monitoring weld pool surface and penetration using ...€¦ · formation remains a challenging task...

Introduction Incomplete penetration has been along-standing issue affecting the safe-ty and integrity of welded structures(Refs. 1–3). In manual welding, askilled welder determines the penetra-tion state by observing the shape ofthe weld pool top surface. The experi-ence could be used to develop an intel-ligent welding system to automaticallycontrol the welding process and reducethe chance of incomplete penetration(Refs. 4–6). Two-dimensional (2D) weld poolgeometry has been studied and ap-

plied to welding quality control in pre-vious research (Refs. 7, 8). Wang et al.(Ref. 9) and Wu and Chen (Ref. 10)utilized the passive vision system forpenetration control during aluminum-alloy gas tungsten arc welding(GTAW). An artificial neural networkmodel was used to extract the weldpool boundary. The 2D features of theweld pool were used to control pene-tration via the Hammerstein model. Liu et al. (Ref. 11) introduced an ac-tive contour method to extract theweld pool boundary from the passivevision images. However, in the passivevision system, arc light that was uti-

lized as the light source may haveoverwhelmed important features inthe image. An alternative approach is to usethe auxiliary laser light to illuminatethe weld pool area (Refs. 12, 13). Inthe studies conducted by Zhang et al.(Ref. 7) and Kovacevic et al. (Ref. 14),the weld pool boundary was extractedwith the laser light that illuminatedthe active vision image with sufficientaccuracy. In principle, three-dimensional (3D)geometry of the weld pool surfacewould provide more comprehensive in-formation for penetration detection(Ref. 8). However, extracting the 3D in-formation remains a challenging taskfor welding automation. In previouswork (Ref. 15), the dot-matrix patternof the structured laser light was project-ed onto the weld pool surface, and thedistorted reflected pattern was capturedfrom an image plane. The 3D-surface reconstruction algo-rithm was developed and verified on aspherical mirror with sufficient accura-cy. Zhao et al. (Ref. 16) used an im-proved shape from shading (SFS) algo-rithm to recover the weld pool surfaceheight (SH) from the passive image.The 3D weld pool surface was recon-structed offline. Wang et al. (Ref. 17)estimated the weld pool surface heightusing the machine vision method andrelated it to penetration during thepulsed gas metal arc welding (GMAW-P) process. In this research, a new method wasproposed using a passive vision systemto determine the weld pool SH duringGTAW. Gas tungsten arc welding uses

WELDING RESEARCH

OCTOBER 2017 / WELDING JOURNAL 367-s

SUPPLEMENT TO THE WELDING JOURNAL, October 2017Sponsored by the American Welding Society and the Welding Research Council

Monitoring Weld Pool Surface and Penetration Using Reversed Electrode Images

A new method was developed to relate weld pool surface height to the reversed electrode image on the weld pool surface during GTAW

BY Z. CHEN, J. CHEN, AND Z. FENG

ABSTRACT The threedimensional weld pool top surface shape provides important information about the state of weld penetration during welding. In this study, a method wasdeveloped to quantitatively relate weld pool surface height to the reversed electrodeimage (REI) on the weld pool surface. This new feature was extracted from the weldpool image using a passive visionbased monitoring system during gas tungsten arcwelding (GTAW). Due to the specular reflection of the weld pool top surface, the REIis visible on the weld pool surface during GTAW. The position of the REI wasdetermined with a robust image processing algorithm. Based on the principle of lightreflection, the distance between the electrode tip and the REI (DERI) was related tothe weld pool surface height. By assuming the weld pool surface was a spherical mirror, a reflection model was established to calculate the surface height (SH) indexbased on the measurement of the DERI, arc length, and weld pool geometry. The proposed method was verified with beadonplate welding experiments. The SH waspositively related to the face reinforcement or depression of the weld bead. Thismethod was applied to monitor the penetration state during beadonplate autogenous welding, particularly when a complete penetration weld was formed.

