- Email: [email protected]
Visual-based spatter detection during high-power disk laser welding
Optics and Lasers in Engineering 54 (2014) 1–7
Contents lists available at ScienceDirect
Optics and Lasers in Engineering journal homepage: www.elsevier.com/locate/optlaseng
Visual-based spatter detection during high-power disk laser welding$ Deyong You a,b, Xiangdong Gao a,n, Seiji Katayama b a Faculty of Electromechanical Engineering, Guangdong University of Technology, No. 100 West Waihuan Road, Higher Education Mega Center, Panyu District, Guangzhou 510006, China b Joining and Welding Research Institute, Osaka University, 11-1 Mihogaoka, Ibaraki, Osaka 567-0047, Japan
art ic l e i nf o
a b s t r a c t
Article history: Received 13 December 2011 Received in revised form 4 September 2013 Accepted 15 September 2013
The spatters created during laser welding are considered as essential information for welding quality inspection. This paper proposes a laser welding quality inspection method based on the high-speed visual detection. A high-power (10 kW) disk laser bead-on-plate welding of Type 304 austenitic stainless workpiece was carried out and two high-speed cameras were used to capture the spatter images. The first one was used to measure the near infrared (IR) light emitted from a molten pool, and the second was utilized to measure the ultraviolet (UV) light and visible light. By comparing the images captured from two different cameras, it was found that the measurement of UV light and visible light was more appropriate for spatter detection. Based on image process technology, a novel spatter detection algorithm was presented. A shape similarity function and angle similarity function of the spatters were established for spatter recognition by defining the spatter features, such as centroid position, gray-value, average gray-value and radius. The spatter volume, spatter gray-value and spatter radius were used to evaluate the welding quality. By comparing the spatter information with the molten pool width, this paper investigates the internal relationship between the welding quality and the spatter feature parameters. Experimental results proved that the proposed spatter feature extraction method could guarantee an accurate evaluation on the quality of high-power disk laser welding. In addition it was demonstrated in this study that, by using high-speed visual detection and image processing technology, the quantities and feature of spatters could be measured during the welding process. & 2013 Elsevier Ltd. All rights reserved.
Keywords: High power laser welding Spattering Visual detection Quality inspection
1. Introduction Laser welding has been widely used for its advantages due to high intensity heat sources in narrowly focusing a laser beam to a small area, which is instrumental in realizing deep-penetration and highspeed welding, and improving mechanical properties [1–5]. An effective method for monitoring the welding process is essential to improve welding quality [6–8]. In order to optimize the productivity of laser welding, electromagnetic emissions from the weld zone, such as reflected light, thermal radiation and plasma radiation, are the most important information during on-line monitoring [9,10]. Several fundamental studies on plasma monitoring have been performed to evaluate the stability of laser welding [11,12]. Highspeed photography has been proved to be an effective method in analyzing the structure and dynamic behavior of a keyhole during the laser welding process [13,14]. Also, metal spatters incurred during laser welding affect the molten pool shape and the welding
☆ This work is an expanded version of the paper published at International Conference on Advanced Design and Manufacturing Engineering (ADME) 2011 in Guangzhou, China, September 16–18, 2011. n Corresponding author. Tel.: þ 86 137 11457326; fax: þ 86 208 5215998. E-mail address: [email protected] (X. Gao).
0143-8166/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.optlaseng.2013.09.010
quality. During laser welding, spattering refers to the scattering of fluid metal made by laser-induced plume steams from the keyhole. Effective monitoring and control over metal spatters are prerequisites to the production of high-quality welds. Sehun Rhee in Korea used IR and UV sensors to collect plasma and spatter signals during welding, and analyzed them by way of multiple characteristics pattern recognition in order to evaluate welding quality [15,16]. A study shows the optimal model for estimating the amount of spatter when considering arc extinction in short circuit transfer mode, and the optimal model for estimating spatter rate using an artificial neural network in the short circuit transfer region of GMAW [17,18]. Also some methods for measuring spatter in gas metal arc welding by optical monitoring of the weld pool and surrounding area via a highframe-rate camera were proposed and have been proven to be effective [19]. Pattern recognition and Kalman filter algorithm were adopted to enhance spatter recognition accuracy and the effectiveness of spatter feature analysis [20]. Some research works have been conducted to monitor the laser welding defects like spatters [21,22]. It was possible to accurately position the welding spatter by using image recognition technology, thus eliminating the spatter influence on molten pool recognition [23]. Previous research works have proved that keyhole wall shows smooth distribution even at a stable status, and only a few small
2
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
spatters are produced inside the keyhole [24]. However, keyhole wall shows distinguishable wave distribution at an unstable status [25]. If the unstable status remains, molten metal will be separated from the molten pool because of vaporized pressure and produce spatters of large size [26]. Since spatters are from the inside of the molten pool, the amount of molten metal within weld seam area will surely reduce when spatters of large size are produced. This results in external defects (such as underfill or lack of fusion) of weld seam [27]. In addition, the production of large-sized spatters indicates an unstable status inside the keyhole. Especially, there is wave distribution of the keyhole wall, which might also result in internal defects (such as porosity) of weld seam [27]. Therefore, the key for spatter detecting is to accurately identify spatters of large size. Up to the present time, however, accurate quantification of spatter during laser welding process is still a formidable task. In this work the high-power disk laser bead-on-plate welding of Type 304 austenitic stainless steel plates was carried out and two high-speed cameras were used to capture the spatter images of welding process in order to extract instantaneous variation of metal spatters. It was followed by the performance of image preprocessing algorithm to extract spatter information of each image, including spatter centroid, area, grayscale, average grayscale and direction. A spatter searching matrix was then built up accordingly. The spatters generated at time t and prior to time t could be distinguished by comparing the spatters feature similarity functions. The radiuses of spatters were extracted to calculate the volume of spatters generated at time t, which was used as quantification information for evaluating the welding quality. It was also demonstrated that the spatters gray-value could be used for welding quality inspection.
of camera 1 was 01, while that of camera 2 was 751. The spatter images, shown in Figs. 3 and 4, were captured by different cameras. The image resolution was 512 pixel 512 pixel. The images shown in Fig. 3 contained clear information of plume (UV light) and spatters (visible light). The images shown in Fig. 4 reflected detail information of molten pool (IR band). The images captured by camera 1 are used for spatter detection, while images captured by camera 2 are more conducive to analyze thermal distribution of molten pool. The image sequence captured by highspeed camera 1 is shown in Fig. 5. By using high-speed photography and image processing technology the spatters could be detected during the laser welding process. Thus the detailed information of spatters, such as spatter volume, spatter grayvalue and spatter radius, could be extracted to evaluate the laser welding quality.
