Laser VisionBased Detection of Weld Penetration in GTAW · the welding process to prevent certain defects (Ref. 31). However, for real-time detection during welding, the ability of

Introduction Gas tungsten arc welding (GTAW) isa major arc process for precision join-ing in reactor pressure vessels, nuclearpower plants, and other industries dueto its ability to produce high-qualitywelds (Ref. 1). However, it is conven-tionally performed by skilled humanwelders who can adapt to the variedwelding conditions and control weldquality timely and correctly. As laborcosts and a shortage of skilled welderscontinuously increase, automatedGTAW becomes urgent (Ref. 2). Unfor-tunately, in the absence of humanadaptation, varying or inconsistentconditions such as in joint geometry,root opening, and mismatch will likely

cause defects when using welding pa-rameters predetermined from nominalconditions. These defects include in-complete joint weld penetration, ex-cessive weld penetration, weld poolcollapse, and porosity. Repairing is notpreferred because of the high costs aswell as the increased production time.Equipping machines/robots withadaptation abilities appears to offer abetter solution but presents technicalchallenges. Adaptation ability depends on theavailability of methods that can detectpossible weld defects. They can espe-cially detect the weld penetration tomeet the minimal requirement forweld integrity. Possible methods in-clude the following: machine vision

(Refs. 3–7), radiography (Refs. 8, 9),infrared sensing (Refs. 10–12), ultra-sonic (Refs. 13–16), magnetic detec-tion (Ref. 17), acoustic measurements(Refs. 18, 19), and weld pool oscilla-tion (Refs. 20–23). Nacereddine et al.(Ref. 24) used the radiographic sensorto automatically detect the weld de-fects, which was considered as a non-destructive examination method. However, radiographic sensors can-not be installed on weld lines to sup-port online detection. Ultrasonicmethod lacks a visual record and re-quires highly skilled operators to per-form defect recognition, although it isbeing widely implemented due to itsflexibility and lower cost (Ref. 14).Carvalhoa et al. (Ref. 25) used themagnetic testing method to inspectpipeline weld defects and recognizethe defect pattern. However, thismethod was limited to ferromagneticmaterials for off-line detection. Alfaroet al. (Ref. 26) utilized the infraredsensor to measure weld penetration byreal-time monitoring the weld pooltemperature, but the temperature gra-dient of the weld pool was extremelyhigh. To use the measurement to de-rive the weld penetration, the pointbeing measured must be accurately po-sitioned. In practice, an error can beexpected under the varying condi-tions. Tao et al. (Ref. 27) investigatedacoustic emissions to detect weldingporosity and incomplete joint weldpenetration using the amplitude andcentroid frequency of acoustic emis-sion. This method requires noise re-duction as well as an efficient and fastdata processing system when weldingdefects are encountered. Recently, Zhang et al. (Ref. 28) pro-

WELDING RESEARCH

Laser VisionBased Detection ofWeld Penetration in GTAW

An inspection method was proposed to monitor weld penetrationfor investigating the control of weld defects

BY G. ZHANG, Y. SHI, Y. GU, D. FAN, AND M. ZHU

ABSTRACT In automatic gas tungsten arc welding (GTAW), incomplete joint weld penetration, excessive weld penetration, or weld pool collapse may occur due to changes in weldingconditions or weld structures so that the required weld integrity cannot be assured. Inorder to realtime monitor the weld penetration to prevent these defects, an innovativemethod was proposed that detects the dynamic change of the reflection image area(RIA) for the laser pattern reflected by the specular weld pool surface. The dynamicchange in the RIA was found to be determined by the weld pool surface, which was inturn determined by the weld penetration. Because of this physics mechanism, the RIAdemonstrated two peaks that corresponded to the occurrences of the complete jointpenetration and excessive penetration, respectively. This unique characteristic eases thedetermination of the weld penetration status based on a simple analysis of the onedimensional RIA signal. Further, in comparison with the computation of the 3D surfaceof the weld pool, the RIA was computed simply by image binarization. The proposed

method is not only innovative and physics based but also easy to implement online inreal time. Experiments and comparative analyses from highspeed and backside images,weld beads, and RIA signals verified the physics mechanism and the effectiveness of theproposed method.