KEYWORDS • Passive Vision • Reversed Electrode Image • Weld Pool Surface Height • Penetration

Chen Supplement Layout.qxp_Layout 1 9/13/17 4:28 PM Page 367

a nonconsumable tungsten electrodeto establish an arc with the base metal.The intensive arc melts the base metaland forms a weld pool. Due to thespecular reflection of the weld poolsurface, a virtual reversed electrodeimage (REI) was formulated on thesurface of the weld pool. In this paper, the REI was investi-gated and related to weld pool SH andwelding penetration. A robust imageprocessing algorithm was developedto detect the position of the REI fromthe weld pool image. A mathematicalmodel that calculates the SH was alsodeveloped based on the law of reflec-tion. In this model, the weld pool topsurface is considered a spherical mir-ror with a changing curvature. TheSH was quantitatively related to thedistance between the electrode tipand the REI (DERI), arc length, andweld pool width. To verify thismethod, bead-on-plate welding exper-iments were performed on an SS304workpiece. The calculated SH was val-idated through the welding experi-ments with different weld pool sur-face shapes. The relationship betweenthe SH and penetration during bead-on-plate autogenous welding was fur-ther investigated through these exper-iments. The study of REI can be fur-ther applied to control penetrationduring the GTAW process.

Principles of Measurement

Acquisition of the ReversedElectrode Image during GTAW

The diagram of the passive visionweld pool monitoring system is shownin Fig. 1. The moving welding torchwas perpendicular to the stationaryworkpiece. The weld pool SH was de-fined as the vertical distance from thevertex of the weld pool surface to theworkpiece. Considering the weld pooltop surface as a spherical surface mir-ror, a virtual reversed image of thetungsten tip was formulated on theother side of the weld pool due tospecular reflection on the weld poolsurface. The DERI is defined as the following:

DERI = D0 + D1 (1)

Object distance, D0, is equal to thearc length during welding. As men-tioned in previous research (Refs. 20,21), the arc length is defined as thedistance from the tip of the weldingelectrode to the adjacent surface of theweld pool. The image distance, D1, isthe distance from the REI to the weldpool surface, which is determined bythe shape of the weld pool surface. A high-speed charge-coupled device

(CCD) camera was mounted behindthe moving torch to capture the weldpool images. The welding angle anddistance between the camera and thetorch could be adjusted in the experi-ment. The distance between the weldcamera lens and the tungsten tip wasapproximately 150 mm. Based on thepinhole camera model, both of the im-ages of the electrode and the REI werevisible from the camera with an ade-quate camera posture. In this study, the passive vision sys-tem utilized arc light as the major lightsource. A narrow band-pass filter wasadded in front of the camera lens toreduce the intensity of the arc light inthe captured image. The exposure timeand aperture were also adjusted tocontrol the amount of light reachingthe CCD camera. Figure 2 shows the passive visionimage obtained during GTAW withwelding wire. When the exposure timewas selected at 500 s, the weld poolwas fully visible as shown in Fig. 2A.However, the intensive arc light over-whelmed the internal information ofthe weld pool. By reducing the exposuretime to 50 s, the intensive arc lightwas largely suppressed, and only a spin-dle-shaped area was illuminated. Theelectrode tip and REI were clearly visi-ble inside the illuminated area. There-fore, we selected a low exposure time

WELDING RESEARCH

WELDING JOURNAL / OCTOBER 2017, VOL. 96368-s

Fig. 1 — Diagram of passive vision monitoring system.

Fig. 2 — The passive vision image of the weld pool during GTAW withwelding wire: A — Image captured with 500s exposure time; B — imagecaptured with 50s exposure time.

Fig. 3 — Weld pool surface shape in four penetration states during autogenous GTAW: A — Partial penetration; B — critical penetration; C —complete penetration; D — overpenetration.

A

A B C D

B


for the DERI measurement. The dropletin Fig. 2B was introduced by the weld-ing wire with a low feeding speed. Thedisturbance from the droplet was mini-mized because the welding wire wassmoothly fed into the pool.

Reflection Model on the WeldPool Surface

Previous research found that arcpressure and gravity deform the weldpool top surface during GTAW (Refs.22, 23). Arc pressure formulates a de-pression area underneath the elec-trode if the welding current is above200 A. For the experiment with weld-ing current less than 200 A, this ef-fect of arc pressure may be insignifi-cant, but gravity and surface tensionplay an important role on the defor-mation of the weld pool.

