a very thin sample of the manufacturers is conditioned by pursuing the maxi-gasket (length:2–3mm),(ii)comparison between mum quality at minimum costs.This goal can onlythe sample and a reference drawing of the transversal be achieved by means of an industrial managementsection of the gasket by means of a magnifying lens,policy fully oriented to quality and based on in-and (iii)measurement of some geometrical parame-dustrial process integration.ters of the sample by means of a drawing rule.These These requirements are more pressing in theoperations must continuously be repeated at the production of automotive gaskets where automationbeginning of the extrusion process until the gasket and integration seems to be still difficult.Further-has the required characteristics,and periodically more,the incidence of the quality control and,during steady-state conditions.They take experts more than 6–7min during which the production cycle is running;about 80–100m of scraps (taking into account typical extrusion speed)are produced *Corresponding author.Tel.:139-89-964248;fax:139-89-964218;e-mail:daponte@unina.it when the gasket does not match quality 0263-2241/99/$–see front matter ©1999Elsevier Science Ltd.All rights reserved.PII:S0263-2241(98)00076-1170L.Angrisani et al./Measurement25(1999)169–181requirements.Furthermore,relative percentage un-of a CCD camera underexposed.The underexposed certainties on measured values never smaller than image is characterised by high contrast between the 5%are reached and,as for gasket sponginess estima-gasket transversal section and the overexposed en-tion,more complex operations are required.vironment thus allowing easy contour extraction.A Automatic in-process quality controls could allow further image with overhead illumination shows a lower percentage of scraps thus reducing the non-itself suitable for an accurate analysis of the micro-quality costs related to the storage and shipping of scale bubble structure of the transversal section. defective products,and increasing customer trust.The paper proposes an automatic system for themeasurement both of geometrical and physical pa- 3.The hardwarerameters of automotive gaskets,which is able toreduce both the time required by gasket quality To reduce costs and make successive additions of control and measurement uncertainty.The system,computational and image digitising blocks very easy, based on a general purpose camera interfaced to a general purpose hardware is usually adopted in this PC,executes an original image-based measurement kind of applications[5–8].The proposed measure-procedure;this procedure is implemented by a series ment system(Fig.2)adopts a commercial CCD of software modules able to characterise the gasket camera which is connected to a DSP-based frame from its digitised image[5].The software modules,grabber hosted by a personal computer(PC).TheE(i)localise the contours of the gasket transversal CCD camera is the TM526A by Current Technolo-section in the digitised image,(ii)measure both gy Inc.,whose horizontal resolution is more than450 geometrical dimensions and physical parameters of rows and has a vertical resolution of more than350Ethis section,and(iii)verify that quality requirements columns.The frame grabber is the FF1by Current are satisfied.Technology Inc.,whose input section consists of an Both reliability and effectiveness of the measure-A/D converter capable of digitising data at10MHz ment procedure are assessed by analysing the results with8bits of resolution.of the metrological characterisation of the proposed The gasket sample is placed on an opaque glass measurement system.The results obtained from the sheet,as shown in Fig.3.Somefluorescent lamps application of the measurement system to an actual are suitably positioned below the glass sheet in order production cycle arefinally given.to provide maximum uniformity of illumination.Thelighting system includes also a3000lumen circularlamp placed over the gasket sample at the height of 2.The basic idea of the measurement system the camera lens.The two lighting sources aremutually excluding;thefluorescent lamps are The goal of the measurement system is the switched on for contour extraction,the circular lamp, automatic evaluation of some geometrical(height,on the contrary,for sponginess evaluation. width,thickness)and physical(sponginess)parame-The hardware is completed by a network link ters of the transversal section of automotive gaskets.between the Quality Control Laboratory and the For the sake of clarity,Fig.1shows a typical automatic measurement system[7].location of an automotive gasket in a car,a