3. Extraction and quantification of spatter features For the convenience of image processing, the spatter images were converted to grayscale images, which are shown in Fig. 6(a) and (d). Also, the grayscale images were pre-processed to eliminate plume information in order to obtain accurate spatter features, which are shown in Fig. 6(b) and (e). An individual spatter j at sample time t was defined as ct(i,j). This spatter j contained six characteristics (i ¼ 1; 2; :::; 6), which were centroid positions ðct ð1; jÞ; ct ð2; jÞÞ, area ct ð3; jÞ, gray-value ct ð4; jÞ, average gray-value ct ð5; jÞ and radius ct ð6; jÞ. The characteristics were defined as follows: ct ð1; jÞ ¼
∑x;y A Dj Pt ðx; yÞx ; ∑x;y A Dj Pt ðx; yÞ
ct ð2; jÞ ¼
∑x;y A Dj Pt ðx; yÞy ; ∑x;y A Dj ðtÞ Pt ðx; yÞ
2. Experimental setup The experimental setup includes a Motorman 6-axis robot, high-power disk laser TruDisk-10003(10 kW), NAC high-speed camera (2000 f/s), shielding gas (argon), and Type 304 austenitic stainless workpiece (150 100 10 mm3). The structure of the experimental setup is shown in Fig. 1. The beam diameter of laser focus was 480 μm, and the laser wavelength was 1030 nm. The gas flow was 40 L/min while the nozzle angle was 451. Welding speed was 4.5 m/min. The workpiece was driven by a precise servo motor installed on the working table. The spectral band-pass filter setup on camera 1 was 320–750 nm (measurement of UV and visible light induced plume and spatter), while the other setup on camera 2 was 960–990 nm (measurement of near IR light emitted from the molten pool). The response curves of different filters are shown in Fig. 2. As shown in Fig. 2(b), there are two optical filters set up on camera 2 in order to capture the near IR light. The angle
Fig. 1. Experimental setup of high-power disk laser bead-on-plate welding.
ct ð3; jÞ ¼ dj ðtÞ; ct ð4; jÞ ¼
x; y ¼ 1; :::; 512
∑
Pt ðx; yÞ;
x;y A Dj ðtÞ
∑x;y A Dj ðtÞ Pt ðx; yÞ ; dj ðtÞ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ∑x;y A Ej ðtÞ ðx c1;j Þ2 þ ðy c2;j Þ2 ct ð6; jÞ ¼ ; ej ðtÞ
ð1Þ
x; y ¼ 1; :::; 512
ð2Þ
ct ð5; jÞ ¼
x; y ¼ 1; :::; 512
ð3Þ
where ct(i,j) is the data matrix of spatters on the images captured at time t, Pt is the spatter image at time t, and Pt ðx; yÞ is the grayvalue of position (x,y); Dj and Ej represent the covering area and edge of spatter j, respectively, while dt ðjÞ and et ðjÞ are the pixel numbers of the covering area and edge of spatter j at time t respectively. To recognize the spatter generated at time t, spatter searching matrix ot ðu; vÞ was established, which contained information of all spatters generated before time t (including time t), such as centroid positions ðot ð1; vÞ; ot ð2; vÞÞ, area ot ð3; vÞ, gray-value ot ð4; vÞ, average gray-value ot ð5; vÞ and radius ot ð6; vÞ. In order to distinguish the spatters generated at time t from the spatters generated prior to time t, shape similarity function simiðv; jÞ and angle similarity function dgsimiðv; jÞ of the spatters were established which are expressed in Eqs. (4) and (5). Both functions were used for comparing the spatter feature similarity between searching matrix ot 1 ðu; vÞ at time t 1 and data matrix ct ði; jÞ at time t. By using the shape similarity function and angle similarity function, it is possible to extract the position and feature parameters of the new spatter generated at time t, and add its new information to ot ðu; vÞ as searching basis at time t þ 1. The process
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
3
Fig. 2. Response curves of camera filter. (a) Response curves of UV and visible light band-pass filter. (b) Response curves of near IR light band-pass filter (combination of optical filter 1 and optical filter 2).
Fig. 3. Image with plume and spatter information captured from camera 1.
Fig. 4. Image with molten pool distribution and spatter information captured from camera 2.
