Measuremets of weld geometry usig image processig techology Chia-Lug Chag a, Ye-Hug Che b a Departmet of Mechaical Egieerig, Natioal Yuli Uiversity of Sciece ad Techology, Douliu, Yuli, Taiwa, ROC b Departmet of Mechaical ad Computer Aided Egieerig, Chug Chou Istitute of Techology, Yuali Towship, Chaghua, Taiwa, ROC Tel.: + 886-5-5342601 ext.4126 Fax: +886-5-5312062 E-mail: chagcl@yutech.edu.tw Abstract The weld geometry is directly related to weld stregth. The weld geometry by maual measuremet may be iaccurate due to huma errors. I this study, the image processig is proposed to measure the geometrical dimesios of bead-o-plate weld. The image oise of weld was smoothed by media filter. The boudary of fusio area was ehaced by histogram processig. The edge of deposit ad fusio area is the obtaied usig Laplacia-Gaussia operator ad zero crossig detectio. The demostrated example shows that the image processig ca effectively ad accurately measure the weld geometry. Keywords: Weld geometry, image processig, bead-o-plate weld, deposit area, fusio area 1. Itroductio The weld geometry is a fuctio of weldig process parameters. It is importat to study the effect of weldig process parameters o weld geometry, especially whe the automated weldig equipmet is used. A isufficiet throat or cocave type weld profile reduces the cross-sectioal area of the weld. Therefore, the cocave type weld is a poit of weaess. Excessive covexity or reiforcemet causes stress cocetratio that creates problems uder impact, fatigue loads, or low-temperature service. Iadequate peetratio is a weld defect, which would leave a ufused area that could cause failure. The effect of process variables o weld geometry has bee studied by Targ ad Yag [1]. The weld profiles were measured usig the plaimeter. I order to study the correlatio betwee the weld geometry ad weld failure, Bowma ad Qui [2] ad Mashiri et al. [3] made duplicated molds from welds, ad the weld geometry was measured from a mold usig a profile projector. The weld measuremet by either the plaimeter or the profile projector is ot fully automated. Maual measuremet may be iaccurate due to huma errors. Wide applicatios have bee employed to measure the geometrical characteristics of fillig or sprayig process due to the techological advacemet of image processig. Qu et al. [4] used image processig to measure the deposit dimesio i spray formig. Kuo ad Wu [5] used fuzzy theory coupled with image processig to trac the welded seam ad to sed a appropriate cotrol sigal to the umerical cotrol machie. I this study, the image processig techology is employed to measure the weld geometry of bead-o-plate weld. 4-036
2. Image processig A image ca be represeted by two-dimesioal fuctio. The image fuctio f x, y represets the gray level value, ad x ad y represet space coordiates. The image cosists of m poits, where every poit is ow as a pixel. The f x, ydigitizatio value of a pixel is 8 represeted by a byte. f x, y 0 represets blac, while f x, y 2 1 255 represets white, ad the itermediate values of x y 2.1 Mas f, represet the graduated variatio from blac to white. The mas is a operator that obtais the ew value of fuctio for a pixel. A pixel ad its eighborig pixels are operated upo ad a ew value is calculated. After the pixel is operated upo by the mas, the value of the fuctio is the umerical value of this poit. 2.2 Filter If the image is oisy ad blurred, it is ecessary to use a filter to smooth the image prior processig. The smaller the pixel area chose, the less effective the filter will be, while, with a larger chose pixel area, the more effective the filter operatio will be, but the operatio will be more complex. (1). Low pass filter A image with a large variatio i the gray level value will result i high frequecy oise followig Fourier trasfer. If the high frequecy compoets ca be removed, the image will be smoothed. The mas