gasketpiece(gasket length ranges within3–5m),a gasketsample like those required by the measurement 4.The measurement proceduresystem,and some results of this proposed system(inner and outer contours of the transversal section The measurement procedure is described with of the gasket sample,and its geometrical parame-references to its software implementation.In par-ters).ticular,object-oriented programming(OOP)has The basic idea of the measurement system is that been adopted in order to realise aflexible and illumination from below(Fig.2)can have the modular software structure.OOP ensures exploita-transversal section of a gasket sample placed in front tion of the fundamental issues of modularity,L.Angrisani et al./Measurement25(1999)169–181171 Fig.1.Typical location of an automotive gasket;gasket piece;sample cut from the gasket under test;final results of the measurement system giving the height H,the width L and the inner and outer contours of the gasket sample.Fig.2.Block diagram of the measurement system.172L.Angrisani et al./Measurement25(1999)169–181Fig.3.Measurement system(upper);detail of the lighting system(lower).L.Angrisani et al./Measurement25(1999)169–181173 reusability and extendibility for measurement soft- 4.1.Graphical user interfaceware[9].This is particularly important for industrialapplications the evolution of which has to be pursued This module allows the inputting of data,the at limited cost.selection of the type of gasket to be analysed and the The OOP-based software structure for the pro-outputting of measurement results[9,10].posed measurement procedure is sketched in Fig.4.The software core is the Supervisor which manages 4.2.Image acquisitionboth the execution and data exchange of the follow-ing software modules:(i)Graphical User Interface A suitable module has been set up for capturing [9,10],(ii)Image Acquisition,(iii)Image Process-two images of the gasket sample sequentially. ing,(iv)Gasket Geometrical and Physical Parameter Thefirst image,underexposed,is acquired after Evaluation,and(v)Network Interface for the Quality switching on thefluorescent lamps below the opaque Control Laboratory.glass sheet.An example is given in Fig.5a;it is In the following,the most significant features and worth noting the high contrast between the transver-functionalities of all modules are described;the sal section of the gasket sample and the overexposed description order reflects the execution order.The environment.attention is principally paid to the Image Processing The second image is acquired with only overhead module;it allows both detection and extraction of the illumination thus enhancing the vision of the micro-inner and outer contours of the gasket transversal scale bubble structure of the transversal section of section from the digitised image of a gasket sample.the gasket sample.Fig.4.OOP-based software structure for the measurement system.174L.Angrisani et al./Measurement25(1999)169–181transversal section of the gasket sample.The task isaccomplished by applying a sequence of suitablealgorithms;all of them are original but Snakealgorithm,a well-known edge detector[11].Thefollowing steps can be distinguished.4.3.1.Threshold algorithmDue both to the not negligible height(2–3cm)ofthe gasket sample and theflexibility of the rubber,the digitised image includes some lateral(externaland/or internal)surfaces of the gasket sample(Fig.5a)in addition to its transversal section.Moreover,illumination from below has lateral surfaces betterlighted than the transversal section.As a conse-quence,the grey tone histogram of the whole imageis always bimodal(Fig.6):its right peak(lighterpixels)is related to the background,its left peak(darker pixels)is related to the transversal section,and the grey levels between the two peaks are relatedto noise,shadows,and lateral surfaces.Both loca-tions and heights of the two peaks change both withsample dimensions and lighting conditions thusmaking it difficult to establish afixed,optimalthreshold value.For this reason,an adaptive thres-hold is adopted(a new value is established for eachacquired image).The Threshold algorithm analyses the grey tonehistogram of the digitised image and computes thethreshold as weighted average of the level of the leftpeak and those of the contiguous grey tones.Itproduces a binary(two colours)image;any pixelcharacterised by a grey tone greater than the thres-hold is set to white(255),otherwise is set to blackFig.5.