KEYWORDS • Gas Tungsten Arc Welding (GTAW) • Weld Penetration • Laser Vision • Image Processing • Weld Pool Surface

MAY 2017 / WELDING JOURNAL 163-s

ZHANG SUPP MAY 2017 WJ 201689_Layout 1 4/6/17 4:09 PM Page 163

posed an alternative method to senseweld penetration based on arc voltage,and obtained the step-by-step pulsedgas tungsten arc welding (GTAW-P)complete joint weld penetration. Yooet al. (Ref. 29) had monitored the in-tensity of arc light caused by pool os-cillation and then ensured the weldpenetration status in step-by-stepGTAW. Shi et al. (Ref. 30) had pro-posed a new method to detect pool os-cillation frequency, and identified theweld penetration. However, it is diffi-cult to separate the characteristic sig-nal from the small variations in the arcvoltage, which can be easily disturbed.Those methods can only detect thepartial and complete joint penetration. Vision sensing methods have great-ly improved our abilities to monitorthe welding process to prevent certaindefects (Ref. 31). However, for real-time detection during welding, theability of conventional vision methodis affected by the strong arc light andthe specular weld pool surface. Theweld pool surface that can reflect welddefects cannot be clearly imaged. Fur-thermore, the process to extract fea-ture signals of weld defects from theweld pool images is complex and time-consuming. To overcome such short-comings, an innovative inspectionmethod was proposed in this work todetect the weld penetration status toprevent possible incomplete joint weldpenetration, excessive weld penetra-tion, or weld pool collapse. The objective of this study was to es-tablish the experimental system and ac-quire the characteristic signal from thereflection laser images by the developedimage processing algorithm, which rep-resents the weld penetration status.Verification experiments were then con-ducted, and the correlation between thecharacteristic signal and the weld pene-tration status was derived.

Proposed Method

The proposed laser vision-basedmethod monitors the change in the re-flection laser stripes on the imagingplane that reflects the variation of theweld pool surface to determine/esti-mate the weld penetration status frompartial, complete joint to excessive, orpool collapse. The principle can be il-lustrated in Fig. 1. To this end, Kovacevic et al. (Ref.

32) first proposed the basic method,and it was developed by the Universityof Kentucky’s Zhang and Saeed et al.(Refs. 33, 34) as well as Song et al.(Refs. 35, 36) to monitor the weld poolsurface. In their method, a dot matrixlaser pattern was projected onto theweld pool surface, which was specularand reflected as a mirror. The reflec-tion of the projected laser patternfrom the weld pool surface was used tocompute the 3D weld pool surface andestimate the weld penetration as wellas control it (Refs. 37–41). However,the computation for the weld pool sur-face was relatively complex. The novel-ty of our proposed method lies in cor-relating the reflection characteristic tothe status of the weld penetration —partial, complete joint, excessive, orweld pool collapse — without actuallycomputing the 3D weld pool surface.As a result, the status of the weld pen-etration can be determined rapidly in

real time. Figure 1 depicts the correlation ofthe change in reflection laser imagesand the weld pool surface as well as theweld penetration in stationary GTAW.In this case, five laser stripes were pro-jected onto the weld pool surface. Theirreflections from the weld pool surfacegathered as the weld pool surface con-cavity and its back-side width increasedwhen the penetration status changedfrom partial to complete joint and ex-cessive. However, as the back-sidewidth of the weld pool further in-creased to the level comparable withthe top-side width of the weld pool, anexcessive weld penetration level, thegathered laser stripes on the imagestarted to disperse with a larger curva-ture because of the more concave poolsurface. Hence, the number of the pix-els of the reflection laser stripes in agiven area on the image varies accord-ingly during the penetration develop-

WELDING RESEARCH

WELDING JOURNAL / MAY 2017, VOL. 96164-s

Fig. 1 — Schematic diagram of the correlation between the characteristic of reflectionlaser stripes, weld pool surface, and weld penetration in stationary GTAW.

A B

C D


ment process as the reflection changesfrom gathering to dispersing. To simplyand quickly achieve the change in thelaser stripe characteristics, the sum ofthe gray value of the pixels (SGVP) in agiven area on the reflection image wascomputed and used to reflect thechange in the reflection, and thechange in the weld penetration status.It had been demonstrated that the con-vexity of the weld pool surface closelycorrelated with the back-side width ofthe weld pool (Refs. 40–42).

ExperimentalExperimental System

Figure 2 is a schematic diagram ofthe system that real-time monitors thelaser stripes reflected by the weld poolsurface. The system includes a struc-tured-light laser generator, an imagingplane, two GZL-CL-22C5M-C high-speed cameras, and cameras 1 and 2produced by Point Grey Research Co.While camera 1 captured the reflec-tion laser stripes, camera 2 capturedthe back-side surface of the workpieceto observe the actual weld penetrationstatus. A 50-mW Stocker Yale’sLasirisTM SNF continuous illuminationlaser with variable focus was utilizedto generate the five-line laser pattern.