This research focuses on bead-on-plate welding under a low current (< 200 A) and low travel speed (1.5–3mm/s) level. Under this condition,previous numerical studies of weldpool behaviors indicated that the weldpool surface shape is an importantcharacteristic associated with weldpenetration, as shown in Fig. 3 (Refs.18, 22, 24). During the partial penetrationstate, as shown in Fig. 3A, the weldpool surface was convex due to the ex-pansion of liquid metal. The criticalpenetration was the transition statefrom partial penetration to completepenetration, when the bottom surface

WELDING RESEARCH


Fig. 4 — A — Reflection diagram of the concave weld pool surface;B — reflection diagram of the convex weld pool surface.

Fig. 6 — Flow chart of the SH calculation.

Fig. 5 — DERI vs. SH and weld pool width (D0 = 4.5 mm).

A B

Table 1 — Algorithm to Extract Position REI from Image

Input: An image frame, location of the electrode tip in the image Output: The location of the reverse image of the electrode tip ZB in the image, DERIs

1: Apply the median filter to the image. 2: Obtain the binary image (Fig. 9B). 3: Determine that the ROI contains the REI from the binary image by scanning the

white area vertically. 4: Determine the location ZB of REI from the original image by searching the

maximum intensity point inside the ROI. 5: Compute the DERIs by |ZA − ZB|.

Chen Supplement Layout.qxp_Layout 1 9/13/17 4:28 PM 69

of the weld pool started to form on thebackside of the workpiece. For the complete penetration weldin Fig. 3C, the area of the weld poolbottom surface was increased, and thesurface tension became the majorforce to support the liquid metal.Therefore, the gravity force formed aconcave-shaped weld pool top surface.The depression of the weld pool sur-face further increased at the overpene-tration state, as shown in Fig. 3D. When joining with a welding wire,along with an increase of liquid weldpool volume, the convexity of the weldpool surface in Fig. 3A can be furtherincreased. Depression of the weld poolsurface in Fig. 3C–D can be reduced. In this research, the sphere-shapemodel was used to describe the weldpool top surface in Equation 2. R is theradius of the sphere, and d is the totallength of the partial sphere, whichequals to the width of the weld pool.Thus, the index of the SH is the ap-proximate calculation of the weld poolsurface height.

(R)2 = (R + SH)2 + (d/2)2 (2)

The reflection diagram of two typesof spherical weld pool surfaces areshown in Fig. 4. O is the vertex pointon the spherical surface. Point C is thecenter of the curvature, where OC = R.F is the focal point of the curved mir-ror. The distance between points F andO is the focal length, ,which is equalto R/2. The concave weld pool had apositive SH, and the convex one had a

negative SH. Based on the reflection law, the vir-tual image was formed in the oppositeside of the mirror because the objectdistance D0 is smaller than the focuslength . The mirror equation de-scribes the relationship between ob-ject distance, D0, and image distance,D1, as follows:

The object distance, D0, is less thanthe absolute value of the image dis-tance |D1| in the concave surface (SH< 0 and F > 0) in Fig. 4A. A magnifiedREI is formed on the other side of theweld pool surface. For the convex sur-face in Fig. 4B, a shrunken REI appearson the other side of the weld pool. Theobject distance D0 is larger than imagedistance |D1|. When SH = 0, the weldpool surface is a flat mirror, and D0 =|D1|. [An REI image with the samesize as the electrode tip is formed.]Based on Equations 2 and 3, the rela-tionship between DERI and SH is fur-ther deduced to the following:

Simulation

In this proposed reflection model,the DERI is determined by SH, weldpool width, and arc length. In Fig. 5,

the simulation curve describes the re-lationship between the DERI, weldpool SH, and weld pool width. The arclength is the only constant variable inthe simulation case (D0 = 4.5 mm),and the 3D surface plot describes therelationship between the DERI, weldpool SH, and weld pool width. Thetrend of the plot shows that increasingthe SH causes the DERI to decrease.

Experimental Procedureand Method

Procedure for Calculating theSurface Height Index

The procedure for calculating theSH is presented in Fig. 6. Camera cali-bration was performed to identifycamera parameters before the weldingexperiments. The position of the REIwas automatically extracted from theimage obtained using the designed im-age processing algorithm in real time.The DERI in the real-object coordinatesystem was further calculated. Mean-while, the arc length was determinedfrom the arc voltage. With the meas-urement of the weld pool width fromthe passive vision image, the SH wascalculated based on the proposed re-flection model.