(a)Underexposed image:the arrows indicate the lateralsurfaces of the gasket sample which make difficult contourdetection and extraction process;(b)threshold algorithm results;(c)closing algorithm results;(d)tracking algorithm results;(e)snake algorithm results.4.3.Image processingThis module receives as input the aforementionedunderexposed image.It is mandated to the detectionand extraction of the inner and outer contours of the Fig.6.Grey tone histogram of the image shown in Fig.5a.176L.Angrisani et al./Measurement25(1999)169–181and its width along the horizontal one as shown in values of which are strongly related to the value of the samefigure;the section area considered is the‘empty-to-filled’ratio[20,21].bounded by the inner contour;thickness is thedistance between inner and outer contour.Finally, work interface for the quality control sponginess is defined as the percentage‘empty-to-laboratoryfilled’ratio characterising the spongy part of thetransversal section.Its value,which ranges within This module establishes a dynamic data exchange 0–100%,has to be assigned to one of the following between the measurement system and the data base five classes:0–20%,20–40%,40–60%,60–80%,on quality trend of the factory.As a result,(i)a and80–100%.quality check on gasket production,providing the With regard to height and width,their values are customer(automotive producer)with more accurate obtained by circumscribing the outer contour with a information about gasket characteristics,and(ii)a rectangle,the sides of which are parallel to the axes better correlation between the used compound and of the coordinate system adopted.Thickness is gasket features are possible.The last information measured for each point of the outer contour as the helps the compound designer in choosing both the length of the shortest path between it and a point of best elements for the compound and the optimal ratio the inner contour(Fig.8).The value of the section between them.area is established by dividing it into triangles withthe same vertex(the centre of the section),the basesof which are consecutive chords of the inner contour. 5.Experimental resultsAs for sponginess evaluation,the second acquiredimage(the sharper one)is needed.By superimposing A prototype of the proposed measurement system the effective contours,extracted by the Snake algo-was arranged and tested at the Laboratories of the rithm,on this image,the pixels belonging to the area University of Salerno.An industrial version of the included by the two contours are selected and its system was then installed at the BTR SAIAG Sealing grey tone histogram is evaluated.Sponginess class is Systems S.p.A.factory in Battipaglia,Italy,in order established on the basis of the chief parameters of to measure the characteristics of three different types the histogram(mean,mode,standard deviation),the(A,B,C)of automotive gaskets(Fig.9).Fig.8.Thickness measurement.A part of the transversal section of the gasket sample is shown:for each point of the outer contour,P theiE procedure searches the minimum path,th between P and any point P,of the inner contour.i iE jIL.Angrisani et al./Measurement25(1999)169–181177During the former stage,both the calibration andmetrological characterisation of the measurementsystem were carried out[22–24].The calibrationaimed at,(i)tuning both the characteristics of thelighting system and distance between the camera andthe gasket sample,(ii)evaluating the numericalconstants needed for the conversion of the measure-ment results from pixels to millimetres,and(iii)establishing the relationship between the sponginessclasses and the statistical parameters of the grey tonehistogram.In particular:1.The characteristics of the lighting system and thedistance between the camera and the gasketsample had to satisfy two opposite needs:(a)improving image definition and resolution,and(b)allowing the visualisation of the whole trans-versal section even when its dimensions are largerthan the nominal ones2.The pixel-to-millimetre conversion factor wasevaluated by measuring(in pixels)the knowndimensions of the transversal section of an ironparallelepiped3.For each gasket,thefive sponginess classes wereassociated to the corresponding values(intervals)of the statistical parameters of the grey tonehistogram by means of several tests on gasketscharacterised by known sponginessA further experimental tuning has been conductedon the parameters characterising the Image Process-ing Module in order to make measurement resultsindependent from any environmental change in thefactory.With regard to the metrological characterisation,three reference iron targets with the same nominalshapes of the three aforementioned gaskets wereused.In particular,200observations were carried outfor each target.A statistical analysis