The imaging plane (a thin glass with asheet of white paper attached) wasused to intercept/image the reflectionlaser stripes. The corresponding high-speed camera can take 60 to 1800 pic-tures at one second. The direct currentelectrode negative (DCEN) GTAWwithout filler metal was used to per-form bead-on-plate experiments ontype 304 stainless steel sheet with di-mensions 100 50 2 mm. The welding direction is the negativey axis in Fig. 2. The welding torch wasperpendicularly installed and kept sta-tionary during the welding processwhile the workpiece was moving. Thetravel speed was controlled by a com-puter-assisted servomotor. The weldingtorch, laser generator, and imagingplane were carefully aligned. To reducethe influence of strong arc light, thehigh-speed camera was fitted with acomposite filter lens that most of theinterference of arc radiation, which wasdistributed along the entire visiblewavelength spectrum, was effectivelyfiltered out. The sampling frequency oftwo cameras were the same at 200 Hz.Pure argon was used as the shielding gasat a flow rate of 10 L/min. The systemwas programmed to simultaneouslycapture the weld pool reflection andworkpiece back-side images as well ascontrol the workpiece movement.

Image Processing Algorithm

A novel, specific image processing al-gorithm that can quickly address thespecific characteristics of the reflectionlaser images and the particular informa-tion of our concern was developed. Figure 3 shows the principle/flow-chart of the proposed algorithm con-sisting of the following: image prepro-cessing, selecting the region of inter-est (ROI), and identifying and record-ing the two characteristic peaks thatare to be detailed in later experiments.In the image preprocessing section,linear grayness conversion was utilizedto enhance the contrast of the reflec-tion laser stripes from the back-ground. A homomorphic filter wasused to reduce/eliminate the effect ofthe noises produced in the backgroundsuch as signal transmission, experi-ment surroundings, and strong arclight. Because the change of the ROIwas represented by the SGVP definedin the experimental system section,the reflection laser image was furtherbinarized to easily calculate the grayvalue. The selection of the ROI (a rec-tangular window) was crucial for effec-tively processing the image to extractthe relevant information. While a larger rectangular window(e.g., rectangle 1 in Fig. 4) increases

WELDING RESEARCH


Fig. 2 — Schematic diagram of an experimental system. Fig. 3 — Flow chart of the image processing algorithm.


the processing time, a too small win-dow (e.g., rectangle 3 in Fig. 4) wouldlose the characteristic information re-flecting the variation in the weld pool.Hence, in this work, a special algo-rithm was developed to adaptively se-lect the ROI, as shown in Fig. 5. As canbe seen, by finding the upmost, lower-most, leftmost, and rightmost brightpoints, the ROI of a rectangle can beadaptively defined. Furthermore, therunning time of this algorithm was theshortest, which will be demonstratedby later experimental data. In particular, as shown in Fig. 5, aftersearching and ensuring the four charac-teristic points, the rectangular windowwould be acquired and its four sidesshould be located in the above recordedpoint respectively, such as rectangle 2 inFig. 4. Thus, this rectangular windowcan encompass all the laser stripes that

were reflected by the weld pool surfaceand imaged on the plane. Compared with the algorithm pro-posed in Ref. 30, which was utilized toextract the weld pool oscillation fre-quency in GTAW, this algorithmprocessed the images simply andavoided the errors caused by the devia-tion that the laser stripes were reflect-ed on the relative side of the imageplane. In addition, this algorithm over-comes the blindness of first selectingthe rectangular window for correctingthe reflection laser image.

Experimental Resultsand Discussion

A series of stationary and travelingGTAW experiments have been con-ducted to verify the effectiveness and

feasibility of the proposed weld pene-tration monitoring method and imageprocessing algorithm. The parametersfor the six representative experimentscorresponding to different cases to bediscussed in detail are listed in Table 1.Meanwhile, the verified back-side im-age of the weld pool was captured tocalculate the times when completejoint weld penetration and excessiveweld penetration occurred.

Stationary GTAW Experiment

The images shown in Fig. 6 (weld-ing current was 70 A) start at the 9th sduring the experiment. The laserstripes reflected by the weld pool sur-face are shown in Fig. 6A while theworkpiece back-side images are shownin Fig. 6B. Figure 6A gives three apparent ob-

WELDING RESEARCH


Fig. 4 — Schematic diagram of the ROI selection. Fig. 5 — Flow chart of the adaptive selection of the ROI.

Table 1 — Welding Parameters

Welding and Laser Parameters

Welding Distance from the Distance from the Center of the Angle of the Laser Arc Length/ Welding Traveling Condition Weld Tungsten Axis to Laser Generator to the Center Projected to the mm Current/A Speed/mms–1

the Imaging Plane/mm of the Weld Pool Surface/mm Weld Pool Surface (deg)