is expressed as follows: jct ð3; jÞ ot 1 ð3; vÞj ot 1 ð3; vÞ jct ð4; jÞ ot 1 ð4; vÞj jct ð5; jÞ ot 1 ð5; vÞj þ þ ot 1 ð4; vÞ ot 1 ð5; vÞ
simiðv; jÞ ¼
ð4Þ
oct ði; lÞ ¼ fct ði; jÞjð \ vr ¼ 1 simiðr; jÞ 4 0:9Þ [ ð \ vr ¼ 1 dgsimiðr; jÞ 4 20Þg;
ct ð1; jÞ ot 1 ð1; vÞ dgsimiðv; jÞ ¼ arctanð Þ ot 1 ð7; vÞ ct ð2; jÞ o ð2; vÞ
ð5Þ
t1
( ot ðu1 ; vÞ ¼
ct ði; jÞ; ot 1 ðu; vÞ;
minðsimiðv; jÞÞ o S \ dgsimiðv; jÞ oDeg ; minðsimiðv; jÞÞ 4 S [ dgsimiðv; jÞ 4Deg
u1 ¼ i ¼ 1; :::; 6
ð6Þ ( ot ð7; vÞ ¼
ot 1 ð1;vÞ arctanðcctt ð1;jÞ ð2;jÞ ot 1 ð2;vÞÞ;
minðsimiðv; jÞÞ o S \ dgsimiðv; jÞo Deg
ot 1 ð7; vÞ;
minðsimiðv; jÞÞ 4 S [ dgsimiðv; jÞ4 Deg
( ot ð8; vÞ ¼
at time t and any spatters in searching matrix ot 1 ðu; vÞ, spatter j was used as new information oct ði; lÞ at time t, and then added to the searching matrix ot ðu; vÞ as searching basis at time t þ 1. This process is shown as follows:
ot 1 ð8; vÞ þ 1;
minðsimiðv; jÞÞ o S \ dgsimiðv; jÞ o Deg
ot 1 ð8; vÞ;
minðsimiðv; jÞÞ 4 S [ dgsimiðv; jÞ 4 Deg
ð7Þ
i ¼ 1; :::; 6; l ¼ 1; :::; m ot ðu1 ; v þ lÞ ¼ oct ði; lÞ;
where ot ð7; vÞ is the moving direction of spatters and is expressed in degree, S is the shape similarity threshold 0.9, Deg is the angle similarity threshold 20, and ot ð8; vÞ represents the occurrence frequency of spatters. If there was no similarity between spatter j
u1 ¼ i ¼ 1; :::; 6;
l ¼ 1; :::; m
ð10Þ
ot ð7; v þ lÞ ¼ ot ð8; v þ lÞ ¼ 0
ð11Þ
ot ð9; v þ lÞ ¼ t
ð12Þ
ot ðu2 ; v þ lÞ ¼ oct ði; lÞ;
ð8Þ
ð9Þ
u2 ¼ i þ 9 ¼ 10; :::; 15;
l ¼ 1; :::; m
ð13Þ
where m is the number of new spatter information generated at time t, and ot ð9; vÞ represents the time when new information is created. An accumulator ot ð8; vÞ for recording spatter occurrence frequency was established to filter the noise information of the images. The information was considered as spatter information generated at time t only when its occurrence frequency reached SN¼ 5, and was then saved to the new-spatter matrix N(t). The spatters detection results are shown in Fig. 6(c) and (f). The
4
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
Fig. 5. Spatters image sequence captured by high-speed camera 1.
Fig. 6. Spatter detection. (a) Time t þ 0 spatters image P t ðx; yÞ. (b) Time t þ 0 spatters image without plume. (c) Time t þ 0 spatters detection. (d) Time t þ 1 spatters image P t þ 1 ðx; yÞ. (e) Time t þ 1 spatters image without plume. (f) Time t þ 1 spatters detection.
process is shown in Fig. 7.
process is expressed as follows: NðtÞ ¼ nt ða; bÞ;
Nð1Þ ¼ zeroð7; 1Þ
nt ða1 ; b þ1Þ ¼ ot ðu2 ; vÞ; nt ð7; b þ 1Þ ¼ ot ð9; vÞ; a1 ¼ u2 10 ¼ 1; :::; 6; o8;v ðtÞ ¼ SN
ð14Þ
SVðtÞ ¼
∑
4π½nf ð6; bÞ3
nf ð7;bÞ ¼ t
ð15Þ
where N(1) represents the initial value of new-spatter matrix. The new-spatter matrix made it possible to extract the total volume SVðtÞ, gray-value SGðtÞ and radius SRðtÞ of spatters generated at time t, which are given in Eqs. (16)–(18). The spatter detection
3
SGðtÞ ¼ ∑ nf ð4; bÞ; nf ð7;bÞ
SRðtÞ ¼
∑
nf ð7;bÞ ¼ t
nf ð6; bÞ;
;
t ¼ 1; 2; :::; f
t ¼ 1; 2; :::; f
t ¼ 1; 2; :::; f
ð16Þ
ð17Þ
ð18Þ
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
5
Fig. 7. Flowchart of spatter detection process.
Fig. 8. Comparison between weld seam and spatters volume detected from spatters image sequences.
where f is the total number of spatter images (in this case f¼2400), with the condition that SN þt o f . The results of spatters volume detection, spatters gray-value detection and spatters radius detection are shown in Figs. 8, 9 and 10 respectively. Ten frames of spatter images were chosen to illustrate the detail information of spatters and corresponding weld seam, as shown in Table 1. The bead width was narrow while larger spatters were generated from the molten pool, such as samples 5–7 in Table 1. The number of spatters created at different times is varying. The largest number of spatters created at the same time is 4, which is shown in sample 4 of Table 1. Also, sometimes there are no spatters generated from the molten pool, as seen in samples 2 and 8 in Table 1. Due to the limited depth of field of the camera, a few spatters are out of focus as shown in Fig. 3. However, this does not pose much influence on the inspection result. The proposed system automatically identified the spatters as soon as they were ejected from the keyhole, which is shown in Fig. 6, and saved the parameters of the spatters. The
size of the spatters changed as they moved, and the identification system synchronously updated spatter parameters to ensure the accuracy of spatter tracking. To be more specific, the system would save two groups of spatter parameters, including the original status parameters and the real-time status parameters. However, only the original status parameters were used as the basis for evaluating welding quality. This is because the spatters were within the focus of the camera (that is, within the welding area) at their original status, which therefore yields accurate parameters. It helps to avoid measurement error of the spatter size caused by defocus.
4. Conclusions Experimental results showed that there was a dramatic width decrease in the middle area of the weld seam, and the corresponding detection results indicated that spatter volume and gray-value within
6
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
Fig. 9. Comparison between weld seam and spatters gray-value detected from spatters image sequences.
Fig. 10. Comparison between weld seam and spatters radius detected from spatters image sequences.
this range were relatively large. These characteristics help to prove the effectiveness of applying a high-speed visual detection method to extract the instantaneous information of spatters during laser welding. By using the proposed spatter recognition detection algorithm to extract feature parameters of the spatters, the spatter volume and gray-value could be used as accurate quantification information for
evaluating weld seam quality. The experimental results indicated that there was a close relationship between the spatter feature information and the welding quality, since the weld seam width decreased as the numbers and mass of the spatters increased. Accordingly, an effective control over spatter numbers was crucial for improving laser welding quality.