was thenperformed in order to evaluate,(i)the systematicerror of the measurement system,and(ii)thestandard uncertainty[25].The systematic error oneach parameter of the transversal section was givenby the difference between the mean value of200measurements and the corresponding known value ofthe iron target.The maximum systematic error(20.92mm)resulted in area measurements due to the Fig.9.Gaskets analysed by the measurement system:type A(upper),type B(middle)and type C(lower).triangular approximation.All errors were then intro-178L.Angrisani et al./Measurement25(1999)169–181duced in the Gasket Geometrical and Physical Pa-ment,the grey tone histogram depicted in the right rameter Evaluation module as corrections.As for side of the same Fig.10singles out the class that the (ii),the total standard uncertainty(u)was given by gasket sample belongs to.For the sake of clarity, the experimental standard deviation(s)resulting Fig.12shows the results related to two different negligible the uncertainty in systematic error estima-gasket samples belonging to an acceptable(Fig.12a) tions.For each gasket,the total uncertainties on and unacceptable(Fig.12b)sponginess class,respec-height and width measurements were not greater than tively.Finally,Table2gives some values measured 0.2mm,which is lower than the manufacturing on samples taken during a production period of10h tolerance(1mm).These uncertainties were also and transmitted,via the factory network,to the confirmed by the tests carried out directly on samples Quality Control Laboratory.In the specific case,the of the rubber gaskets.The results of all tests are numerical values refer to data concerning the type A collected in Table1.gasket.Typical video results provided by the industrialversion of the measurement system are shown in Fig.10.The inner and outer contours of the transversal 6.Conclusionssection of the gasket sample and the values of thegeometrical parameters can be noticed.Only the The paper has presented an automatic image-based most significant thickness measurements,according measurement system whose image-processing pro-to the type of gasket,are displayed.The whole cedure allows the characterisation of automotive thickness trend is saved into an externalfile that can gaskets.Such a measurement system has proved to be shown off-line at operator’s request and looks like be very effective,especially with veryflexible that in Fig.11;the dashed boundary limits stand for objects.In these cases,all the traditional techniques, manufacturing tolerance.As for sponginess measure-based on contact measurement devices,fail because Table1Systematic errors and standard deviations on height(H),width(L),thickness(d1,d2,d3),and area(A)measurements,evaluated by analysing the results of200tests both on iron targets and rubber gasketsGasket type Nominal value Systematic error s(iron target)s(rubber gasket) AH(mm)15.64220.220.0400.029L(mm)15.56320.260.0570.049d1(mm) 2.710.000.0560.063d2(mm) 2.870.000.0980.089d3(mm) 3.010.000.110.070 2A(mm)86.510.610.320.40BH(mm)20.66320.240.0850.060L(mm)19.23420.310.0570.061d1(mm) 3.250.000.0940.093d2(mm) 4.280.000.0660.036d3(mm) 2.690.000.130.11 2A(mm)155.080.830.450.22CH(mm)19.84020.160.0640.059L(mm)17.29220.290.0390.030d1(mm) 2.080.000.0530.040d2(mm) 5.470.000.0770.057d3(mm) 2.190.000.0340.049 2A(mm)145.770.890.450.42180L.Angrisani et al./Measurement25(1999)169–181Fig.12.Sponginess measurements.Sponginess value belonging to an,(a)acceptable class,and(b)unacceptable class.Appendix1where l[[0,1]are the regularisation parametersithat significantly influence the solution to Eq.(A1) Snakes is an active edge model for representing[17].Setting l4(12l)emphasises regularisation, deformable contours in noisy images,proposed in yielding strongly model-driven solutions which are[18].A snake is an ordered set of points V5[v,robust to noise.In contrast,small values of l enable1v,...,v],where each snaxel v is defined on the the snake to effectively capture boundary discon-2n ifinite grid:v[E5h(x,y):x,y51,2,...,M j.tinuities,but they also render the solutions very iContour extraction by means of a snake model sensitive to noise.involves two energy functionals E,and E[17].From amongst the proposals for snake im-int extWhile the internal energy E imposes continuity and plementation,a modified version of the algorithm intsmoothness constraints,the external energy E given in[17]has been used in this paper.It includesextattracts the snake to salient image features.The a formulation of energy functionals in order to snake turns boundary extraction into an energy incorporate the gradient direction and an automatic