Stationary 50 40 30 5 70 0Experiment 50 40 30 5 75 0 50 40 30 5 65 0 Traveling 50 40 30 5 70 1.3~1.0~0.8Experiment 50 40 30 5 67 2.5~2.3~2.0 50 40 30 5 63 2.3~2.0~1.7


servations: 1) the number of the re-flected stripes that were imaged in-creased from three to four in frame(c1); 2) the reflected stripes gatheredand changed to a spot in frame (e1)and then were separated in frame (h1)and clearly separated in (i1); 3) thenumber of the identifiable reflectedstripes became five in frame (i1). While the projected laser has fivestripes, only those projected on theweld pool surface were specularly re-flected and imaged. Observation 1clearly suggests that the weld pool sur-face had been increased from the timeof frame (a1) (t = 9 s) to interceptthree stripes to the time of frame (c1)(t = 10 s) to intercept four stripes. What is interesting is that, whilethe weld pool surface has been in-creased and more stripes were inter-cepted as explained, the image regionof the reflection laser stripes was de-creased. When the incident light wasunchanged, the reflected pattern wasonly determined by the shape of thereflection mirror, which was the weldpool surface in this case. This clearlysuggests that the reflection mirror in-

creased its concavity degree. Beforethe back-side of the workpiece be-comes free surface, the liquid metal inthe weld pool, due to the increasedvolume because of the thermal expan-sion, could only elevate from the work-piece. The weld pool surface must beconvex unless the arc force creates alocal concavity. The weld pool surfacemay become globally concave only af-ter the back-side surface of the work-piece becomes a free surface that is nolonger able to constrict the weld poolmetal. Hence, the contraction of thereflection laser stripes not only re-flects that the weld penetration be-came complete, such that the back-side surface was free, but also that theconcavity degree or the back-side freesurface increased, i.e., the weld pene-tration increased. When the concavity degree furtherincreased, the focal length of the con-cave reflection mirror (weld pool sur-face) reduced to deviate the focal pointfrom the imaging plane toward themirror. While the light (reflection ofthe laser stripes) converged before thefocus point, it diverged after the focal

point. Hence, the divergence of the re-flection, from frame (g1) to (j1), re-flects the change of the focal point tobecome more deviated toward the re-flection mirror or the increase of theconcavity degree. The increased con-cavity degree was caused by the in-creased back-side free surface, i.e., theincreased weld penetration. Hence, thedivergence of the reflection fromframe (g1) to (j1) reflects the increasein the weld penetration. The back-side images in Fig. 6B,captured from camera 2, which wasplaced in the yoz plane (shown in Fig.2) at 30 deg with the z axis from be-hind, show the variation of back-sidehigh-temperature region of the work-piece. As the weld penetration in-creased, the variation of the high-tem-perature region also extended. Whenthe workpiece is just penetrated, onlya little liquid weld metal may exceedbeyond the bottom of the workpiece,as can be seen in Fig. 6B frame (d2). In the zoomed-in images shown inFig. 6C, the red and green lines, re-spectively, represent the high-temper-ature region edge of weld metal andthe edge of liquid metal exceeding theworkpiece bottom surface that wasalso the back-side surface edge of theweld pool. Figure 6C frame (e2) showsthat parameters h and w were utilizedto represent the height and width ofthe liquid metal exceeding the bottomof the workpiece, respectively. The ex-ceeding liquid metal increased as thefree surface on the bottom of theworkpiece increased, i.e., as the weldpenetration increased. From thezoomed-in frames (d2~f2), it was easi-ly found that h and w both increasedas the weld penetration increased.This is understandable because whenthe arc force, which tends to press the

WELDING RESEARCH


Fig. 6 — Sampled images in the stationary experiment. Starting at the 9th s with the sampling interval 0.5 s.

A

B

C


liquid metal beyond the bottom of theworkpiece (if it has become free liquidsurface), remains the same, an in-creased free bottom surface, i.e., an in-creased penetration, an increasedgravity, will allow the liquid metal tomove further beyond the bottom ofthe workpiece. An easy way was needed to conve-niently and accurately detect thechanges of the weld penetration statusfor real-time control. To this end, weproposed to compute the area of thereflected laser on the image. We haveproposed, in the image processing al-gorithm section, to binarize the imageto separate the laser reflection fromthe background such that the SGVPcan be used to measure the total areaon the image that intercepted the re-flected laser. This area, referred to asthe reflection image area (RIA) in this

paper, is suggested to first increase asthe weld pool surface increases suchthat more laser stripes are reflected,and then decrease as the reflectionmirror (weld pool surface) changesfrom convex to concave such that thereflection becomes from divergent toconvergent, as the weld penetrationincreases. After the mirror surface becameconcave, the reflection became conver-gent and the focal point moved towardthe mirror as the concavity degree in-creased (as the weld penetration in-creased). When the focal point movedto the imaging plane, all the laser re-flection converged to form a verybright area. While this area was verybright, its area became minimum be-cause of the best focus. Then, as theweld penetration increased such thatthe concavity degree further increased,

the focal point moved away from theimaging plane when moving towardthe reflection mirror. The laser reflec-tion diverged again but from the focalpoint. Because of the divergence, theRIA increased as the focal pointsmoved toward the weld pool surfacewhen the weld penetration increased.After the concavity increases to a de-gree such that the concavity becomesdeep, some reflection will be blockedby the workpiece such that the RIAwill decrease. As such, we expect thatthe RIA will demonstrate a twin peakscharacteristic during the completejoint development process of the weldpenetration. For the stationary experi-ment designed to examine this entireprocess, the authors expect to observethis phenomenon. Figure 7A shows the change of theRIA during the stationary experiment.