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
Table 1 Details of spatters and weld seam during laser welding process. Sample Bead width (mm)
Image sequence (frame)
1
1.71
402
2 3
1.52 1.67
508 857
4
1.71
900
5
0.95
1099
6
1.00
1198
7 8 9 10
1.04 1.43 1.62 1.57
1203 1340 1685 2144
Spatter volume (pixel3)
Spatter grayvalue ( 104)
Spatter radius (pixel)
46.4 4.2 0 33.5 4.2 91.9 33.5 46.4 4.2 652.3 11.5 4.2 33.5 1480.3 0 11.8 4.2
1.2069 0.2092 0 0.7154 0.6078 1.0561 0.6272 1.0097 0.2201 3.0063 0.7730 1.1571 0.4672 11.3400 0 1.0326 0.5518
2.2 1 0 2 1 2.8 2 2.2 1 5.3 1.4 1 2 7.1 0 1.4 1
Acknowledgments This work was supported in part by the National Natural Science Foundation of China (51175095), the Guangdong Provincial Natural Science Foundation of China (10251009001000001 and 915100900 1000020), the Specialized Research Fund for the Doctoral Program of Higher Education of China (20104420110001) and the State Scholarship Fund from China Scholarship Council (2011844250). References [1] Katayama S, Kawahito Y. Laser direct joining of metal and plastic. Scripta Materialia 2008;59:1247–50. [2] Gao X, You D, Katayama S. Infrared image recognition for seam tracking monitoring during fiber laser welding. Mechatronics 2012;22(4):370–80. [3] Mei L, Chen G, Jin X, Zhang Y, Wu Q. Research on laser welding of highstrength galvanized automobile steel sheets. Optics and Lasers in Engineering 2009;47(11):1117–24. [4] Yan J, Gao M, Zeng X. Study on microstructure and mechanical properties of 304 stainless steel joints by TIG, laser and laser–TIG hybrid welding. Optics and Lasers in Engineering 2010;48(4):512–7. [5] Ni X, Zhou Z, Wen X, Li L. The use of Taguchi method to optimize the laser welding of sealing neuro-stimulator. Optics and Lasers in Engineering 2011;49 (3):297–304. [6] Pal K, Pal SK. Monitoring of weld penetration using arc acoustics. Materials and Manufacturing Processes 2011;26(5):684–93. [7] You D, Gao X, Katayama S. Multiple-optics sensing of high-brightness disk laser welding process. NDT & E International 2013;60:32–9.
7
[8] Rizzia D, Sibillanoa T, Calabresea PP, Anconaa A, Lugaràa PM. Spectroscopic, energetic and metallographic investigations of the laser lap welding of AISI 304 using the response surface methodology. Optics and Lasers in Engineering 2011;49(7):892–8. [9] Olsson R, Eriksson I, Powell J, Kaplan AFH. Challenges to the interpretation of the electromagnetic feedback from laser welding. Optics and Lasers in Engineering 2011;49(2):188–94. [10] Olsson R, Eriksson I, Powell J, Kaplan AFH. Advances in pulsed laser weld monitoring by the statistical analysis of reflected light. Optics and Lasers in Engineering 2011;49(11):1352–9. [11] Colombo D, Colosimo BM, Previtali B. Comparison of methods for data analysis in the remote monitoring of remote laser welding. Optics and Lasers in Engineering 2013;51(1):34–46. [12] Liu L, Chen M. Interactions between laser and arc plasma during laser–arc hybrid welding of magnesium alloy. Optics and Lasers in Engineering 2011;49 (9–10):1224–31. [13] Moradi M, Ghoreishi M, Frostevarg J, Kaplan AFH. An investigation on stability of laser hybrid arc welding. Optics and Lasers in Engineering 2013;51 (4):481–7. [14] Eriksson I, Powell J, Kaplan AFH. Melt behavior on the keyhole front during high speed laser welding. Optics and Lasers in Engineering 2013;51 (6):735–40. [15] Park H, Rhee S. Analysis of mechanism of plasma and spatter in CO2 laser welding of galvanized steel. Optics & Laser Technology 1999;31(2):119–26. [16] Park YW, Park H, Rhee S, Kang M. Real time estimation of CO2 laser weld quality for automotive industry. Optics & Laser Technology 2002;34 (2):135–42. [17] Kang MJ, Rhee S. The statistical models for estimating the amount of spatter in the short circuit transfer mode of GMAW. Welding Journal 2001;80(1):1–8. [18] Kang MJ, Kim Y, Ahn S, Rhee S. Spatter rate estimation in the short-circuit transfer region of GMAW. Welding Journal 2003;82(9):238–47. [19] Schwab G, Steele JPH, Vincent TL. Vision based spatter classification for contaminant detection. Welding Journal 2009;88(6):121–30. [20] Jäger M, Humbert S, Hamprecht FA. Sputter tracking for the automatic monitoring of industrial laser-welding processes. IEEE Transactions on Industrial Electronics 2008;55(5):2177–84. [21] Kaplan A.F.H., Norman P., Eriksson I., Analysis of the keyhole and weld pool dynamics by imaging evaluation and photodiode monitoring. In: Proceedings of LAMP2009 – the 5th international congress on laser advanced materials processing, 2009, Japan. [22] Katayama S, Mizutani M, Tarui T, Mori K. Monitoring and phenomena observation during YAG laser lap welding of Zn-coated steel sheets. Journal of Lanzhou University of Technology 2004;30(4):31–6. [23] Stache NC, Zimmer H, Gedicke J. Robust high-speed melt pool measurements for laser welding with sputter detection capability. Pattern Recognition 2007;4713:476–85. [24] Kawahito Y, Matsumoto N, Abe Y, Katayama S. Relationship of laser absorption to keyhole behavior in high power fiber laser welding of stainless steel and aluminum alloy. Journal of Materials Processing Technology 2011;211 (10):1563–8. [25] Zhang M, Chen G, Zhou Y, Li S. Direct observation of keyhole characteristics in deep penetration laser welding with a 10 kW fiber laser. Optics Express 2013;21(7):19997–20004. [26] Kaplan AFH, Powell J. Spatter in laser welding. Journal of Laser Applications 2011;23(2):032005. [27] Kawahito Y, Matsumoto N, Katayama S. High quality welding of stainless steel with 10 kW high power fibre laser. Science and Technology of Welding and Joining 2009;14(4):288–94.