operatio is ofte used for the smoothig of a image. This method reduces the high frequecy compoets ad smoothes the image. (2). High pass filter This filter is used to reiforce the part where the image varies greatly, ad to reduce the low frequecy compoet of the image. After treatmet by this filter, the outlie of the image will become clearer. (3). Media filter The filter is used to remove radom oise ad to maitai image sharpess. Let each poit i the pixel rage be arraged accordig to its umerical value. The media value will be obtaied. The gray level of each poit is replaced by the media of the gray levels i a eighborhood of that poit, istead of by the average. 2.3 LOG operator The edge of a image ca be efficietly detected usig LOG (Laplacia of Gaussia ) operator. The operator combied both low pass filter ad high pass filter. The low pass filter is Gaussia fuctio, which smooth the iput image. The high pass filter is Laplacia fuctio, which reiforces the outlies of a image. I smooth ( x, G( x, I ( x, (1) iput 2 2 1 ( x y ) G ( x, exp 2 2 (2) 2 2 4-037
Where * represets a covolutio operator ad is the spread parameter. I output 2 ( x, I ( x, (3) smooth The resolutio of a digit camera is gettig higher, hece the captured pixels of a image is growig larger. It is ecessary to implemet a larger mas to calculate LOG operator. A 11 11 mas with 2 2 is used to approximate LOG operator i this study [6]. The mas computes a respose by itegratig the iput of 121 pixels ad thus respods a larger image feature ad ot a smaller oe. 2.4 Zero-crossig detectio The zero-crossig detectio is to mar the poits where the result of LOG respose has zero value. Huertas ad Medioi [7] have developed a systematic method for classifyig 33patters i order to determie edge directio. A 33widow of LOG respose is compared with the zero-crossig patters as show i referece [7]. If the33widow of LOG respose is idetical to oe of zero-crossig patters, the the ceter poit of the widow is the edge poit. The gray level value of edge poit is 0, otherwise the gray level value is 255. 2.5 Histogram processig The histogram with sigificat spread correspods to a image with high cotrast. If the histogram of a image with gray levels i the rage [0, L-1] ca be described as a discrete probability fuctio of r equals, where r is the th gray level, is the umber of pixels i the image with that gray level, is the total umber of pixels i the image, ad 0,1,2,, L 1,the a uiform histogram of the image ca be obtaied usig histogram equalizatio as Equ.(4) [8]. The treatmet of histogram equalizatio ca spread the gray level rage ad thus ehace the edge detectio. Where 3. Results S K is the equalizatio gray level correspodig to th gray level. 3.1 Accuracy verificatio S K j0 I order to verify the measuremet accuracy, the evaluatio compariso has bee carried out betwee a group of cocetric circles raged from 2 to 10 mm radius with 1 mm itervals ad a referece circle of 3 mm radius. The images of both the cocetric ad referece circle acquired from SONY DSC-H5 digital camera (7.2M Pixel) were processed by LOG operator ad zero-crossig detectio. The ceter ad average radius R avg for each circle were calculated by the least square method. The legth of pixel was estimated by dividig R avg of the referece circle by the pixel umber o R avg. Similarly, the area of pixel was calculated by dividig area of the referece circle by the pixel umber withi the circle. Based o the pixel j (4) 4-038