minimisation problem.Under this framework,one and implicit selection of local regularisation parame-seeks to obtain V satisfying ters.The algorithm is mainly based on an active Lcontour model which adapts an invariant matrix to V5argmin O[l E(v)L L i i int i the rigid displacements,while control of the con-1(12l)E(v)]i51,...,n(A1)tour’s local characteristics is given to a Markoviani ext iL.Angrisani et al./Measurement25(1999)169–181181 Table2[8]F.van der Heijden,Image-Based Measurement Systems, Values measured on gasket samples taken during a production Wiley,New York,1994.period of10h and transmitted,via the factory network,to the[9]P.Daponte,L.Nigro,F.Tisato,Object-oriented design of Quality Control Laboratory measurement systems,IEEE Trans.Instrum.Meas.41(6)(1992)874–880.N H L d1d2d3A[10]P.Daponte,L.Nigro,F.Tisato,Integrating the user-interface2(mm)(mm)(mm)(mm)(mm)(mm)in an object-oriented measurement system,Measurement15 115.1815.17 2.71 2.87 2.9485.34(2)(1995)91–101.215.6015.49 2.78 2.90 3.0189.66[11]i,Deformable contours:modeling,extraction,de-315.9615.45 2.77 2.80 2.8785.48tection and classification,PhD.Thesis,University of Wis-415.9315.38 2.81 2.94 3.1188.97consin,Madison,1994.515.3815.21 2.81 2.88 2.9689.68[12]D.H.Ballard,C.M.Brown,Computer Vision,Prentice-Hall, 614.9515.09 2.79 3.09 2.8482.33Englewood Cliffs,NJ,1982.715.5715.52 2.58 2.90 3.0292.41[13]D.Geiger, A.Gupta,L.A.Costa,J.Vlontzos,Dynamic 815.1415.02 2.63 2.77 3.1389.22programming for detecting,tracking,and matching deform-915.3315.28 2.76 2.90 3.0188.61able contours,IEEE Trans.Pattern Analysis and Machine 1014.9115.17 2.63 2.91 3.0284.98Intelligence17(3)(1995)294–302.1115.2115.21 2.66 2.83 3.0185.99[14]H.Li,B.S.Anjunath,S.K.Mitra,A contour-based approach 1215.3615.00 2.76 2.89 3.0283.77to multisensor image registration,IEEE Trans.Image Proc.4 1315.2315.08 2.71 2.82 3.0883.97(1995)320–334.1414.8514.91 2.80 2.87 3.0284.72[15]F.Leymarie,M.D.Levine,Tracking deformable objects in 1515.3114.62 2.60 2.90 3.0787.69the plane using an active contour model,IEEE Trans.Pattern 1614.8614.86 2.75 2.97 3.0590.20Analysis and Machine Intelligence15(6)(1993)617–633. 1714.6314.82 2.58 2.93 3.0289.21[16]m,H.Yan,Fast greedy algorithm for active con-1814.7815.14 2.59 2.66 3.0385.53tours,Electron.Lett.30(1)(1994)21–23.1915.3815.50 2.73 2.89 2.9382.15[17]B.Su, A.Nabout,H.A.N.Eldin,Measurement of ring-2015.4515.59 2.70 2.94 2.9386.98shaped and toothed mechanical parts through image process-ing,in:Proc.1990IEEE Ind.Appl.Soc.Annu.Meeting,Seattle,WA,USA,7–12October,1990,pp.396–401. operator.Furthermore,an original procedure for the[18]M.Kass,A.Witkin,D.Terzopoulos,Snakes:active contourmodels,in:Proc.1st puter Vision,1987,pp. snaxel initialisation has been proposed(Section4).259–269.[19]S.Harrington,Computer Graphics,McGraw Hill,New York,1987.References[20]C.W.Moon,Error analysis for dimensional measurementusing computer vision techniques,in:Proc.IEEE Instrum.Meas.Tech.Conf.,San Diego,CA,USA,20–22,April, [1]G.Stella,New structures and technologies EPDM for1988,pp.372–376.automotive gaskets,in:Proc.PLAST-RUBBER1991,Milan,[21]M.A.Coulthard,Image processing for automatic surfaceItaly,May9,1991.defect detection,in:Proc.3rd IEE Int.Conf.on Image Proc.[2]H.Focht,International Polymer Science and Technology7and its Applications,Warwick,UK,18–20July,1989,pp.(T/65)(1980).192–196.[3]V.L.Lue,Rubber World26(1980).[22]M.A.Sid-Ahmed,M.T.Boraie,Dual camera calibration for [4]R.Tessieri,D.Piccinini,L’evoluzione e le prospettive dei3-D machine vision metrology,IEEE Trans.Instrum.Meas.profilati per tenuta di vetri e portiere di autoveicoli,Macplas39(3)(1990)512–516.176(March)(1996)99.[23]J.J.Aguilar,F.Torres,M.A.Lope,Stereo vision for3D [5]E.Harms,Machine vision:what can it do for you,in:Proc.measurement:accuracy analysis,calibration and industrial IEEE45th Conf.Electr.Eng.Problems in the Rubber andapplications,Measurement18(4)(1996)193–200.Plastic Industries,Akron,OH,USA,13–14April,1992,pp.[24]Y.S.Li,T.Y.Young,J.A.Magerl,Subpixel edge detection30–34.and estimation with a microprocessor-controlled line scan [6]R.L.Horst,M.Negin,Vision system for high-resolutioncamera,IEEE Trans.Ind.Electron.35(1)(1988)105–112.dimensional measurements and on-line SPC:web process[25]Guide to the expression of uncertainty in measurement,1stapplication,IEEE Trans.Ind.Appl.28(4)(1992)993–997.ed.,BIPM,IEC,IFCC,ISO,IUPAC,IUPAP,OIML,1993, [7]Z.Bien,S.R.Oh,J.Won, B.J.You, D.Han,J.O.Kim,ISBN92-67-10188-9.Development of a well-structured industrial vision system,in:Proc.16th Annu.Conf.of IEEE Ind.Electron.Soc.,Pacific Grove,CA,USA,27–30November,1990,pp.501–506.。