Contents lists available at ScienceDirect
Optics and Lasers in Engineering journal homepage: www.elsevier.com/locate/optlaseng
Visual-based spatter detection during high-power disk laser welding$ Deyong You a,b, Xiangdong Gao a,n, Seiji Katayama b a Faculty of Electromechanical Engineering, Guangdong University of Technology, No. 100 West Waihuan Road, Higher Education Mega Center, Panyu District, Guangzhou 510006, China b Joining and Welding Research Institute, Osaka University, 11-1 Mihogaoka, Ibaraki, Osaka 567-0047, Japan
art ic l e i nf o
a b s t r a c t
Article history: Received 13 December 2011 Received in revised form 4 September 2013 Accepted 15 September 2013
The spatters created during laser welding are considered as essential information for welding quality inspection. This paper proposes a laser welding quality inspection method based on the high-speed visual detection. A high-power (10 kW) disk laser bead-on-plate welding of Type 304 austenitic stainless workpiece was carried out and two high-speed cameras were used to capture the spatter images. The first one was used to measure the near infrared (IR) light emitted from a molten pool, and the second was utilized to measure the ultraviolet (UV) light and visible light. By comparing the images captured from two different cameras, it was found that the measurement of UV light and visible light was more appropriate for spatter detection. Based on image process technology, a novel spatter detection algorithm was presented. A shape similarity function and angle similarity function of the spatters were established for spatter recognition by defining the spatter features, such as centroid position, gray-value, average gray-value and radius. The spatter volume, spatter gray-value and spatter radius were used to evaluate the welding quality. By comparing the spatter information with the molten pool width, this paper investigates the internal relationship between the welding quality and the spatter feature parameters. Experimental results proved that the proposed spatter feature extraction method could guarantee an accurate evaluation on the quality of high-power disk laser welding. In addition it was demonstrated in this study that, by using high-speed visual detection and image processing technology, the quantities and feature of spatters could be measured during the welding process. & 2013 Elsevier Ltd. All rights reserved.
Keywords: High power laser welding Spattering Visual detection Quality inspection
1. Introduction Laser welding has been widely used for its advantages due to high intensity heat sources in narrowly focusing a laser beam to a small area, which is instrumental in realizing deep-penetration and highspeed welding, and improving mechanical properties [1–5]. An effective method for monitoring the welding process is essential to improve welding quality [6–8]. In order to optimize the productivity of laser welding, electromagnetic emissions from the weld zone, such as reflected light, thermal radiation and plasma radiation, are the most important information during on-line monitoring [9,10]. Several fundamental studies on plasma monitoring have been performed to evaluate the stability of laser welding [11,12]. Highspeed photography has been proved to be an effective method in analyzing the structure and dynamic behavior of a keyhole during the laser welding process [13,14]. Also, metal spatters incurred during laser welding affect the molten pool shape and the welding
☆ This work is an expanded version of the paper published at International Conference on Advanced Design and Manufacturing Engineering (ADME) 2011 in Guangzhou, China, September 16–18, 2011. n Corresponding author. Tel.: þ 86 137 11457326; fax: þ 86 208 5215998. E-mail address: [email protected] (X. Gao).
0143-8166/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.optlaseng.2013.09.010
quality. During laser welding, spattering refers to the scattering of fluid metal made by laser-induced plume steams from the keyhole. Effective monitoring and control over metal spatters are prerequisites to the production of high-quality welds. Sehun Rhee in Korea used IR and UV sensors to collect plasma and spatter signals during welding, and analyzed them by way of multiple characteristics pattern recognition in order to evaluate welding quality [15,16]. A study shows the optimal model for estimating the amount of spatter when considering arc extinction in short circuit transfer mode, and the optimal model for estimating spatter rate using an artificial neural network in the short circuit transfer region of GMAW [17,18]. Also some methods for measuring spatter in gas metal arc welding by optical monitoring of the weld pool and surrounding area via a highframe-rate camera were proposed and have been proven to be effective [19]. Pattern recognition and Kalman filter algorithm were adopted to enhance spatter recognition accuracy and the effectiveness of spatter feature analysis [20]. Some research works have been conducted to monitor the laser welding defects like spatters [21,22]. It was possible to accurately position the welding spatter by using image recognition technology, thus eliminating the spatter influence on molten pool recognition [23]. Previous research works have proved that keyhole wall shows smooth distribution even at a stable status, and only a few small
2
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
spatters are produced inside the keyhole [24]. However, keyhole wall shows distinguishable wave distribution at an unstable status [25]. If the unstable status remains, molten metal will be separated from the molten pool because of vaporized pressure and produce spatters of large size [26]. Since spatters are from the inside of the molten pool, the amount of molten metal within weld seam area will surely reduce when spatters of large size are produced. This results in external defects (such as underfill or lack of fusion) of weld seam [27]. In addition, the production of large-sized spatters indicates an unstable status inside the keyhole. Especially, there is wave distribution of the keyhole wall, which might also result in internal defects (such as porosity) of weld seam [27]. Therefore, the key for spatter detecting is to accurately identify spatters of large size. Up to the present time, however, accurate quantification of spatter during laser welding process is still a formidable task. In this work the high-power disk laser bead-on-plate welding of Type 304 austenitic stainless steel plates was carried out and two high-speed cameras were used to capture the spatter images of welding process in order to extract instantaneous variation of metal spatters. It was followed by the performance of image preprocessing algorithm to extract spatter information of each image, including spatter centroid, area, grayscale, average grayscale and direction. A spatter searching matrix was then built up accordingly. The spatters generated at time t and prior to time t could be distinguished by comparing the spatters feature similarity functions. The radiuses of spatters were extracted to calculate the volume of spatters generated at time t, which was used as quantification information for evaluating the welding quality. It was also demonstrated that the spatters gray-value could be used for welding quality inspection.
of camera 1 was 01, while that of camera 2 was 751. The spatter images, shown in Figs. 3 and 4, were captured by different cameras. The image resolution was 512 pixel 512 pixel. The images shown in Fig. 3 contained clear information of plume (UV light) and spatters (visible light). The images shown in Fig. 4 reflected detail information of molten pool (IR band). The images captured by camera 1 are used for spatter detection, while images captured by camera 2 are more conducive to analyze thermal distribution of molten pool. The image sequence captured by highspeed camera 1 is shown in Fig. 5. By using high-speed photography and image processing technology the spatters could be detected during the laser welding process. Thus the detailed information of spatters, such as spatter volume, spatter grayvalue and spatter radius, could be extracted to evaluate the laser welding quality.