WELDING RESEARCH


Fig. 7 — The change of the RIA in the stationary experiment (welding current was 70 A): A — Original; B — fivepoint smooth filtered.

A B

Fig. 8 — The change of RIA in the stationary experiment (welding current was 75 A): A — Original; B — fivepoint smooth filtered.

A B


The adaptive ROI selection, proposed inthe image processing algorithm section,was used to obtain the ROI withinwhich the binarization was performedand the sum of gray value of pixels wascounted as the area for the RIA plottedin Fig. 7A. Figure 7B shows a filteredRIA during the experiment. It was clearthat the twin peaks characteristic in-deed exists in the penetration develop-ment process as analyzed above. To convincingly show this inherentcharacteristic of the RIA with varyingweld penetration, results from two ad-ditional experiments with a differentwelding current are shown in Figs. 8and 9. Observations of Figs. 8B and 9Bshowed the twin peaks were clearlyseen in the welding process as the weldpenetration increased despite differ-ent welding currents that generateddifferent arc forces acting on the weldpool and produced diverse weld poolgeometry. In addition, these two peakswere easily distinguished from the en-tire change of the RIA. Hence, thischaracteristic of the RIA is inherentfor the penetration developmentprocess, independent from the adap-tive ROI selection that only reducedthe computation to improve the real-time ability to implement, and can be

used to monitor the status of weldpenetration in stationary welding.

Traveling GTAW Experiment

In stationary GTAW, the fluid flowstatus and weld pool surface geometrywere relatively stable such that the laserstripes may be reflected and interceptedwith desirable regularity. The RIA wasalso easy to obtain. Verification experi-ments have also been conducted to veri-fy the phenomenon observed earlierfrom the stationary experiment, as wellas our observations and analyses as theeffectiveness of the proposed imageprocessing algorithm. Now the questionis if all these will hold for a movingwelding process where the weld metalfluid flow and the weld pool surfacemay be less stable and more complex.To answer this question, experimentswith varying travel speed have beenconducted using the parameters givenin Table 1. Because of the movement,the images can be mapped, in time, tothe back-side weld bead that bettershowed the change of the weld penetra-tion than the back-side surface image. In the moving experiment with 70 Awelding current, the travel speed wasfirst set at 1.3 mm/s to run 4.6 s, and

then decreased to 1.0 mm/s to run 2 s,and finally decreased to 0.8 mm/s tolast to the end of the experiment in or-der to obtain an excessive weld penetra-tion or weld pool collapse. The imagespresented in Fig. 10 start 1 s before thespeed is changed to 1.0 mm/s. Figure 11A plots the change of theRIA characteristic on the reflectionlaser image with the time. Figure 11Bgives the weld bead on both sides withthe first vertical line representingwhere the tungsten axis (arc center)was when the welding started, the sec-ond vertical line representing wherethe speed was changed to 1.0 mm/s,and the third line representing wherethe speed was changed to 0.8 mm/s. When the travelling speed is 1.3mm/s (frame (a) and to (b) in Fig. 10,first second in Fig. 11A, and the seg-ment before the first and second verti-cal lines from the left), the weld pene-tration was partial. After the speedwas decreased to 1.0 mm/s, the pene-tration changed from partial to com-plete joint (between the second andthird vertical lines from the left in Fig.11B and the period from t = 2 to 4 s inFig. 11A, and frame (c) to (f) in Fig.10). This can be easily seen from theback-side weld bead in Fig. 11B. Figure11A suggests that the first peak of the

WELDING RESEARCH


Fig. 9 — The change of RIA in the stationary experiment (welding current was 65 A): A — Original; B — fivepoint smooth filtered.

Fig. 10 — Reflection image series during the weld penetration establishment process with a moving workpiece. The image sampling interval was 0.5 s.