System Calibration

A vision approach was developed tocalculate the DERI in the object sys-

�4* SHd / 2( )2 +SH2

= 1D0

+ 1D0�DERI

(4)

1f

= 1D0

+ 1D1

(3)

WELDING RESEARCH


Fig. 7 — Schematic of the pinhole camera model. Fig. 8 — Calibration of the camera system.


tem from the image based on the pin-hole camera model as shown in Fig. 7.First, a standard calibration methodwas performed to identify the cameraparameters. A flat checker board planewith known grid size was set under-neath the electrode tip. From the(checker board) image shown in Fig. 8,the intrinsic and extrinsic parameterswere identified (Ref. 25). The cameraintrinsic parameter includes the cam-era focus length, , the stand-off dis-tance S between the camera lens to-ward the object plane, and the princi-ple point O at the center of the sensorplane. The camera setup angle can becalculated from the extrinsic parame-ters, which determined the camera’spose and position in the object coordi-nate system. In the object coordinate systemDERI = |Z'A – Z'B|, Z'A represents theposition of the electrode tip while Z'Brepresents the position of the REI. Inthe camera sensor coordinate system,ZA represents the position of the elec-

trode tip and ZB represents the posi-tion of the REI. The distance betweenZA and ZB was nominated as DERIs =|Z'A – Z'B|. The transformation equa-tion is shown as follows

Equations 5 and 6 can be furthersimplified as Equation 7 if S >> DERI.

DERI DERIs S/(cos ) (7)

From the obtained image, the DERIs can be measured in the unit ofpixel, which was determined by theCCD size and camera resolution. Thecamera resolution was 800 600 pix-els, and the CCD size was 11.2 8.4mm. The pixel size was 0.014 mm.

Algorithm to Extract the REIfrom the Image

To calculate DERI in real time, it wasnecessary to automatically obtain theposition of the electrode tip and the REIfrom the image plane. The location ofthe electrode tip ZA was fixed becausethe camera was mounted on the movingwelding torch. It was challenging to de-termine the location of REI (ZB) in thepassive vision image due to the distur-bance of the droplet and arc. Hence, asshown in Table 1 and Fig. 9, a robust al-gorithm was developed to automaticallyextract the ZB location and compute theDERIs from the image sequence. Based on the proposed algorithm,DERIs can be calculated in real timewith the error range in ±2 pixel.

Experimental Setup and Welding Parameters

The experimental setup of the weldpool surface monitoring system isshown in Fig. 10. The welding torch wasset perpendicular to the workpiece, andit travelled toward the right at a con-stant speed driven by the servomotorand transmission unit. The automaticvoltage control unit (AVC) was appliedto maintain a constant arc length byclose-loop control of the welding voltagein real time. The position of the weldingtorch was automatically adjusted bycomparing the measuring voltage withthe setting voltage. An increment of 0.2V triggered the control unit to adjust itsposition to maintain constant weldingvoltage and arc length during welding.

Z'B cos�S–Z'B sin�

f = ZB (6)

Z'A cos�S–Z'A sin�

f = ZA (5)

WELDING RESEARCH


Fig. 9 — A — Location of ZB in the image; B — image binarization and window search. Fig. 10 — Diagram of experiment setup.

A B

Table 2 — Welding Parameters

Number Material Thickness Current Travel Voltage Wire Feed of Experiment Speed Rate

1 SS304 3 mm 120 A 2 mm/s 11 V / 2 SS304 3 mm 120 A 3 mm/s 11 V / 3 SS304 6 mm 150 A 2 mm/s 11 V 10.6 mm/s 4 SS304 6 mm 150 A 2 mm/s 11 V / 5 SS304 6–1 mm 120 A 2 mm/s 11 V / 6 SS304 6–1 mm 120 A 1.6 mm/s 11 V /


