Detection of color faults on cable insulation - the IWCS Proceedings ...

<strong>Detection</strong> <strong>of</strong> <strong>color</strong> <strong>faults</strong> on cable insulationHorst A. ScheidR&D, Siebe Engineering Germany53577 Neustadt / Wied, Germany+49-2683-3003-0 · scheid@siebe.deAbstractThe difference <strong>of</strong> <strong>color</strong>s between two cables can be safetyrelevant. For controlling purposes, <strong>of</strong>ten visual inspection with<strong>color</strong> cards is used to check the acceptance limit <strong>of</strong> the specified<strong>color</strong>s. Visual assessments are subject to subjective impressionsarising from factors such as illumination, angle <strong>of</strong> vision and the<strong>color</strong> sensitivity <strong>of</strong> the observer, whereas an automatic system getsobjective, reproducible and comparable results. This paperintroduces some <strong>color</strong> basics and depicts important differences in<strong>color</strong> measurements between cable applications and the paintindustry.A new developed <strong>color</strong> checking system has been shown to beeffective for cable production lines as well as for <strong>of</strong>fline testing.Various tests to define the limitations on geometry and movement(speed, turning) are described for single <strong>color</strong>ed and stripe codedcables. The measurement techniques ‚best-fit-scan’ and ‚tolerancescan’ will be explained.The result is a quasi continuous inline <strong>color</strong> control similar todiameter or spark testing and a simple <strong>of</strong>fline quality checkindependent from the human eye <strong>color</strong> perception.2. Color metricsFor a better understanding <strong>of</strong> <strong>color</strong> measurements, it is useful firstlyto define some basics <strong>of</strong> <strong>color</strong> perception and <strong>color</strong> metrics. Just todepict the difficulties in interpretation ‚<strong>color</strong>’ by the human eye,picture 1 shows two squares A and B. Every person classifies A tobe more dark than B, even though they both have the same grayvalue. This (like many other optical illusions) explains, whyobjective <strong>color</strong> specification by the human eye is nearly impossible.Keywords: wire; cable; quality control; <strong>color</strong>; colour; checking;testing; fault detection1. IntroductionIn today’s cable production, it is a common standard and state <strong>of</strong> theart in automotive wire production to use automatic <strong>color</strong> changingsystems and automatic <strong>color</strong> batch dosing systems on extrusionlines. On such production lines for automotive wires a huge number<strong>of</strong> combinations <strong>of</strong> main and stripe <strong>color</strong> are used and can be presetwithin the line control menu.For quality control, concentricity, diameter, capacitance and spark<strong>faults</strong> are constantly measured and protocolled. Readings canautomatically influence and correct the extrusion parameters. Butthe correctness <strong>of</strong> cable <strong>color</strong>s is still left to the imagination and skill<strong>of</strong> the line operator, to recognize the correct <strong>color</strong>s in accordancewith relevant standards and auditing procedures.The proper <strong>color</strong> is checked either visually inline or after thecompletion <strong>of</strong> a drum by inspection <strong>of</strong> the top layer. Start and end <strong>of</strong>the <strong>color</strong> changing process is normally not monitored duringrunning production. The scrap length is set by means <strong>of</strong> empiricalvalues under consideration <strong>of</strong> a safety value which is longer thanactually necessary.It is therefore obvious, that wrong <strong>color</strong>s cost valuable productiontime and material scrap. The logical consequence <strong>of</strong> theseconsiderations is the need for some automatic inline <strong>color</strong>measurement.Figure 1: Optical Illusion : Square A and B have the samegray value, but are interpreted by human eye as differentbecause the nearest neighbourhood is different [1].To describe <strong>color</strong> in physical terms, the base is a part <strong>of</strong> theelectromagnetic spectrum that has wavelengths from 350 to 800 nmand will be recognized by the human eye as ‚<strong>color</strong>’ (in ascendingorder violet-indigo-blue-green-yellow-orange-red). A betterphysiological representation is the so called <strong>color</strong> wheel (or <strong>color</strong>circle), where different