A B

A B C D E F G H I J


RIA, at which the convexity of theweld pool surface reaches its maxi-mum, occurs less than 1.0 s after thespeed decreases. This should be themoment at which the complete jointpenetration was sufficiently estab-lished, such that the free surface onthe bottom can allow the liquid metalto move beyond the bottom surface tooff-set the increase in the volumethermal expansion. The complete jointpenetration must have been estab-lished before this moment. As can be seen from the back-sideweld bead in Fig. 11B, the completejoint penetration indeed happens inthe first half of this 2-s period. In Fig.10, the laser reflection is more concen-trated in frame (d) than that in frame(c). This again suggests that at thetime of frame (d), the convexity of thereflection mirror has been reducedfrom that at the time of frame (c). Atthe time of frame (f), the laser reflec-tion has been well converged. This im-

plies that the focal point is at the im-aging plane. The concavity of the weldpool surface was large but not too ex-cessive to block the laser reflection.The weld penetration should be signif-icant but not yet excessive. This can beverified from the back-side weld beadin Fig. 11B. During the welding period withspeed at 0.8 mm/s, the back-side weldpool shape between the third verticalline and the rightmost vertical line inFig. 11B and the typical frame (g) ofFig. 10, combining the change of theSGVP in the RIA in Fig. 11A (t > 3 s)showed that the weld penetration fur-ther increased. This implies that theconcavity will further increase, and itbecomes possible that the focal pointmoves away from the imaging planetoward the weld pool surface such thatthe reflection on the image starts todiverge. This is clear in Fig. 10 wherean increasing reflection divergence canbe observed after frame (h). As such,

one can expect that the second peak ofthe RIA should occur before the timeof frame (h), i.e., less than 1 s after thespeed is decreased to 0.8 mm. Figure 11A shows that the secondpeak occurs approximately 0.6 s afterthe speed decreases at t = 3 s in Fig.11B. Of course, this further increase ofthe concavity was a result of the fur-ther increase in the weld penetration.As can be seen from Fig. 11B, the pen-etration indeed becomes larger on theright of the last (from left) of the ver-tical line. An excessive weld penetra-tion occurred but no melt-through occurred. The welding current and weldingspeed significantly affected the arc stability and the liquid metal fluidflow, which, in turn, determined thegeometry of weld pool surface and thecharacteristics of the reflection laserstripes. Hence, to test if this character-istic of the reflection laser stripes wasalso produced in different traveling

WELDING RESEARCH


Fig. 11 — The traveling experimental result (speed 1.3~1.0~0.8 mm/s): A — The change of RIA; B — weld bead.

Fig. 12 — The traveling experimental result (speed 2.5~2.3~2.0mm/s): A — The change of RIA; B — weld bead.

A

A

B

B


conditions, two traveling experimentsperformed with different welding cur-rent and welding speed were selected,and the corresponding data were de-tailed. The results can be shown inFigs. 12 and 13, respectively. From Figs. 12A and 13A, it is clear-ly observed that the twin peaks weredetected from the reflection laserstripes as the penetration increasedfrom partial to excessive as shown inFigs. 12B and 13B. As such, the phe-nomena observed in the stationaryweld pool development were also ob-served at the moving weld pool.

Conclusions In this paper, an innovative inspec-tion method was first proposed tomonitor the weld penetration for in-vestigating the control of weld defects.The following conclusions can bedrawn: 1) A novel laser-vision basedmethod has been proposed to real-time detect the change of the weldpool surface in GTAW. Experimentsshow that the change of the reflectionimage area (RIA) for the projectedlaser lines reflected by the pool surfaceenables representing the weld penetra-tion status from partial, completejoint, and excessive to weld pool col-lapse in the stationary and travellingcondition. 2) A new image processing algo-rithm had been developed and suc-cessfully applied to quickly process theseries of intercepted images and ob-tain the change of the RIA characteris-tic. Twin peaks illustrated in the RIA

during the penetration developmentprocess provided a robust method toanalyze for the occurrences of criticaladequate penetration and excessivepenetration. 3) Stationary and traveling experi-ments together with the comparativecross analyses of different data sourcesverified the effectiveness of the pro-posed method as a unique novel weldpenetration monitoring approach forits principle, physics mechanism, andimage processing algorithm.

This work was funded by the Nation-al Natural Science Foundation of China(#61365011), Young Creative TalentSupport Program of Long Yuan of Chi-na and Hong Liu Outstanding TalentTraining Plan of Lanzhou University ofTechnology (#J201201), State Key Lab-oratory of Advanced Processing and Re-cycling of Nonferrous Metals of China(#SKLAB02015008), and Natural Sci-ence Foundation of Gansu Province ofChina (#1508RJZA070).