The digital camera was fixed on thewelding torch. The weld pool image andarc voltage can be obtained in real timeat the sampling rate of 30 Hz. To validate the proposed method tocalculate the SH, bead-on-plate weldingexperiments were conducted on anSS304 workpiece with uniform thick-ness from Experiments 1 to 4. Theworkpiece was clamped tightly beforewelding. The welding parameters areshown in Table. 2. All welds were con-ducted at a constant current and weld-ing speed. Experiments 1 and 2 wereconducted on a 3-mm-thick SS304workpiece with the same welding cur-rent and welding voltage. The weldingspeed of Experiment 1 was much lowerthan that of experiment 2. Experiments3 and 4 were conducted on a 6-mm-thick SS304 workpiece. Experiment 3was conducted with welding wire to in-crease the convexity of the weld poolsurface. To further study the relationshipbetween SH and penetration, in Ex-periments 5 and 6, bead-on-plate au-togenous welding was conducted on aspecially designed SS304 workpiecewith decreasing thickness from 6 to 1mm, as shown in Fig. 10. Welding pa-rameters including current, voltage,and travel speed of the torch remainedconstant during each weld. Due to thethickness change of the material, vary-ing penetration states were found inone pass of weld.

Results and Discussion Determination of Arc Length

The object distance, D0, in Equation

3 equals the arclength. The meas-urement of arclength from the arcvoltage has been ap-plied to weldingpenetration controlby many researchers(Refs. 26–28). Thearc voltage wasfound to be propor-tional to the arclength with the given welding current(Refs. 21, 28). In this experimental set-up, arc length was controlled at a con-stant level via the arc voltage controlunit. The measured voltage signal fromthe hall sensor from Experiments 1 and4 are shown in Fig. 11A–B. At 2 s afterthe arc started, the measured arc volt-age became stable at 11 ± 0.1 V. The relationship between arc volt-age and arc length was establishedthrough the calibration curve asshown in Fig. 11C. During the calibra-tion tests, around 30 data points werecollected at different arc lengths withthe welding current at 150 and 120 A.At the same welding current, the arcvoltage increased with the increase inarc length. The calibration curve wasobtained through linear regression.The R-squared value of the regressionwas 0.97–0.98, which indicated thepoints were well in line with the re-gression result. Based on the calibration curveshown in Fig. 11C, the estimated arclength for Experiments 1, 2, 5, and 6was 4.5 mm. For Experiments 3 and 4,the estimated arc length was 3.6 mm.The standard error of the arc lengthestimation was ±0.32 mm.

SH Calculation fromExperiments

The weld bead cross-section view ofExperiments 1–4 are shown in Fig. 12.In Experiment 1, a completely pene-trated weld was produced on a 3-mm-thick plate. A concave-shaped weldbead was made on the workpiece. Thedepression of the weld bead was –1.1mm. In Experiment 2, a partially pene-trated weld was produced on the sameplate by increasing the travel speed. InExperiment 3, a convex weld bead wascreated on a 6-mm-thick workpiece us-ing welding wire. The face reinforce-ment measured from the cross-sectionview was 1.2 mm. The face reinforce-ment of the weld bead in Experiments2 and 4 were close to zero. For Experiments 1–4, the DERI wascalculated from the REI using the de-veloped method as shown in Fig. 13.Before welding, the distance betweenthe electrode and the workpiece wasclose to zero. As welding started, theweld torch was automatically raisedup. In the first 10 s of the weld, thecalculated DERI continuously in-creased from 0 s until the weldingprocess became relatively stable. The

WELDING RESEARCH


Fig. 11 — A — Measured arc voltage for Experiment 1 with a current of 120 A; B — measured arc voltage for experiment 4 with acurrent of 150 A; C — arc voltage vs. arc length.

A B

C


measured data in Fig. 13 after 18 s canbe used to calculate the SH of the fourexperiments. The images in Fig. 14 were obtainedat 20 s when the weld pool became sta-ble. The REI in Fig. 14 were symmetri-cal because the shape of weld pool sur-face was close to a symmetrical sphere.Also, the unsymmetrical weld poolsurface caused aberration of the REIimage, which may have affected theDERI measurement. From the REI image in Fig. 14A–B,the measured DERI was larger in Ex-periment 1 than that in experiment 2due to the decreased weld pool SH. InExperiment 2, the weld pool surfacewas close to flat when the partiallypenetrated weld occurred. The weldpool surface was considered a flat mir-ror. The size of REI was close to theimage of the electrode tip. As the pen-etration increased in Experiment 1, aconcave weld pool surface was formed.Compared to Fig. 14B, the specularweld pool surface served as a concavemirror, which increased the size of the