circular sectors are filled with different<strong>color</strong>s. Colors in opposite sectors are designated as complimentary,that leads to the well known RGB model : With the three basic<strong>color</strong>s red, green and blue. All other <strong>color</strong>s can be created bysuitable mixing. Mixing complementary <strong>color</strong>s 1:1 results in aneutral gray or white (additive RGB-mixing). This model is verycommon for camera or monitor applications, but it is a puremathematical description without any feeling for human <strong>color</strong>depiction. In 1927, the German ‚Reich-Ausschuß fürLieferbedingungen’ (an organisation for quality assurance) arrangeda <strong>color</strong> chart, which should serve as reference for <strong>color</strong>ed parts. Thattable is nowadays still very common in industry as ‚RAL Paletteclassic/design/effect' [2]. But also this attempt does not include thecomplete continuum <strong>of</strong> <strong>color</strong> variations and so it is not suitable foran automated system. In the first decade <strong>of</strong> the 20 th century, AlbertMunsell separated hue, value, and chroma into independentInternational Wire & Cable Symposium567Proceedings <strong>of</strong> the 61st IWCS Conference

• With a very short sampling time an averaging over acertain number <strong>of</strong> single spots eliminates local deviations.This is justifiable, as <strong>color</strong> changes in extrusion have arelative long transition time caused by mixing effects in thebarrel• Object movement (Jitter) has to be minimized at thesensor position. This is important for the object-sensordistance ds (illumination reduces with ds 2 ) as well as fortransversal movement, where the object is leaving the scanspot partially or completely.• The wire geometry is detected as a side view on acylinder surface. This results in a <strong>color</strong> variation from thecylinder center view towards the cylinder border. This effectis additionally influenced by the surface roughness. As bothconditions cannot be changed, the final <strong>color</strong> value cannotbe interpreted as an absolute measurement but as relativemeasurement with high reproducibility.Normally one line runs different conductor/insulation diameters.The device should be able to work with various geometries (over acertain range) without mechanical preparation or sensorrecalibration.One more challenge is the measurement in a production <strong>of</strong> <strong>color</strong>coded wires (one or two stripes). As the final <strong>color</strong> establishes afterthe cooling down <strong>of</strong> the polymer, sampling has to be done behindthe cooling trough. Caused by redirecting wheels and the productitself (particularly stranded conductor), the wire can turn around thelongitudinal axis in an irregular way. Therefore the sensor detectssometimes the main <strong>color</strong>, sometimes the stripe <strong>color</strong>, or both at thesame time in the scan field. Figure 3 gives an impression <strong>of</strong> thesensors view on a two <strong>color</strong>ed wire. With sophisticated mechanics,the wire turning can be changed to be more regular and used formain and stripe <strong>color</strong> detection with only one sensor.Cable producers and polymer <strong>color</strong> batch suppliers rarely use theo.m. L*a*b* <strong>color</strong> notation. Therefore, databases with reference<strong>color</strong>s should include other notations like RAL or Munsell as well.4. Typical applications and inlinemeasurement test resultsA device (reffered to as SCM=Siebe Color Match) was developedaccording to the rules mentioned in section 3, including optics forcable diameter 1-10 mm, internal illumination to preventmetamerism (a phenomenon <strong>of</strong> matching <strong>of</strong> apparent <strong>color</strong> <strong>of</strong>objects with different spectral power distributions) and an industrialcomputer (IPC) for image processing.System functional design and accuracy was first tested with single<strong>color</strong> wires to verify the aim <strong>of</strong> a resolution <strong>of</strong> at least dE~3, so theresult would be the same or better than checking by the human eye.Figure 4 shows a detailed yt-plot from a measurement period <strong>of</strong> 15minutes for a yellow wire with 2.5 mm diameter. The histogram <strong>of</strong>all dE values in figure 4 depicts a maximum <strong>of</strong> around 0.75 (averagevalue 0.89) and is a pro<strong>of</strong>, that the system has a resolution <strong>of</strong>minimum dE=1. No values higher than 3 are