1. Ravisankar, A., Velaga, S. K., Rajput,G., and Venugopal, S. 2014. Influence ofwelding speed and power on residual stressduring gas tungsten arc welding (GTAW) ofthin sections with constant heat input: Astudy using numerical simulation and ex-perimental validation. Journal of Manufac-turing Processes 16(2): 200–211. 2. Kah, P., Shrestha, M., Hiltunen, E., andMartikainen, J. 2015. Robotic arc weldingsensors and programming in industrial ap-

plications. International Journal of Mechanicaland Materials Engineering 13: 1–16. 3. Kovacevic, R., Zhang, Y. M., andRuan, S. 1995. Sensing and control of weldpool geometry for automated GTA welding.Journal of Engineering for Industry 117(2):210–222. 4. Kovacevic, R., Zhang, Y. M., and Li, L.1996. Monitoring of weld joint penetra-tion based on weld pool geometrical ap-pearance. Welding Journal 75(10): 317-s to329-s. 5. Chen, S. B., Zhao, D. B., Lou, Y. J.,and Wu, L. 2004. Computer vision sensingand intelligent control of welding pool dy-namics. Robot Weld Intelligent AutomationLNCIS 229: 25–55. 6. Ario, S., Baskoro, R. M., and Masashi,K. 2009. Welding penetration control foraluminium pipe welding using omni-direc-tional vision-based monitoring of moltenpool. Jointing Conference 27(2): 17–21. 7. Huang, W., and Kovacevic, R. 2012.Development of a real-time laser-basedmachine vision system to monitor andcontrol welding processes. InternationalJournal of Advanced Manufacturing Technol-ogy 63(1–4): 235–248. 8. Guu, A. C., and Rokhlin, S. I. 1989.Computerized radiographic weld penetra-tion control with feedback on weld pool de-pression. Materials Evaluation 47: 10–14. 9. Rokhlin, S. I., and Guu, A. C. 1990.Computerized radiographic sensing andcontrol of an arc welding process. MaterialsEvaluation 69: 3–8. 10. Beardsley, H., Zhang, Y. M., and Ko-vacevic, R. 1994. Infrared sensing of fullpenetration state in gas tungsten arc weld-ing. International Journal of Machine Toolsand Manufacture 34(8): 1079–1090. 11. Chen, W. H., and Chin, B. A. 1990.Monitoring joint penetration using in-frared sensing techniques. Welding Journal69(4): 181-s to 185-s. 12. Vasudevan, M., Chandrasekhar, N.,Maduraimuthu, V., Bhaduri, A. K., and Raj,

WELDING RESEARCH


Fig. 13 — The traveling experimental result (speed 2.3~2.0~1.7 mm/s): A — The change of RIA; B — weld bead.

A B

Acknowledgments

References


B. 2011. Real-time monitoring of weld poolduring GTAW using infra-red thermogra-phy and analysis of infra-red thermal im-ages. Welding in the World 55(7): 83–89. 13. Siores, E. 1988. Development of areal-time ultrasonic sensing system for auto-mated and robotic welding. PhD disserta-tion. West London, UK, Brunel University. 14. Schuster, G. J., Doctor, S. R., andBond, L. J. 2004. A system for high-resolu-tion, non-destructive, ultrasonic imagingof weld grains. IEEE Transactions on Instru-mentation and Measurement 53: 1526–1532. 15. Carlson, N. M., and Johnson, J. A.1988. Ultrasonic sensing of weld pool pen-etration. Welding Journal 67(11): 11-s to16-s. 16. Mizota, H., Nagashima, Y., andObana, T. 2015. Fundamental study ofmolten pool depth measurement methodusing an ultrasonic phased array system.Japanese Journal of Applied Physics 54(7S1):07HC03-1-4. 17. Shilov, A. V., Kushner, A. V., andNovikov, V. A. 2016. The detection of realdefects in ferromagnetic objects using amagnetic-field imaging film. Russian Jour-nal of Non-destructive Testing 52(4): 220–225. 18. Pal, K., and Pal, S. K. 2011. Moni-toring of weld penetration using arcacoustics. Materials and ManufacturingProcesses 26: 684–693. 19. Lv, Na., Xu, Y. L., Zhong, J. Y., Chen,H. B., Wang, J. F., and Chen, S. B. 2012. Re-search on detection of welding penetrationstate during robotic GTAW process basedon audible arc sound. Industrial Robot: AnInternational Journal 40(5): 474–493. 20. Aendenroomer, A. J. R., and Ouden,G. D. 1998. Weld pool oscillation as a toolfor penetration sensing during pulsed GTAwelding. Welding Journal 77(5): 181-s to187-s. 21. Ju, J. B., Hiroyuki, H., and Suga, A.2004. Oscillation of molten pool by pulsedassist gas oscillating method and penetra-tion control using peculiar frequency. Jour-nal of High Temperature Society 30(5):263–269. 22. Wang, Q. L., and Lin, S. B. 1988. De-tection of the weld full penetration signals

