Welding Penetration Control for Aluminum Pipe Welding Using Omnidirectional Vision-based Monitoring of Molten Pool *

Transcription

1 [ 溶接学会論文集第 7 巻第 号.7s -s(9)] Welding Penetration Control for Aluminum Pie Welding Using Omnidirectional Vision-based Monitoring of Molten Pool * by Ario Sunar Baskoro **, Rui Masuda**, Masashi Kabutomori** and Yasuo Suga*** This aer resents a study on a new method of welding enetration control for aluminum ie in Tungsten Inert Gas (TIG) welding using omnidirectional vision-based monitoring of molten ool. As circumferential butt-welded ies and the alication of aluminum alloy in ie welding that has been use in various industrial sectors requires new technique of welding monitoring rocess. For ie welding using constant arc current and welding seed, the bead width becomes wider as the circumferential welding of small diameter ies rogresses. In order to avoid the errors and to obtain the uniform weld bead over the entire circumference of the ie, the welding conditions should be controlled. This research studies the intelligent welding rocess of aluminum alloy ie 663S-T5 in fixed osition using the AC welding machine. The monitoring system used an omnidirectional camera to monitor backside image of molten ool. A new method of image rocessing algorithm was imlemented to rocess the catured image and to recognize the edge of molten ool. Back bead width as the result of detection was delivered into fuzzy inference system to control welding seed. From the exerimental results it shows the effectiveness of the control system that is confirmed by sound weld of exerimental results. Key Words: Welding Penetration Control, Aluminum Pie, Omnidirectional Vision-based Molten Pool Detection, Fuzzy Inference System. Introduction For many alications in automation of welding rocess, the need for higher weld quality and reduced manufacturing cost has become increasingly imortant. Advanced welding technology takes art to reduce manufacturing cost, however its use requires a means of sensing and monitoring of error in the rocess. As the alication of ie welding in ower stations, offshore structures, and rocess industries, it is imortant to investigate the characteristic of the welding rocess. There have been roblems in automating arc welding rocesses such as sensing, monitoring and line tracking. Therefore, to achieve full automation of aluminum ie welding, the welding enetration should be controlled. Since the early 96 s sensing and control systems have been successfully imlemented in alications where the sensor could be laced on the backside of the weld and moved in synchronism with the welding torch ). Intelligent control systems has been develoed for modeling and controlling the welding rocess using neural network -3) and fuzzy techniques 4-5). Another difficulty to control of an arc welding rocess is how to detect weld ool geometrical features, such as weld bead width and enetration, either from the toside or backside, conveniently and in real-time. The exeriment using the vision sensing to control the TIG weld width for stainless steel ) and aluminum alloy 3,5) ie has been conducted with the algorithm of image rocessing to detect molten ool s edge using *** Received: 8..8 *** Graduate, School of Science and Technology, Keio University *** Member, Faculty of Science and Technology, Keio University lain mirror that rotates along the welding torch during the ie welding rocess. This research studies the intelligent welding rocess of aluminum alloy ie 663S-T5 in fixed osition using the AC welding machine. The roosed monitoring system used an omnidirectional vision-based monitoring of molten ool to cature wide view of backside image of molten ool circumferentially without rotated art of mirror. This camera consists of a ersective camera and a hyerboidal mirror allowing a central rojection by reflected rays. The constant AC current of TIG welding rocess was used and the welding seed derived from the back bead width of molten ool image was controlled. Image rocessing algorithm was constructed to detect edge of molten ool. In this exeriment, the fuzzy inference system was used to control welding enetration by modifying seed.. Exerimental rocedure. Exerimental device Figure shows the ie welding system develoed in this study. The major functional elements of the exerimental system are a circumferential welding maniulator, CCD camera and the image board (56 ixels, 8bit), the ersonal comuter (CPU:.MHz), A/D board to measure arc current and voltage, TIG welding machine, and motor board to control steing motors. Base metal used in this exeriment was aluminum alloy ie A663S-T5. Pulsed TIG AC welding machine with square-wave current and ure argon shielding gas was used.