recorded, so athreshold could be set to values <strong>of</strong> 5-7 for <strong>color</strong> fault alarm.Figure 3 : Simulated 2-<strong>color</strong>ed wire in the scan field. Theupper part is a view into the cables longitudinal directionwith the sensor at the top and its aperture indicated as acone. The lower part shows the sensors ‚camera view’ ata coincidental time (with average <strong>color</strong> values at the rightside).Figure 4 :Left side – dE calculated from L*a*b* values accordingequation (2), setpoints 87.62 / -66.04 / 39.10)Right side – Histogram <strong>of</strong> dE with a binning <strong>of</strong> 0.05.Average dE = 0.89.Due to the above mentioned jitter and surface variations, FWHMvalue <strong>of</strong> the luminance channel L* is typically higher than the purechromaticity channels a* and b*. Therefore in equation (2), factork L >1 is used to compensate for this. As result, dE tolerance field isslightly elliptical instead <strong>of</strong> spherical with k L =1.By putting one grain <strong>of</strong> blue masterbatch into the feeding <strong>of</strong> thescrew, dE was increasing significantly to values above 10 (middle <strong>of</strong>figure 5) for 1-2 minutes. The smaller increase <strong>of</strong> dE some 3minutes later can be interpreted by blue residues that were stillsomewhere on the screw for a certain time. Only the main deviationwas found later by visual inspection (left side <strong>of</strong> figure 5, upper andlower sample, in the middle the undisturbed cable is shown).To check the geometrical resolution, some small geometries withdifferent <strong>color</strong>s are measured and evaluated similar to the firsttest. Histograms are fitted to Poisson-distribution (alternativelyWeibull-distriubution) for easier comparison between the runs. Infigure 6, the ‘green production’ has significantly better averageInternational Wire & Cable Symposium569Proceedings <strong>of</strong> the 61st IWCS Conference

and deviation values than the ‘blue production’ (Poissonparameter λ is the indicator for distribution). A possible reasonmay be less melt homogeneity when using blue masterbatch. Asmentioned in section 2, with a tolerance setting <strong>of</strong> dE=7, allsamples would pass, but ‘blue’ is produced near the limits.switching <strong>of</strong>f the co-extruder for stripe in a running production foraround 40 seconds.Figure 5 : Forced <strong>color</strong> fault by putting blue masterbatch into the barrel feedingTo give a better feeling <strong>of</strong> <strong>color</strong> distances between these 4samples, their reference setting and dE to the other references arelisted in table 2.Table 2 : L*a*b* reference <strong>color</strong>s measured via ‘teach in’and <strong>color</strong> distance between reference <strong>color</strong>s.COLORNAMEReferenceOrange (Or)ReferenceBrown (Bn)ReferenceBlue (Bl)ReferenceGreen (Gn)L* a* b* dE 76 toOrdE 76to BndE 76to BldE 76to Gn39.1 26.3 44.0 ⎯ 24.8 54.3 53.633.2 15.8 22.3 . ⎯ 29.9 40.032.1 6.6 -6.1 . . ⎯ 41.749.7 -20.6 20.2 . . . ⎯The second main step was to measure on a stripe coded wire. Fora separation <strong>of</strong> both <strong>color</strong>s from the raw signal, statistical methodsare used as the portion <strong>of</strong> main and stripe <strong>color</strong> in the scan field isvariable. So the raw L*a*b* values vary between main and stripvalues. As the longitudinal wire rotation speed changes, theresidence time <strong>of</strong> one <strong>color</strong> under the sensor position cannot bepredicted. A ‚turn mechanism’ was tested to make the rotationmore regular and to insure, that both <strong>color</strong>s come into the scanfield within a time period shorter than the alarm time. Acontinuous dE plot similar to figure 6 is not useful. The solutionis to trigger each tolerance channel when the corresponding <strong>color</strong>is found. As long as the alarm time is longer than the re-triggeringtime, the tolerance keeps in ‘ok’ status. With very small wiregeometry and/or with small stripe width, even when the stripeposition is in the scan field middle, the sensor detects a bit <strong>of</strong>main <strong>color</strong> at the stripe borders. This is limiting the <strong>color</strong>separation as there is more ‚mixing’ between main and stripe<strong>color</strong> at smaller geometries.After testing and verifying that main and stripe <strong>color</strong> could bedetected in this way, a stripe missing fault was forced byFigure 6 : Different <strong>color</strong>ized samples <strong>of</strong> the same cablegeometry and a test