legth ad area, the average radius, maximum radius R max, miimum radius R mi, rage radius R ra, area of comparative circle A ca be estimated. Table1 shows the geometric data of the evaluated circles, where A 0 equals 2 R ad σ is the stadard deviatio. 3.2 Weld geometry Table 1. Results of evaluated circles. A A R avg (mm) R max (mm) R mi (mm) R ra (mm) σ (mm) A(mm 2 ) 0 100% 3mm 3.00 3.02 2.98 0.04 0.016 28.27-2mm 1.99 2.01 1.97 0.04 0.015 12.44-1.0 6mm 6.01 6.03 5.99 0.04 0.018 113.48 0.33 10mm 10.02 10.05 10.00 0.05 0.023 315.42 0.40 The base material used for this study was a 9 mm-thic plate of low carbo steel. The plate was cut ito 120 65mm pieces ad both surfaces were sad blasted to remove dirt ad oxides. A CO 2 arc weldig machie was used to deposit experimetal bead-o-plate welds. The electrode diameter 1.2mm, weldig voltage 28V, weldig curret 200A ad weldig speed Item Radius 250 mm were used i the experimet. The weld was cross-sectioed at mid-legth, mi polished ad macro etched with 5% ital. A digital camera was used to acquire the image of a macro etched sectio of weld alog with a referece circle of 3 mm radius. The image of a weld cross sectio is show i Fig. 1. The image of referece circle was treated by LOG operator ad zero crossig detectio to obtai the outlie. The pixel legth ad area were estimated based o the pixel umber o the diameter ad withi the area of the circle, respectively. I order to decrease the oise disturbace of the weld image, a media filter was used to smooth the weld image. The LOG operator ad zero crossig detectio were the processed to obtai the edge of weld deposit. Due to the cotrast of gray level betwee the fusio area ad base material was isigificat, the histogram processig was used to ehace the juctio of fusio area. The edge of fusio area was obtaied by LOG operator ad zero crossig detectio. The edge of deposit ad fusio area after image processig is show i Fig. 2, where A d represets deposit area, A f represets fusio area, W is weld width, R is weld reiforcemet ad P is weld peetratio. Based o the estimated pixel legth ad area, the geometrical dimesios of wed ca be calculated: A d = 22.30 mm 2, A f = 23.13 mm 2, W= 14.94 mm, R= 2.67 mm, ad P= 2.77 mm. w A 0 Base material A d A f R P Fig. 1. Image of macro etched sectio of Bead-o-plate weld. Fig. 2. Edge of deposit ad fusio area. 4-039
4. Coclusio The macro etched sectio of weld ca reveal the weld deposit ad fusio juctio betwee weld ad base material. The deposit area is directly related to the load resistace. The fusio area is a idicatio of the compositio chage of weld material caused by mixig of base material ad related to joit peetratio. I this study, the image processig is used to measure the geometrical dimesios of bead-o-plate weld. The edges of deposit ad fusio area are obtaied by LOG operator ad zero-crossig detectio. The results show that the image processig ca effectively ad accurately measure the weld geometry. 5. Acowledgmets This wor was supported by the atioal sciece coucil of ROC uder cotract No. NSC-96-2221-E-224-051. The authors would lie to express their thas herewith. Refereces 1. Y.S. Targ, W.H. Yag. Optimisatio of the Weld Bead Geometry i Gas Tugste Arc Weldig by the Taguchi Method. It. J. Adv. Mauf. Techol. 1998, vol. 14, pp. 549-554. 2. M.D. Bowma, B.P. Qui. Fillet Weld Profile Measuremets. Experimetal Techiques. 1995, No. 5, pp. 21-24. 3. F.R. Mashiri, X.L. Zhao, P. Grudy. Effects of Weld Profile ad Udercut o Fatigue Crac Propagatio Life of Thi-Walled Cruciform Joit. Thi-Walled Structures. 2001, vol. 39, pp. 261-285. 4. Yig-dog Qu, Cheg-sog Cui, Sha-be Che, Qig-chu Li. O-lie Measuremet of Deposit Dimesio i Spray Formig Usig Image Processig Techology. Joural of Materials Processig Techology. 2006, vol. 172, pp. 195-201. 5. Hsig-Chia Kuo, Li-Je Wu. A Image Tracig System for Welded Seams Usig Fuzzy Logic. Joural of Materials Processig Techology. 2002, vol. 120, pp. 169-185. 6. Lida G. Shapiro, George C. Stocma. Computer Visio. Pretice-Hall, Upper Saddle River, New Jersey, 2001. 7. A. Huertas, G. Medioi. Detectio of Itesity Chages Usig Laplacia-Gaussia Mass. IEEE Tras. Patter Aalysis ad Machie Itelligece (PAMI-8). 1986, 5, pp. 651-664. 8. Rafael C. Gozalez, Richard E. Woods. Digital Image Processig. Addiso-Wesley Publishig Compay, New Yor. 1993. 4-040