2 8s 研究論文 Ario SUNAR BASKORO et al.:welding Penetration Control for Aluminum Pie Welding Using Omnidirectional Personal Comuter A/D board V I Motor board Image board Welding torch Fig. Schematic of ie welding system CCD camera TIG Welding Machine mirror and its dimension, resectively. Detail of monitoring system is shown in Fig. 3. Table Material roerties and welding conditions Base metal Al-663S-T5 Diameter of ie (mm) 37.8 Thickness of ie (mm). Density (g/cm 3 ).69 Melting oint ( o C) Thermal conductivity (W/m.K at 5 o C) 9 Welding machine AC Electrode % Th-W (.4 mm) Nominal arc length (mm).5 Welding current, I (A) 5 ~ 7 Pulse current frequency, f (Hz) 5 ~ EN ratio.5 Welding seed, v (mm/s).83 ~ 3.33 Shielding gas % Ar Shielding gas, q (l/s).3 ~.5. Edge Detection of Molten Pool (a) To view (b) Size of hyerboloidal mirror [mm] Fig. Hyerboloidal mirror Fig. 3 Detail of monitoring system Molten ool image Create set window Histogram analysis Set window Edge detection's range Vertical scanning Panorama transformation To and bottom edge Transformed image Find the max width of edge Find center of gravity Width of molten ool Fig. 4 Flowchart of image rocesing In this aer, omnidirectional vision-based molten ool detection system consists of a CCD camera and a hyerboloidal mirror. A CCD camera catured the backside image of molten ool and sent the image to ersonal comuter through the image digitizer. Time required for caturing a single frame was /6s. Figure (a) and (b) resent the to view of the hyerboloidal Figure 4 resents the flowchart of image rocessing algorithm. Due to the low melting oint of aluminum, brightness of molten ool is low; therefore the stable and robust image rocessing algorithm must be constructed. The detail of edge detection of molten ool will be discussed as follows. (a) Histogram Analysis: First, the image of molten ool was analyzed to get the histogram information. Original image size was 56 x 9 ixels. Comaring with the stainless steel, the edge of molten ool in aluminum is very difficult to be detected. After the observation to find the exact osition of molten ool and comared to the real back bead width, it is found that the edge of molten ool aligned between the edge detection range. This range defined as the range between inner and outer brightness as shown in Fig. 5 (a). In this study, the same method of 4) to find the edge was alied. From the histogram analysis, the frequency of brightness value of the image was obtained. The average brightness, g and accumulation of the ercentage of average brightness, were obtained by the Eqs. () and (), resectively. g i55 i i55 i i i i i i55 i % where f (i) is the frequency of brightness at i. Unlike the revious method which alied some rule to obtain the ercentage of to threshold or outer brightness, out and bottom threshold or inner brightness, 4) in, the new method directly add the constant into the value of those ercentages, which are defined as: () ()

3 溶接学会論文集第 7 巻 (9) 第 号 9s (a) Illustration image of molten ool (b) Inner and outer brightness frequency (=8 o ) Fig. 5 Histogram analysis for edge detection X Reflection x=6 Set window Y Molten ool = 36 O Welding direction = O (a) Panorama transformed image (b) Differential brightness at x=6 (c) Bottom edge osition (d) To edge osition Fig. 6 Edge detection of molten ool out in (%) (3) out in(%) (4) In this exeriment as shown in Fig. 5 (b), the value of out and in are.3% and.8%, resectively. Then the value of outer brightness, gout = 4 and outer brightness, gin = 45 can be obtained. (b) Panorama Transformation: All of the image will be transformed into anorama extended image as shown in Fig. 6 (a). The anorama transformation algorithm used in this aer was develoed from 6). Assumed an image coordinate (u,v) and a real-world three-dimensional osition P(X,Y,Z) are defined. The hyerboloidal equation of the roosed mirror is reresented as: X Y Z C D (5) A B where A and B are arameters of the hyerboloidal mirror shae, C X Y, and D is the distance between lens to a center