length <strong>of</strong> around 300m each.Left side : Direct deviation signal from the device.Right side : Poisson fit curve and mean dE-value.Figure 7 illustrates the result in the raw data (only showing the<strong>color</strong> channels a* and b*) : during normal production, valuestoggle between main and stripe <strong>color</strong>. After the co-extruder wasswitched <strong>of</strong>f (at 10 seconds on x-scale), the stripe signal slowlydisappears towards the main <strong>color</strong> simultaneously to thedecreasing stripe width. After around 5 seconds, the raw signalmoves only within the main <strong>color</strong> tolerances and the togglingdisappeared. The co-extruder was switched on again at around 50seconds on the x-scale and stripe signal ramped up in 5 seconds tothe same condition as before.International Wire & Cable Symposium570Proceedings <strong>of</strong> the 61st IWCS Conference

Figure 7 : Stripe missing test – only shown raw data onthe a*- and b’- channel. Co-extruder was switched <strong>of</strong>f atx-scale position 10s and switched on again at position50s.Another point <strong>of</strong> interest might be the stripe to main ratio. As thesensor detects only the average <strong>color</strong> in the spot field, it is notpossible to measure directly the stripe width. In case <strong>of</strong> a constantlongitudinal product rotation, the time interval <strong>of</strong> main t m and stripet s <strong>color</strong> found in their tolerance interval can be integrated for acertain time T and the resulting time ratio∑T∑Tttmsshould be almostthe same as the geometrical ratio. First trials under optimumconditions gave almost satisfying results with T>10s, but scan fieldsize, jitter and rotation irregularities are still a challenge for anevaluation with high evidence. In figure 8, a short run with a yellowgreencable shows a stripe ratio <strong>of</strong> around 40% in total,corresponding to 20% <strong>of</strong> the circumference for each <strong>of</strong> both stripes.Towards the end <strong>of</strong> the run, co-extruder was ramped down firstbefore stopping, indicated by retrigger breakdown as o.m. andreduced stripe to main ratio.5. Actual device specifications andforthcoming developmentThe user interface <strong>of</strong> the device should be quite easy for the lineoperator without losing setup flexibility or detail information,comparable with inline wire centricity measurement. Based on anIPC, control <strong>of</strong> the sensor is completely transparent for the user. Anon-contact measurement reduces the risk <strong>of</strong> sensor damage. Very<strong>of</strong>ten the wire isn’t completely dry while passing the device. Thiscaused contamination <strong>of</strong> the sensor surface but could be solved byinstallation <strong>of</strong> protecting windows. To start up the line ormaintenance, it is possible to swing the complete device in a ‘safeposition’. Connection to the line PLC is possible via simple digital24 DCV signals (signal status and device enable; <strong>color</strong> faultmessages could be implemented to the coil protocol like spark ordiameter <strong>faults</strong>) or with a more complex network communication viaTCP/IP to have access to the recipe database or to support hostcomputer link. To be independent <strong>of</strong> all production lightingconditions, the scan field is almost covered and illumination is doneby an internal LED light source. Line speed is no limitation, just thegap between subsamples increases.Two measurement modes are implemented :• In the ‘best-fit-mode’, the <strong>color</strong> difference to eachdatabase <strong>color</strong> is calculated. After comparing, the bestmatching <strong>color</strong> from the database is shown as the result.In case this <strong>color</strong> is same as a specified <strong>color</strong>, the statusis raised (‘ok’ for single <strong>color</strong> check, status traffic light+1 for dual <strong>color</strong> check). Same is done when a second<strong>color</strong> is specified. This mode is a simple procedure if noclear specification is known. By just running ameasurement, possible <strong>color</strong> shades can be tested asreference.