3. Extraction and quantification of spatter features For the convenience of image processing, the spatter images were converted to grayscale images, which are shown in Fig. 6(a) and (d). Also, the grayscale images were pre-processed to eliminate plume information in order to obtain accurate spatter features, which are shown in Fig. 6(b) and (e). An individual spatter j at sample time t was defined as ct(i,j). This spatter j contained six characteristics (i ¼ 1; 2; :::; 6), which were centroid positions ðct ð1; jÞ; ct ð2; jÞÞ, area ct ð3; jÞ, gray-value ct ð4; jÞ, average gray-value ct ð5; jÞ and radius ct ð6; jÞ. The characteristics were defined as follows: ct ð1; jÞ ¼
∑x;y A Dj Pt ðx; yÞx ; ∑x;y A Dj Pt ðx; yÞ
ct ð2; jÞ ¼
∑x;y A Dj Pt ðx; yÞy ; ∑x;y A Dj ðtÞ Pt ðx; yÞ
2. Experimental setup The experimental setup includes a Motorman 6-axis robot, high-power disk laser TruDisk-10003(10 kW), NAC high-speed camera (2000 f/s), shielding gas (argon), and Type 304 austenitic stainless workpiece (150 100 10 mm3). The structure of the experimental setup is shown in Fig. 1. The beam diameter of laser focus was 480 μm, and the laser wavelength was 1030 nm. The gas flow was 40 L/min while the nozzle angle was 451. Welding speed was 4.5 m/min. The workpiece was driven by a precise servo motor installed on the working table. The spectral band-pass filter setup on camera 1 was 320–750 nm (measurement of UV and visible light induced plume and spatter), while the other setup on camera 2 was 960–990 nm (measurement of near IR light emitted from the molten pool). The response curves of different filters are shown in Fig. 2. As shown in Fig. 2(b), there are two optical filters set up on camera 2 in order to capture the near IR light. The angle
Fig. 1. Experimental setup of high-power disk laser bead-on-plate welding.
ct ð3; jÞ ¼ dj ðtÞ; ct ð4; jÞ ¼
x; y ¼ 1; :::; 512
∑
Pt ðx; yÞ;
x;y A Dj ðtÞ
∑x;y A Dj ðtÞ Pt ðx; yÞ ; dj ðtÞ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ∑x;y A Ej ðtÞ ðx c1;j Þ2 þ ðy c2;j Þ2 ct ð6; jÞ ¼ ; ej ðtÞ
ð1Þ
x; y ¼ 1; :::; 512
ð2Þ
ct ð5; jÞ ¼
x; y ¼ 1; :::; 512
ð3Þ
where ct(i,j) is the data matrix of spatters on the images captured at time t, Pt is the spatter image at time t, and Pt ðx; yÞ is the grayvalue of position (x,y); Dj and Ej represent the covering area and edge of spatter j, respectively, while dt ðjÞ and et ðjÞ are the pixel numbers of the covering area and edge of spatter j at time t respectively. To recognize the spatter generated at time t, spatter searching matrix ot ðu; vÞ was established, which contained information of all spatters generated before time t (including time t), such as centroid positions ðot ð1; vÞ; ot ð2; vÞÞ, area ot ð3; vÞ, gray-value ot ð4; vÞ, average gray-value ot ð5; vÞ and radius ot ð6; vÞ. In order to distinguish the spatters generated at time t from the spatters generated prior to time t, shape similarity function simiðv; jÞ and angle similarity function dgsimiðv; jÞ of the spatters were established which are expressed in Eqs. (4) and (5). Both functions were used for comparing the spatter feature similarity between searching matrix ot 1 ðu; vÞ at time t 1 and data matrix ct ði; jÞ at time t. By using the shape similarity function and angle similarity function, it is possible to extract the position and feature parameters of the new spatter generated at time t, and add its new information to ot ðu; vÞ as searching basis at time t þ 1. The process
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
3
Fig. 2. Response curves of camera filter. (a) Response curves of UV and visible light band-pass filter. (b) Response curves of near IR light band-pass filter (combination of optical filter 1 and optical filter 2).
Fig. 3. Image with plume and spatter information captured from camera 1.
Fig. 4. Image with molten pool distribution and spatter information captured from camera 2.