4 s 研究論文 Ario SUNAR BASKORO et al.:welding Penetration Control for Aluminum Pie Welding Using Omnidirectional Fig. 7 Comarison result of measured and detected molten ool where g(i,j) is the brightness value of a ixel at (i,j) as shown in Fig. 6 (b). Edge osition was determined within this range of searching. By alying the inner brightness and outer brightness, the bottom edge osition can be detected by finding the minimum osition within the range as shown in Fig. 6 (c). In contrary, the to edge osition was detected from maximum osition within the range as shown in Fig. 6 (d). This rocess was reeated along x osition inside the set window. (e) Width Detection: In order to find the width of molten ool, the scanning of widest value of to and bottom edges was conducted. Figure 7 shows the examle of detected edge and width of molten ool..3 Result of Image Processing Algorithm (a) Inut: back bead width error (e) Fig (c) Outut: welding seed correction (v) Fuzzy sets and decision table for fuzzy control oint of the camera. Relationshi equations between Z and X, Z and Y in the image coordinate (u,v) are described as follows: C C f B C BC u v f Z X C D B u f B C BC u v f Z Y C D B v (b) Inut: back bead width error deviation (e) where f is the focal length of a CCD camera. For the roosed omnidirectional camera, A =, B =, D = 6 mm, f =.6 mm and the size of CCD is 6.59 mm 4.94 mm. Panorama image size was 5 x 86 ixels. (c) Set Window: After finding the center of gravity, set window was created automatically to locate the scanning area and reduce the time of edge detection. The set window was created from the oints of left, right, to and bottom as shown in Fig. 6 (a). Maximum size of set window was 5 x ixels. (d) To and Bottom Edge Detection: From the original image, the vertical scanning was erformed to find to and bottom edge detection. Edge detection was erformed on the differential value of brightness along vertical axis g (i,j) 4) : g' ( i, j) g( i, j ) g( i, j) (8) Δe N Z P e N Z N N Z Z P P (d) Decision table P (6) (7) Figure 7 shows the comarison of measured back bead width and detected molten ool width using roosed image rocessing algorithm in reliminary exeriment without control. Image resolution is.93 mm/ixel. It is clearly seen that image rocessing algorithm could detect the molten ool width with good aroximation. However, some errors still occur during the monitoring rocess with the Root Mean Square Error (RMSE) is.66 mm and standard deviation is.4 mm. The cause of the errors might come from the inner and outer brightness as the range for scanning the edge of molten ool. Another reason of this error came from the error of the measurement of bead width..3 Fuzzy Inference System Welding rocess was conducted autogenously for 36 o of circumference and in fixed osition of ie. In constant welding current of 6 A, the torch started to initiate the arc until the initial enetration was roduced. The exeriment with control conducted using fuzzy inference system as control system 7). The outut of the image rocessing which is the width of molten ool, w will become the inut of fuzzy control. The outut of fuzzy control is the correction of welding seed. In this ste, the roosed fuzzy control took two variables to be fuzzified. One was an error (e), which was the difference of back bead width (w(n)) at the concerned time ste (n) from the reference back bead width (w r ): e = w r w n (9) and w r was set at 5 mm. The other was the change of an error defined as: e = w n+ w n () Three kind of membershi (N Negative, Z Zero, P Positive) and triangular membershi functions were used to fuzzify the inuts. Figure 8 (a), (b) and (c) shows the membershi functions and ranges for each fuzzy variable. The decision table

5 溶接学会論文集第 7 巻 (9) 第 号 s (a) Back bead width and welding seed (b) Back bead aearance Fig. 9 Result of exeriment with control for the fuzzy control of welding seed is shown in Fig. 8 (d). 3. Result and discussion In control exeriment, to roduce stable arc condition, the welding seed of.7 mm/s at = o 45 o was ket constant. Figure 9 (a) shows the exeriment result of back bead width and welding seed using fuzzy controller. Fuzzy control determined the correction of welding seed to kee the back bead width in the target range of 5 mm. The result of exeriment with control yields the Root Mean Square Error (RMSE) is.9 mm and standard deviation is.54 mm. At = 45 o 35 o, the welding seed increased from mm/s and the back bead width increased. However, by maintaining and increasing the welding seed until.33 mm/s, the back bead width was ket to 5 mm at = 8 o. Accordingly, by roer control of welding seed, it will roduce excellent back bead width. The exeriment result shows that the bead was smooth in aearance and there was no crack, orosity, undercut and burn through along the circumference as shown in Fig. 9 (b). The back bead width also aligned in the range target of 4 6 mm. In this study, the welding condition was changed during the welding roceeded along the circumference of the ie. However, the suitable welding condition for roducing the good result was obtained at welding current, I = 65 A, ulsed current frequency, f = 5 Hz, minimum welding seed, v min =.83 mm/s, and maximum welding seed, v max =.5 mm/s. In general, the roosed automatic welding system roduced sound weld of aluminum ie by monitoring backside image of molten ool using omnidirectional camera. 4. Conclusions Welding enetration control of aluminum ie using omnidirectional vision-based monitoring of molten ool was constructed. An algorithm to obtain to detect edge of molten ool from anorama image of molten ool was roosed. From the exerimental results using fuzzy inference system, it shows the effectiveness of the control system. References ) E.P. Vilkas: Automation of the Gas Tungsten Arc Welding, Welding J., 45-5 (966), ) M. Muramatsu, Y. Suga, K. Mori: Autonomous Mobile Robot System for Monitoring and Control of Penetration during Fixed Pies Welding, JSME Int. J. Series A, 46-3 (3), ) A.S. Baskoro, M. Kabutomori, Y. Suga: Automatic Welding System of Aluminum Pie by Monitoring Backside Image of Molten Pool Using Vision Sensor, Journal of Solid Mechanics and Materials Engineering, JSME, -5 (8), ) K.Y. Bae, T.H. Lee, K.C. Ahn: An otical sensing system for seam tracking and weld ool control in gas metal arc welding of steel ie, Journal of Materials Processing Tech., --3 (3), ) A.S. Baskoro, M. Kabutomori, Y. Suga: Welding Penetration Control of Aluminum Pie Using Fuzzy Logic Control, The 8 th Sring Round Precision Engineering Conference, The Jaan Society for Precision Engineering, Jaan (7), ) J.C. Kim, M. Muramatsu, Y. Murata, Y., Suga: Omnidirectional Vision-Based Ego-Pose Estimation for an Autonomous In-ie Mobile Robot, Advanced Robotics, -3-4 (7),