• dE measuring (=tolerance scan)is the recommendedmode if a <strong>color</strong> shade is given. The device is comparingthe measured <strong>color</strong> parameters with those <strong>of</strong> thespecified <strong>color</strong> even when there is a better matching<strong>color</strong> in the database. When the dE value is better thanthe specified tolerance, the status is raised (same as inbest fit mode).Figure 8 : Stripe to main ratio test – yellow ratio signal attop is the sum for both stripes. Falling trigger level signal<strong>of</strong> stripe <strong>color</strong> below and decreasing ratio indicate rampdown <strong>of</strong> co-extruder (arrow) screw speed.Figure 9 : User interface <strong>of</strong> the SCM. In the middle left aschematical cross section <strong>of</strong> the wire shows detectedmain and stripe <strong>color</strong>. Lower middle shows the statustransferred to the PLC (green=both <strong>color</strong>s in tolerance,yellow=one is missing / out <strong>of</strong> tolerance, red=double faultor wrong recipe). At the right, <strong>color</strong> setting is displayed.International Wire & Cable Symposium571Proceedings <strong>of</strong> the 61st IWCS Conference

If the reference <strong>color</strong> is not included in the actually used database,the system needs a teach-in. This is done once when the wire runs ingood production and the detected new reference can be stored forfurther production <strong>of</strong> this wire type. Any number <strong>of</strong> recipe can bestored. Until now the recipe database functions are very simple andshould include search functions in future s<strong>of</strong>tware. Default databasefiles available until now are L*a*b*, LCH, RAL and Munsell.Figure 9 shows the main screen that includes the most importantinformation for the user. Additional windows can display moredetails, like separated L*/a*/b*-channel measurements, varioussignal history or trend information. For later evaluation, a subsample<strong>of</strong> raw data is stored to the IPC hard disk in an Excel compatibleformat.6. ConclusionsEven under cable production conditions that are difficult for <strong>color</strong>detection, a reproducible <strong>color</strong> response could be obtained with theSCM. Using the turn characteristics <strong>of</strong> the cable during production,stripe coded cables can be checked with only one sensor, keepingthe system less expensive.SCM is not designed to be a spectrometer with absolutemeasurements within each wavelength interval but a <strong>color</strong>imetergiving deviation signals from reference <strong>color</strong> values. It has beenshown that this is sufficient for quality control in most cableapplications.Point <strong>of</strong> discussion should be the combination <strong>of</strong> stripe and main<strong>color</strong>s. For example main <strong>color</strong> red with a brown stripe is difficult toseparate – for an automatic device as well as for a line operator. Sothis is no problem in <strong>color</strong> measurement itself but a possibleimprovement <strong>of</strong> quality control by a simple re-definition <strong>of</strong> <strong>color</strong>pairs with high dE.7. AcknowledgmentsSpecial thanks to the German AIF (Ministry <strong>of</strong> Economics andTechnology) for temporary financial support for this project.8. References[1] Edward H. Adelson, MIT, 1995[2] https://www.ral-farben.de/492.html?&L=1[3] Joint ISO/CIE Standard: CIE Colorimetry — Part 1-5, ISO11664-1··5 / CIE S014-1··5/E:2006-2007[4] Masataka Okabe, Kei Ito (2008-02-15). Color blind barrierfree. http://jfly.iam.utokyo.ac.jp/<strong>color</strong>/[5] R.S. Berns, Billmeyer and Salzman’s principles <strong>of</strong> <strong>color</strong>technology, John Wiley & Sons, 3 rd edition, 2000Figure 10 : Inline version <strong>of</strong> SCM with touch screen forstand alone operating.In addition to the inline version that is suitable for measurementduring cable production, a second setup was build up for test bayusage. Provided with screen, keyboard and a sample turnmechanism and completely installed on a desk, its purpose is <strong>of</strong>flinemeasurement for short sample testing.To increase the application range, tests with cable diameter down to0.2 mm (single <strong>color</strong>ed) are prepared now.9. About the AuthorDr. Horst Scheid, born 1958 in Germany, studied Solid StatePhysics at the University <strong>of</strong>Saarbruecken. He has worked since1986 in the injection mouldingmachinery research group <strong>of</strong> Arburg.In 1990 he joined the COSY beamaccelerator group in Juelich NuclearResearch Centre and received hisdoctoral degree in Nuclear Physicsfrom the University <strong>of</strong> Bonn. Since1995, his activities focus on foamextrusion, new extrusion processingtechniques and enhanced inlinequality control at Siebe Engineering.International Wire & Cable Symposium572Proceedings <strong>of</strong> the 61st IWCS Conference