REI image and DERI in Fig. 14A. During Experiment 4, the weld poolsurface was considered a flat surfacemirror because the convexity of theweld pool was relatively small. In Exper-iment 3, the convex weld pool with anincreased convexity was produced byadding welding wire. A small and fo-cused REI image was found in Fig. 14Cdue to the specular reflection on theconvex shape weld pool surface. Com-pared with experiment 4, the increase inthe weld pool SH caused a decrease inthe measured DERI in Experiment 3, asshown in the measured curve in Fig. 13. The change in arc length was anotherfactor that caused the difference in themeasured DERI in the four experi-ments. Changing the welding currentcaused different arc lengths, which was4.5 ± 0.32 mm in Experiments 1 and 2,as well as 3.6 ± 0.32 mm in experiments3 and 4. Thus, the measured DERI in ex-periments 1 and 2 largely increasedcompared to experiments 3 and 4. Based on the experimental results,we can conclude that the DERI de-

creased with the increase of the weldpool SH with the same arc length, whichis consistent to the simulation results. In Fig. 15, the SH was calculatedduring the stable stage from 15 to 30 sbased on measured variables includingDERI, arc length, and weld pool width.Consequently, the calculated SH wasclose to the measurement of the weldbead from the cross-section view inFig. 13.

Penetration Detection

In Experiments 5 and 6, weldingwas conducted on a specially designedwedge-shaped SS304 workpiece at aconstant current of 120 A. The arcvoltage was controlled at 11 V via aAVC unit. The thickness of the work-piece changed gradually from 6 to 1mm. The front and back side views ofthe two welds were shown in Fig. 16,with the red arrow bar indicating thetime line of the welding process. The front side width of the weld in-creased with the decrease of thickness.

WELDING RESEARCH


Fig. 12 — Weld bead crosssection view for Experiments 1–4. Fig. 13 — Calculated DERI curve for Experiments 1–4.

A B C D

Fig. 14 — The passive REI image of Experiments 1–4 at 20 s after welding started.

Fig. 15 — Calculated SH from Experiments 1–4.


The different penetration states wereidentified from the back side view ofthe weld as shown in Fig. 16B–D. Par-tial penetration was found at the be-ginning of the weld. As the thicknessof the plate decreased, complete pene-tration was formulated at the end ofthe fifth weld. More penetration wasfound in Experiment 6 with a lowerwelding speed at 1.6 mm/s. Overpene-tration and melt-through were formu-lated at the end of the sixth weld. In Fig. 17A, the SH was calculatedbased on the measured variables. Thecalculated SH curve was close to thesurface depression measured from theweld bead. It was found that the SHdid not change much between partialand critical penetration periods of theweld. The calculated SH decreased sig-nificantly once complete penetrationwas formulated. The back side width (BW) was usedas the major variable to describe thepenetration state. In Fig. 17B–C, thederivative of the calculated SH, whichdescribed the trend of SH, was calcu-lated and compared with the BW. Thedata point of the BW was measuredfrom the backside of the weldment af-ter welding. The BW equaled zero dur-ing the partial penetration period. The critical penetration in thisstudy started at 66 s during Experi-ment 5 and ended at 45 s in Experi-ment 6. During this period, the back-side weld pool surface was formulated,and the BW was above zero. However,the calculated SH and its derivative

did not changesignificantly be-cause the weldpool front sidewas close to a flatsurface. The de-rivative of thecalculated SH sig-nificantly de-creased once adeeper completepenetration wasformulated at 80s during Experi-ment 5, and 63 sduring Experi-ment 6, while thedepression of theweld pool surfacewas produced. This phenome-non has also beenconfirmed by pre-vious research(Ref. 29). Thus,the decrease ofSH and its deriva-tive could be used as an important in-formation to detect complete penetra-tion from the top side of the weld poolsurface. Overpenetration started at 80s in Experiment 6. During this period,the weld pool surface was deeply de-pressed and only surface tension sup-ported the liquid pool. An inflectionpoint was found at 72 s in the deriva-tive of the SH, which indicated thatthe decreasing speed of the SH was re-duced when it got close to overpene-

tration. Thus, the derivative of SH canbe further applied to prevent overpen-etration during the welding controlprocess.