of a traveling arc in GTAW process. Process-ing of JIW Asian Pacific Regional WeldingCongress, Hobart, Australia. 23. Wang, Q., Geng, Z., and Yin, S.1991. Real-time full penetration controlwith sensor in the TIG welding of alu-minum alloy. Proceedings of InternationalConference Joining: 9–16. 24. Nacereddine, N., Hamami, L., Tridi,M., and Oucief, N. 2007. Non-parametrichistogram-based thresholding methods forweld defect detection in radiography. IEEEWord Academy Science, Engineering andTechnology 1(9): 1237–1241. 25. Carvalhoa, A. A., Silvab, R. R., Re-belloc, J. M. A., and Sagrilo, L. V. S. 2010.Pattern recognition techniques applied tothe detection and classification of weldingdefects by magnetic testing. Science andTechnology of Welding Joining 21(2):91–111. 26. Alfaro, S. C., and Franco, A. F. D.2010. Exploring infrared sensing for realtime welding defects monitoring in GTAW.Sensors 10: 5962–5974. 27. Tao, Y., Wang, W., and Sun, B. S.2014. Non-destructive online detection ofwelding defects in track crane boom usingacoustic emission technique. Advances inMechanical Engineering 6: ArticleID.505464. 28. Zhang, S. Q., Hu, S. S., and Wang, Z.J. 2016. Weld penetration sensing inpulsed gas tungsten arc welding based onarc voltage. Journal of Materials ProcessingTechnology 229: 520–527. 29. Yoo, C. D., and Richardson, R. W.1993. An experimental study on sensitivityand signal characteristics of weld pool os-cillation. Transaction of the Japan WeldingSociety 24(2): 140–148. 30. Shi, Y., Zhang, G., Ma, X. J., Gu, Y.F., Huang, J. K., and Fan, D. 2015. Laser-vi-sion based measurement and analysis ofweld pool oscillation frequency in GTAW-P.Welding Journal 94(5): 176-s to 187-s. 31. Li, Y., Li, Y. F., Wang, Q. L., Xu, D.,and Tan, M. 2010. Measurement and de-fect detection of the weld bead based ononline vision inspection. IEEE Transactionon Instrumentation Measurement 59(7):1841–1849. 32. Kovacevic, R., and Zhang, Y. M.

1996. Sensing free surface of arc weld poolusing specular reflection: Principle andanalysis. Journal of Engineering Manufac-ture 210(62): 553–564. 33. Saeed, G., and Zhang, Y. M. 2003.Mathematical formulation and simulationof specular reflection based measurementsystem for gas tungsten arc weld pool sur-face. Measurement Science and Technology14: 1671–1682. 34. Zhang, Y. M., Song, H. S., andSaeed, G. 2006. Observation of a dynamicspecular weld pool surface. MeasurementScience and Technology 17: L9–L12. 35. Song, H. S., and Zhang, Y. M. 2008.Measurement and analysis of three-dimen-sional specular gas tungsten arc weld poolsurface. Welding Journal 87(4): 85-s to 95-s. 36. Song, H. S., and Zhang, Y. M. 2007.Image processing for measurement of three-dimensional GTAW weld pool surface. Weld-ing Journal 86(10): 323-s to 330-s. 37. Liu, Y. K., and Zhang, Y. M. 2015.Controlling 3D weld pool surface by adjust-ing welding speed. Welding Journal 94(4):125-s to 134-s. 38. Wang, Z. Z., Huang, X. Y., Yang, R.G., and Zhang, Y. M. 2013. Measurementof mirror surfaces using specular reflectionand analytical computation. Machine Visionand Applications 24: 289–304. 39. Zhang, W. J., Liu, X. K., and Wang,X. 2012. Characterization of three dimen-sional weld pool surface in GTAW. WeldingJournal 91(7): 195-s to 203-s. 40. Liu, Y. K., Zhang W. J., and Zhang,Y. M. 2013. Estimation of weld joint pene-tration under varying GTA pools. WeldingJournal 92(12): 314-s to 321-s. 41. Wang, X. W., Liu, Y. K., Zhang, W. J.,and Zhang, Y. M. 2012. Estimation of weldpenetration using parameterized three-di-mensional weld pool surface in gas tung-sten arc welding. IEEE International Sympo-sium on Industrial Electronics (ISIE) 12(9):835–840. 42. Zhang, G., Shi, Y., Li, C. K., Huang,J. K., and Fan, D. 2014. Research on thecorrelation between the status of three di-mensional weld pool surface and weld pen-etration in TIG welding. Acta MetallurgicaSinica 50(8): 995–1002.

WELDING RESEARCH


GANG ZHANG, YU SHI ([email protected]), YUFEN GU, DING FAN, and MING ZHU are with the Status Key Laboratory of Advanced Processing and Recycling Nonferrous Metals, Lanzhou University of Technology, Lanzhou, China.


Documents

Laser VisionBased Detection of Weld Penetration in GTAW · the welding process to prevent certain defects (Ref. 31). However, for real-time detection during welding, the ability of