Spatter Behavior in Al-Cu Adjustable Ring Mode Laser Spot Welding

Article information

J Weld Join. 2025;43(6):723-728

Publication date (electronic) : 2025 December 31

doi : https://doi.org/10.5781/JWJ.2025.43.6.15

Yunsoo Lee^*, Inhwan Koo^*, Woohyun Tak^*, Yonghun Lee^**, Jungho Cho^*,

* School of Mechanical Engineering, Chungbuk National University, Cheongju, 28644, Korea

** Digital Twin Team, LG Energy Solution, Cheongju, 28122, Korea

†Corresponding author: junghocho@cbnu.ac.kr

Received 2025 November 28; Revised 2025 December 5; Accepted 2025 December 8.

Abstract

1. Introduction

The transition from internal combustion engine vehicles to environmentally friendly electric vehicles has been accelerating. In particular, various approaches are being explored not only to improve the energy efficiency of individual battery cells, one of the essential components of electric vehicles, but also to arrange and join these cells safely and reliably1). Each cell contains electrode tabs, which are electrically connected to the module through busbars2). Aluminum and copper are widely used as materials for the tabs and busbars3), and laser welding is commonly applied for joining these dissimilar metals due to its advantages of high welding speed, strong joint strength, and low electrical resistance at the weld interface4,5).

However, laser welding inherently suffers from spatter generation, a long-standing issue attributed to the unstable dynamics of the keyhole, which undergoes repeated formation, maintenance, and collapse. When ejected metal spatter penetrates the interior of a cell, it may undergo repeated oxidation and reduction during charge-discharge cycling, forming dendritic structures6). These dendrites impede the movement of lithium ions between the anode and cathode, thereby shortening cell lifespan. Furthermore, dendritic structures can physically damage the separator and cause internal short circuits, which may lead to fire or explosion.

Therefore, reducing spatter during laser welding is a critical challenge. In industrial settings, however, performing exhaustive experiments across numerous process conditions to identify optimal parameters is impractical. Moreover, due to the demand for high productivity at high welding speeds, attempting a wide range of conditions to explore a stable process window is often inefficient. Consequently, fume extractors are frequently employed to capture spatter during welding and prevent its intrusion into the cell. Determining the appropriate extraction power and optimal collector placement requires knowledge of the spatter’s momentum at the moment of ejection.

Numerous studies have investigated spatter behavior in laser welding7-9). For example, Huang et al.10) used a high-speed camera to observe spatter generated during laser keyhole welding of aluminum. By applying adaptive filtering to perform binarization and assuming a constant-velocity linear motion model, they tracked spatter based on similarity in size, angle, and speed. They reported that spatter tends to eject forward and backward relative to the welding direction, and that the frequency of backward ejection increases with welding speed. You et al.11) developed a delay-time tracking algorithm using static and dynamic spatter features and analyzed the size and frequency of spatter with respect to welding speed. However, to the best of the authors’ knowledge, no prior study has investigated the spatter behavior occurring during adjustable ring mode (ARM) laser spot welding of dissimilar thin aluminum and copper sheets. Therefore, in this study, spatter behavior was observed using a high-speed camera, followed by image processing and tracking analysis to extract spatter size, direction, and velocity.

2. Experiment Setup and Image Processing

2.1 Experimental Method

The configuration of the experimental setup is shown in Fig. 1. A fiber laser source with a wavelength of 1070 nm (HighLight FL-ARM 8000, COHERENT) with the maximum power of 4 kW for core and ring beam, respectively, and a galvanometric scanner (SUPERSCAN IV-30, RAYLASE) were used. The core beam and ring beam powers were both set to 1.45 kW. Futhermore, beam spot diameter was 34.4 μm. Three nickel-plated Al1050 tabs (4.5 × 1.5 × 0.4 mm) were stacked on top of a nickel-plated C1100 busbar (7.5 × 1.5 × 4 mm), as illustrated in Fig. 2. The scan speed was fixed at 240 mm/s, and a spiral wobble pattern was applied. A high-speed camera (FASTCAM Mini) equipped with an 808 nm band-pass filter was positioned in front of the optical path to capture only the reflected illuminatiosn laser at that wavelength. The imaging conditions were determined based on preliminary trials. A frame rate of 20,000 FPS and a shutter speed of 1/64,000 s^-1 provided high-quality video with a resolution of 120 × 1280 pixels. Fifteen experimental trials were conducted using these imaging parameters.

Fig. 1

Experimental setup for adjustable ring mode laser spot welding

Fig. 2

Schematic illustration welding of Ni coated aluminum leads and copper busbar with spiral wobble

2.2 Image Processing Pipeline

Foreground extraction was conducted using MATLAB R2023a to isolate only the spatter for tracking. The pipeline shown in Fig. 3, which is based on Gaussian filtering and has been used in previous study9), was adopted. First, a Gaussian filter was applied to the original gray-scale image to suppress high-frequency components. The filtered image was then subtracted from the original image to enhance the prominence of the foreground. Global thresholding was subsequently performed to binarize the image, followed by a watershed transform to separate overlapping foreground regions.

Fig. 3

Image processing pipeline for spatter detection and binarization based on Gaussian filter

However, when applying this procedure, metal vapor and plume surrounding the spatter occasionally remained as noise, as illustrated in Fig. 4. This issue was particularly pronounced when the spatter was partially obscured by the plume. In addition, when the number of pixels constituting the foreground was small, the foreground was often removed during the processing sequence.