Conclusions A new feature of the electrode tipreversed image was defined and ex-tracted from the passive vision imageduring GTAW. A robust algorithm wasdeveloped to automatically track the

WELDING RESEARCH


Fig. 16 — Weld bead on the wedgeshaped workpiece: A — Frontview of Experiment 5; B — back side view of Experiment 5; C —front view of Experiment 6; D — back side view of Experiment 6.

Fig. 17 — A — Comparison of the SH and weld bead in Experiments 5 and 6; B — comparison of the derivative of SH and BW inExperiment 5; C — comparison of the derivative of SH and BW inExperiment 6.

A

A

B

B

C

C

D


position of the REI and calculate theDERI. Based on the spherical mirrorassumption of the weld pool surface,we proposed a reflection model to cal-culate the index of the weld pool sur-face height from the measurement ofthe DERI, arc length, and weldpool width. The method was verifiedwith bead-on-plate experiments. Thefollowing conclusions were drawn: 1. The measured DERI is relevant tothe weld pool surface shape. The in-crease in the weld pool surface heightwill cause a decrease in the DERI. 2. The calculated SH is closely relat-ed to the face reinforcement of theweld bead. The convex and concaveweld pool surface shapes can be deter-mined based on the calculation results. 3. During bead-on-plate autoge-nous welding, no significant change inthe SH was found between the partialpenetration and critical penetrationstates. The SH and the depressiondepth of the weld bead was significant-ly reduced once complete penetrationwas produced. 4. The decreasing speed of the SHwas reduced once the weld was close tooverpenetration.

This research was supported andsponsored by the U.S. Department ofEnergy, Office of Nuclear Energy, fornuclear energy enabling technologiesand crosscutting technology develop-ment efforts, under a prime contractwith Oak Ridge National Laboratory(ORNL), Oak Ridge, Tenn. ORNL ismanaged by UT-Battelle LLC for theU.S. Department of Energy under con-tract DE-AC05-00OR22725.

1. Chen, S. -B., and Wu, J. 2009. Intelli-gentized Methodology for Arc Welding Dy-namical Processes. Springer-Verlag Berlin Heidelberg. 2. Chen, S., Wu, J., and Du, Q. 2011.Non-linear modelling and compound intel-ligent control of pulsed gas tungsten arcwelding dynamics. Proceedings of the Insti-tution of Mechanical Engineers, Part I: Jour-

nal of Systems and Control Engineering225(1): 113–124. 3. Chen, S., and Wu, J. Nov. 5–8, 2007.A survey on intelligentized technologiesfor visual information acquirement, mod-eling and control of arc welding pool dy-namics. Industrial Electronics Society.IECON 2007, 33rd Annual Conference of theIEEE, Taipei, Taiwan. 4. Liu, Y. -K., and Zhang, Y. -M. 2015.Supervised learning of human welder be-haviors for intelligent robotic welding.IEEE Transactions on Automation Scienceand Engineering 14(3): 1532–1541. 5. Chen, S. B, Lou, Y. J, Wu, L., andZhao, D. B. 2000. Intelligent methodologyfor sensing, modeling and control ofpulsed GTAW: Part 1 — Bead-on-platewelding. Welding Journal 79(6): 151-s to163-s. 6. Chen, S., et al. 2000. Intelligentmethodology for sensing, modeling andcontrol of pulsed GTAW: Part 2 — Buttjoint welding. Welding Journal 79(6): 164-sto 174-s. 7. Zhang, Y., Kovacevic, R., and Ruan, S.1995. Sensing and control of weld poolgeometry for automated GTA welding.Transactions of the ASME 117: 210–222. 8. Wang, X. 2014. Three-dimensionalvision-based sensing of GTAW: A review.The International Journal of Advanced Manu-facturing Technology 72(1–4): 333–345. 9. Wang, J., Lin, T., and Chen, S. 2005.Obtaining weld pool vision informationduring aluminium alloy TIG welding. TheInternational Journal of Advanced Manufac-turing Technology 26(3): 219–227. 10. Wu, J., and Chen, S. 2007. Softwaresystem designs of real-time image process-ing of weld pool dynamic characteristics.Robotic Welding, Intelligence and Automa-tion: 303–309. Springer-Verlag Berlin Heidelberg. 11. Liu, J., et al. 2015. Boosting activecontours for weld pool visual tracking inautomatic arc welding. IEEE Transactionson Automation Science and Engineering14(12): 1096–1108. 12. Chen, J., et al. 2015. In situ strainand temperature measurement and model-ling during arc welding. Science and Tech-nology of Welding and Joining 20(3):181–188. 13. Chen, Z., et al. July 12–15, 2016. Insitu strain monitoring in gas tungsten arcwelding processes. IEEE International Con-ference on Advanced Intelligent Mechatronics,Banff, Alberta, Canada. 14. Kovacevic, R., Zhang, Y., and Li, L.1996. Monitoring of weld joint penetrationsbased on weld pool geometrical appearance.Welding Journal 75(10): 317-s to 329-s. 15. Song, H., and Zhang, Y. 2008.