is expressed as follows: jct ð3; jÞ ot 1 ð3; vÞj ot 1 ð3; vÞ jct ð4; jÞ ot 1 ð4; vÞj jct ð5; jÞ ot 1 ð5; vÞj þ þ ot 1 ð4; vÞ ot 1 ð5; vÞ
simiðv; jÞ ¼
ð4Þ
oct ði; lÞ ¼ fct ði; jÞjð \ vr ¼ 1 simiðr; jÞ 4 0:9Þ [ ð \ vr ¼ 1 dgsimiðr; jÞ 4 20Þg;
ct ð1; jÞ ot 1 ð1; vÞ dgsimiðv; jÞ ¼ arctanð Þ ot 1 ð7; vÞ ct ð2; jÞ o ð2; vÞ
ð5Þ
t1
( ot ðu1 ; vÞ ¼
ct ði; jÞ; ot 1 ðu; vÞ;
minðsimiðv; jÞÞ o S \ dgsimiðv; jÞ oDeg ; minðsimiðv; jÞÞ 4 S [ dgsimiðv; jÞ 4Deg
u1 ¼ i ¼ 1; :::; 6
ð6Þ ( ot ð7; vÞ ¼
ot 1 ð1;vÞ arctanðcctt ð1;jÞ ð2;jÞ ot 1 ð2;vÞÞ;
minðsimiðv; jÞÞ o S \ dgsimiðv; jÞo Deg
ot 1 ð7; vÞ;
minðsimiðv; jÞÞ 4 S [ dgsimiðv; jÞ4 Deg
( ot ð8; vÞ ¼
at time t and any spatters in searching matrix ot 1 ðu; vÞ, spatter j was used as new information oct ði; lÞ at time t, and then added to the searching matrix ot ðu; vÞ as searching basis at time t þ 1. This process is shown as follows:
ot 1 ð8; vÞ þ 1;
minðsimiðv; jÞÞ o S \ dgsimiðv; jÞ o Deg
ot 1 ð8; vÞ;
minðsimiðv; jÞÞ 4 S [ dgsimiðv; jÞ 4 Deg
ð7Þ
i ¼ 1; :::; 6; l ¼ 1; :::; m ot ðu1 ; v þ lÞ ¼ oct ði; lÞ;
where ot ð7; vÞ is the moving direction of spatters and is expressed in degree, S is the shape similarity threshold 0.9, Deg is the angle similarity threshold 20, and ot ð8; vÞ represents the occurrence frequency of spatters. If there was no similarity between spatter j
u1 ¼ i ¼ 1; :::; 6;
l ¼ 1; :::; m
ð10Þ
ot ð7; v þ lÞ ¼ ot ð8; v þ lÞ ¼ 0
ð11Þ
ot ð9; v þ lÞ ¼ t
ð12Þ
ot ðu2 ; v þ lÞ ¼ oct ði; lÞ;
ð8Þ
ð9Þ
u2 ¼ i þ 9 ¼ 10; :::; 15;
l ¼ 1; :::; m
ð13Þ
where m is the number of new spatter information generated at time t, and ot ð9; vÞ represents the time when new information is created. An accumulator ot ð8; vÞ for recording spatter occurrence frequency was established to filter the noise information of the images. The information was considered as spatter information generated at time t only when its occurrence frequency reached SN¼ 5, and was then saved to the new-spatter matrix N(t). The spatters detection results are shown in Fig. 6(c) and (f). The
4
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
Fig. 5. Spatters image sequence captured by high-speed camera 1.
Fig. 6. Spatter detection. (a) Time t þ 0 spatters image P t ðx; yÞ. (b) Time t þ 0 spatters image without plume. (c) Time t þ 0 spatters detection. (d) Time t þ 1 spatters image P t þ 1 ðx; yÞ. (e) Time t þ 1 spatters image without plume. (f) Time t þ 1 spatters detection.
process is shown in Fig. 7.
process is expressed as follows: NðtÞ ¼ nt ða; bÞ;
Nð1Þ ¼ zeroð7; 1Þ
nt ða1 ; b þ1Þ ¼ ot ðu2 ; vÞ; nt ð7; b þ 1Þ ¼ ot ð9; vÞ; a1 ¼ u2 10 ¼ 1; :::; 6; o8;v ðtÞ ¼ SN
ð14Þ
SVðtÞ ¼
∑
4π½nf ð6; bÞ3
nf ð7;bÞ ¼ t
ð15Þ
where N(1) represents the initial value of new-spatter matrix. The new-spatter matrix made it possible to extract the total volume SVðtÞ, gray-value SGðtÞ and radius SRðtÞ of spatters generated at time t, which are given in Eqs. (16)–(18). The spatter detection
3
SGðtÞ ¼ ∑ nf ð4; bÞ; nf ð7;bÞ
SRðtÞ ¼
∑
nf ð7;bÞ ¼ t
nf ð6; bÞ;
;
t ¼ 1; 2; :::; f
t ¼ 1; 2; :::; f
t ¼ 1; 2; :::; f
ð16Þ
ð17Þ
ð18Þ
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
5
Fig. 7. Flowchart of spatter detection process.
Fig. 8. Comparison between weld seam and spatters volume detected from spatters image sequences.
where f is the total number of spatter images (in this case f¼2400), with the condition that SN þt o f . The results of spatters volume detection, spatters gray-value detection and spatters radius detection are shown in Figs. 8, 9 and 10 respectively. Ten frames of spatter images were chosen to illustrate the detail information of spatters and corresponding weld seam, as shown in Table 1. The bead width was narrow while larger spatters were generated from the molten pool, such as samples 5–7 in Table 1. The number of spatters created at different times is varying. The largest number of spatters created at the same time is 4, which is shown in sample 4 of Table 1. Also, sometimes there are no spatters generated from the molten pool, as seen in samples 2 and 8 in Table 1. Due to the limited depth of field of the camera, a few spatters are out of focus as shown in Fig. 3. However, this does not pose much influence on the inspection result. The proposed system automatically identified the spatters as soon as they were ejected from the keyhole, which is shown in Fig. 6, and saved the parameters of the spatters. The
size of the spatters changed as they moved, and the identification system synchronously updated spatter parameters to ensure the accuracy of spatter tracking. To be more specific, the system would save two groups of spatter parameters, including the original status parameters and the real-time status parameters. However, only the original status parameters were used as the basis for evaluating welding quality. This is because the spatters were within the focus of the camera (that is, within the welding area) at their original status, which therefore yields accurate parameters. It helps to avoid measurement error of the spatter size caused by defocus.
4. Conclusions Experimental results showed that there was a dramatic width decrease in the middle area of the weld seam, and the corresponding detection results indicated that spatter volume and gray-value within
6
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
Fig. 9. Comparison between weld seam and spatters gray-value detected from spatters image sequences.
Fig. 10. Comparison between weld seam and spatters radius detected from spatters image sequences.
this range were relatively large. These characteristics help to prove the effectiveness of applying a high-speed visual detection method to extract the instantaneous information of spatters during laser welding. By using the proposed spatter recognition detection algorithm to extract feature parameters of the spatters, the spatter volume and gray-value could be used as accurate quantification information for
evaluating weld seam quality. The experimental results indicated that there was a close relationship between the spatter feature information and the welding quality, since the weld seam width decreased as the numbers and mass of the spatters increased. Accordingly, an effective control over spatter numbers was crucial for improving laser welding quality.