Fig. 4

Example of a binarization error in which the plume region is mistakenly segmented as foreground together with spatters

To address these issues, frames in which such problems were prominent were first upscaled three-fold using bicubic interpolation. Using the Video Labeler app, a bounding box enclosing the spatter was manually defined. Within this region, local thresholding based on a global threshold value was performed, while all pixels outside the bounding box were automatically set to zero to generate the final binary image. Furthermore, additional filtering was applied by calculating the average intensity of the objects and retaining only bright and distinct ones that exceeded a specific threshold. Moreover, based on the characteristic that objects deviating from the focal plane tends to contain internal zero-value regions, only objects completely filled with ones were preserved. Consequently, only definitive objects on the two-dimensional plane remain in the final pre-processed images, allowing for the accurate derivation of the statistical distribution of spatter behavior for designing fume extractor.

3. Spatter Tracking Algorithm

In this study, the Hungarian algorithm was employed to track individual spatter particles. After generating a list of binary images following the procedure described in the previous section, the loop illustrated in Fig. 5 was executed. In each iteration, the foreground objects detected in the current frame were matched with those detected in the previous frame by defining independent cost terms based on distance, direction, and area. Each cost component was assigned a corresponding weight, and their weighted sum was used as the total cost.

Fig. 5

Workflow of Hungarian assignment based multiobject tracking procedure applied to binary images

Using this cost matrix, the Hungarian algorithm was applied to compute the globally optimal assignment solution for the matching problem. Additionally, the characteristic motion of spatter, appearing near the bottom of the frame and traveling upward, was incorporated into the matching rule. Any potential match involving a foreground object located below the target spatter was assigned an infinite cost to prevent erroneous associations and ensure stable tracking.

Furthermore, even if a match was valid, it was rejected when the total cost exceeded a predefined threshold to reduce ID switching. Foreground objects identified as representing the same spatter were stored in an active track, and their tracking information was continuously updated in the main track list.

A critical factor in implementing the Hungarian algorithm for tracking is the determination of appropriate weighting factors, which must be obtained through empirical tuning. In this study, reliable tracking performance was achieved when the weights for distance, direction, and area were set to 1.5, 2.0, and 0.5, respectively, and the upper threshold for the total cost was set to 4. An example of the resulting tracking output under these conditions is shown in Fig. 6.

Fig. 6

Tracking result of the ejected spatters in consecutive frames and their trajectories

4. Results

For essach individual spatter particle, the frame index at which it appeared, along with the centroid coordinates and pixel area in that frame, were recorded. Each spatter was approximated as a circle, and its diameter was calculated using the circular area formula. The Euclidean distance between centroid coordinates in consecutive frames was computed, and dividing this distance by the inverse of the frame rate yielded the inter-frame velocity. The average diameter and average velocity for each spatter were then obtained by calculating the arithmetic mean of the corresponding data points.

The ejection angle was defined in the range of 0-180°, measured counterclockwise from the horizontal axis pointing to the right. Additionally, for each spatter, the centroid coordinates at the first frame of appearance were defined as the origin. The subsequent centroid positions along its trajectory were fitted using linear regression, and the slope of the regression line was defined as the mean ejection angle.

Finally, the tracking data obtained from each experimental run were aggregated and normalized by the total number of tracked spatter particles to compute their percentage distribution. The resulting distributions of diameter, velocity, and angle were visualized as histograms, as shown in Fig. 7.

Fig. 7

Statistical distributions of (a) mean diameter of spatter, (b) mean speed, and (c) mean angle

As shown in Fig. 7(a), most spatter particles were distributed below 200 μm in diameter. However, in rare cases, particles with diameters exceeding this range, up to 319 μm, were also observed, while the smallest measured diameter was 30 μm. The average velocities, presented in Fig. 7(b), spanned a wide range from 811 mm/s to 13,400 mm/s, with the majority of data concentrated between 1,000 mm/s and 6,000 mm/s. The measured ejection angles fell within the range of 9.8° to 160.5°.

A noteworthy finding is that, unlike previous observations in keyhole welding, where spatter is predominantly ejected forward and backward along the welding direction with relatively few occurrences in the perpendicular direction10), the present results did not exhibit a bimodal distribution. Instead, during laser spot welding, the spatter was found to disperse more uniformly in both directions relative to the vertical axis.

5. Conclusion

In this study, adjustable ring mode (ARM) laser spot welding of aluminum and copper sheets was performed, and spatter behavior was analyzed through image processing and tracking. Binary images were generated by separating spatter from the background, and the Hungarian algorithm-based tracking procedure was used to extract spatter size, ejection velocity, and angle. The main findings are summarized as follows:

1) Due to the coexistence of spatter and plume in the images, the Gaussian filter-based image processing pipeline alone was insufficient to consistently produce clean binary images. This limitation may be significantly mitigated by employing blob detectors designed to identify small, bright spatter regions more reliably.

2) The Hungarian algorithm, a matrix-based global optimization method, offers the advantage of being relatively straightforward to implement in code. However, achieving sufficient tracking robustness requires extensive trial-and-error parameter tuning. To further improve matching accuracy, ad-ditional features reflecting the physical motion characteristics of spatter, not limited to size, velocity, and direction, should be incorporated. Moreover, integrating the Hungarian algorithm with a Kalman filter is expected to enable more adaptive and predictive tracking performance.

3) Spatter ejection angles were found to range from approximately 10° to 160°, with ejection velocities varying from a minimum of 810 mm/s to nearly 13,000 mm/s. Spatter diameters exhibited substantial variation, ranging from 30 μm to 310 μm.

Article information Continued