Measurement and analysis of three-dimensional specular gas tungsten arc weldpool surface. Welding Journal 87(4): 85-s to 95-s. 16. Chen, S, et al. 2003. Extraction ofthree-dimensional parameters for weldpool surface in pulsed GTAW with wirefiller. Journal of Manufacturing Science andEngineering 125: 493–499. 17. Wang, Z., Zhang, Y., and Wu, L.2010. Measurement and estimation ofweld pool surface depth and weld penetra-tion in pulsed gas metal arc welding. Weld-ing Journal 89(6): 117-s to 126-s. 18. Zhang, G., et al. 2017. Laser vision-based detection of weld penetration inGTAW. Welding Journal 96(5): 163-s to 172-s. 19. Shi, Y., et al. 2015. Laser visionbased measurement and analysis of weldpool oscillation frequency in GTAW-P.Welding Journal 94(5): 176-s to 187-s. 20. Egerland, S. 2015. A contribution toarc length discussion. Soldagem & Inspeçcão20(3): 367–380. 21. Dos Santos, E. B., et al. 2017. Onthe visualization of gas metal arc weldingplasma and the relationship between arclength and voltage. Applied Sciences 7(5):503. 22. Zhang, Y., Cao, Z., and Kovacevic, R.1996. Numerical analysis of fully penetrat-ed weld pools in gas tungsten arc welding.Proceedings of the Institution of MechanicalEngineers, Part C: Journal of Mechanical En-gineering Science 210(2): 187–195. 23. Lin, M., and Eagar, T. 1985. Influ-ence of arc pressure on weld pool geome-try. Welding Journal 64(6): 163-s to 169-s. 24. Zhang, W., et al. 2012. Characteri-zation of three-dimensional weld pool sur-face in GTAW. Welding Journal 91(7): 195-sto 203-s. 25. Zhang, Z. 1999. Flexible camera cal-ibration by viewing a plane from unknownorientations. The Proceedings of the 7th IEEEInternational Conference on Computer Vi-sion, Redmond, Wash. 26. Li, P., and Zhang, Y.-M. 2001. Ro-bust sensing of arc length. IEEE Transac-tions on Instrumentation and Measurement50(3): 697–704. 27. Zhang, Y., Li, X., and Shao, Z. 2017.Method to monitor and control weld pene-tration in gas tungsten welding and full-position pipe welding. Google patents. 28. Li, X., Zhang, Y. M., Kvidahl, L.2013. Monitoring and control of penetra-tion in GTAW and pipe welding. WeldingJournal 92(6): 190-s to 196-s. 29. Rokhlin, S., and Guu, A. 1993. Astudy of arc force, pool depression, andweld penetration during gas tungsten arcwelding. Welding Journal 72(8): 381-s to390-s.

WELDING RESEARCH


ZONGYAO CHEN and ZHILI FENG ([email protected]) are with the University of Tennessee, Knoxville, Tenn. CHEN, FENG, and JIAN CHEN are with the OakRidge National Laboratory, Oak Ridge, Tenn.

References

Acknowledgments


monitoring weld pool surface and penetration using ...€¦ · formation remains a challenging task...

Documents