D. You et al. / Optics and Lasers in Engineering 54 (2014) 1–7
Table 1 Details of spatters and weld seam during laser welding process. Sample Bead width (mm)
Image sequence (frame)
1
1.71
402
2 3
1.52 1.67
508 857
4
1.71
900
5
0.95
1099
6
1.00
1198
7 8 9 10
1.04 1.43 1.62 1.57
1203 1340 1685 2144
Spatter volume (pixel3)
Spatter grayvalue ( 104)
Spatter radius (pixel)
46.4 4.2 0 33.5 4.2 91.9 33.5 46.4 4.2 652.3 11.5 4.2 33.5 1480.3 0 11.8 4.2
1.2069 0.2092 0 0.7154 0.6078 1.0561 0.6272 1.0097 0.2201 3.0063 0.7730 1.1571 0.4672 11.3400 0 1.0326 0.5518
2.2 1 0 2 1 2.8 2 2.2 1 5.3 1.4 1 2 7.1 0 1.4 1
Acknowledgments This work was supported in part by the National Natural Science Foundation of China (51175095), the Guangdong Provincial Natural Science Foundation of China (10251009001000001 and 915100900 1000020), the Specialized Research Fund for the Doctoral Program of Higher Education of China (20104420110001) and the State Scholarship Fund from China Scholarship Council (2011844250). References [1] Katayama S, Kawahito Y. Laser direct joining of metal and plastic. Scripta Materialia 2008;59:1247–50. [2] Gao X, You D, Katayama S. Infrared image recognition for seam tracking monitoring during fiber laser welding. Mechatronics 2012;22(4):370–80. [3] Mei L, Chen G, Jin X, Zhang Y, Wu Q. Research on laser welding of highstrength galvanized automobile steel sheets. Optics and Lasers in Engineering 2009;47(11):1117–24. [4] Yan J, Gao M, Zeng X. Study on microstructure and mechanical properties of 304 stainless steel joints by TIG, laser and laser–TIG hybrid welding. Optics and Lasers in Engineering 2010;48(4):512–7. [5] Ni X, Zhou Z, Wen X, Li L. The use of Taguchi method to optimize the laser welding of sealing neuro-stimulator. Optics and Lasers in Engineering 2011;49 (3):297–304. [6] Pal K, Pal SK. Monitoring of weld penetration using arc acoustics. Materials and Manufacturing Processes 2011;26(5):684–93. [7] You D, Gao X, Katayama S. Multiple-optics sensing of high-brightness disk laser welding process. NDT & E International 2013;60:32–9.
7
[8] Rizzia D, Sibillanoa T, Calabresea PP, Anconaa A, Lugaràa PM. Spectroscopic, energetic and metallographic investigations of the laser lap welding of AISI 304 using the response surface methodology. Optics and Lasers in Engineering 2011;49(7):892–8. [9] Olsson R, Eriksson I, Powell J, Kaplan AFH. Challenges to the interpretation of the electromagnetic feedback from laser welding. Optics and Lasers in Engineering 2011;49(2):188–94. [10] Olsson R, Eriksson I, Powell J, Kaplan AFH. Advances in pulsed laser weld monitoring by the statistical analysis of reflected light. Optics and Lasers in Engineering 2011;49(11):1352–9. [11] Colombo D, Colosimo BM, Previtali B. Comparison of methods for data analysis in the remote monitoring of remote laser welding. Optics and Lasers in Engineering 2013;51(1):34–46. [12] Liu L, Chen M. Interactions between laser and arc plasma during laser–arc hybrid welding of magnesium alloy. Optics and Lasers in Engineering 2011;49 (9–10):1224–31. [13] Moradi M, Ghoreishi M, Frostevarg J, Kaplan AFH. An investigation on stability of laser hybrid arc welding. Optics and Lasers in Engineering 2013;51 (4):481–7. [14] Eriksson I, Powell J, Kaplan AFH. Melt behavior on the keyhole front during high speed laser welding. Optics and Lasers in Engineering 2013;51 (6):735–40. [15] Park H, Rhee S. Analysis of mechanism of plasma and spatter in CO2 laser welding of galvanized steel. Optics & Laser Technology 1999;31(2):119–26. [16] Park YW, Park H, Rhee S, Kang M. Real time estimation of CO2 laser weld quality for automotive industry. Optics & Laser Technology 2002;34 (2):135–42. [17] Kang MJ, Rhee S. The statistical models for estimating the amount of spatter in the short circuit transfer mode of GMAW. Welding Journal 2001;80(1):1–8. [18] Kang MJ, Kim Y, Ahn S, Rhee S. Spatter rate estimation in the short-circuit transfer region of GMAW. Welding Journal 2003;82(9):238–47. [19] Schwab G, Steele JPH, Vincent TL. Vision based spatter classification for contaminant detection. Welding Journal 2009;88(6):121–30. [20] Jäger M, Humbert S, Hamprecht FA. Sputter tracking for the automatic monitoring of industrial laser-welding processes. IEEE Transactions on Industrial Electronics 2008;55(5):2177–84. [21] Kaplan A.F.H., Norman P., Eriksson I., Analysis of the keyhole and weld pool dynamics by imaging evaluation and photodiode monitoring. In: Proceedings of LAMP2009 – the 5th international congress on laser advanced materials processing, 2009, Japan. [22] Katayama S, Mizutani M, Tarui T, Mori K. Monitoring and phenomena observation during YAG laser lap welding of Zn-coated steel sheets. Journal of Lanzhou University of Technology 2004;30(4):31–6. [23] Stache NC, Zimmer H, Gedicke J. Robust high-speed melt pool measurements for laser welding with sputter detection capability. Pattern Recognition 2007;4713:476–85. [24] Kawahito Y, Matsumoto N, Abe Y, Katayama S. Relationship of laser absorption to keyhole behavior in high power fiber laser welding of stainless steel and aluminum alloy. Journal of Materials Processing Technology 2011;211 (10):1563–8. [25] Zhang M, Chen G, Zhou Y, Li S. Direct observation of keyhole characteristics in deep penetration laser welding with a 10 kW fiber laser. Optics Express 2013;21(7):19997–20004. [26] Kaplan AFH, Powell J. Spatter in laser welding. Journal of Laser Applications 2011;23(2):032005. [27] Kawahito Y, Matsumoto N, Katayama S. High quality welding of stainless steel with 10 kW high power fibre laser. Science and Technology of Welding and Joining 2009;14(4):288–94.