Interaction between Local Shielding Gas Supply and Laser Spot Size on Spa�er Formation in Laser Beam Welding of AISI 304

: Background. Spa�er formation at melt pool swellings at the keyhole rear wall is a major issue for laser deep penetration welding at speeds beyond 8 m/min. A gas nozzle directed towards the keyhole, that supplies shielding gas locally, is advantageous in reducing spa�er formation due to its simple utilization. However, the relationship between local gas ﬂow, laser spot size


Introduction
The stainless steel AISI 304 is widely used in home appliances, the food industry, and many engineering applications due to corrosion resistance and weldability [1].A typical application for the use of AISI 304 is the production of continuously manufactured tubes for heating systems, water supply, and further technical applications [2].Most industrially available AISI 304 tubes have diameters <110 mm [2] and wall thicknesses ≤4 mm [3].The production of such tubes can be carried out continuously in a single production line, i.e., a strip of steel coming from a coil is formed into an open-seam tube by multi-stage forming rolls and welded with a longitudinal seam from one side in a single pass [4].Tungsten inert gas and laser beam welding are usually used for the single-pass welding process [5].However, the process speed of the tube manufacturing is limited to ≤5 m/min using tungsten inert gas welding [5].Other arc welding processes would limit the welding speed even further, such as metal inert gas welding (<2 m/min [6]) or submerged arc welding (<1 m/min [7]) as used in automated processing of AISI 304 components.Manual welding processes, like manual metal arc welding, are also not applicable to automated production.
In contrast, laser and electron beam welding of AISI 304 enables high welding speeds of 10 m/min and above [5] due to their high power intensities [8].Unlike electron beam welding, laser beam welding does not normally operate under a vacuum and therefore requires less effort in industrial production [9].Therefore, an enhanced cost-benefit efficiency of the laser welding process for the automated welding of continuously formed open-seam tubes is recognizable [10].In addition, modern solid-state lasers offer an increasing maximum power that could reach even higher welding speeds for AISI 304 [8] but result in an enhanced spa er formation [11].Spa er formation leads to a loss of mass, i.e., the cross-section of the weld is reduced, and requires reworking of components due to adhering spa ers [11].A number of methods are already demonstrating the potential to reduce spa er formation and thus enable high welding speeds.
In order to describe the behavior of AISI 304 during laser deep penetration welding (also known as laser keyhole welding) in the context of spa er formation, a wide range of welding speeds is described to enable a discussion of underlying phenomena and to determine different welding regimes.For lower welding speeds (typically below 5 m/min for AISI 304L) spa ers are formed at the keyhole rim [12].This spa er formation and detachment is undirected in contrast to higher welding speeds [13] and shows a comparatively small spa er size [10].Here a uniform heating of the keyhole surface is observed.Laser welding under this condition is called the "Rosenthal" regime [14].However, at high welding speeds (6 m/min [11] to 10 m/min [15]) the laser welding of AISI 304 results in spa er detachments from the highly fluctuating melt pool swellings at the keyhole rear wall [11].This phenomenon is referred to as a "single wave" regime [12].Spa er formation reduces the seam cross-section and adherent spa ers have to be removed from the sample surface after the welding process [10].An increased welding speed causes an enlargement of the keyhole area and so this process is referred to as an "elongated keyhole" regime.Both, the oscillation of the melt pool swellings and its elevation are decreased in this regime.In all different process regimes, the loss of mass increases with increasing welding speed until the spa er detachment changes to the formation of humps [16].This transition occurs in the "pre-humping" regime and results in the "humping" regime [12].
Spa er detachment in dependence on the required laser energies was described in [17]: to detach a spa er from the melt pool the kinetic energy of the fluid element Ekin,fl must be equal or greater than the sum of the kinetic energy of the spa er Ekin,sp and the surface energy of the melt Esur.To determine the prerequisite for the spa er detachment the kinetic energies were calculated from the melt density ρ, the spa er volume Vsp, and the upward melt flow velocity vfl or the spa er velocity vsp.The surface energy was determined on the basis of the surface tension σ and the spa er surface Ssp [17].A similar condition for spa er detachment was published in [10], but an additional momentum transfer based on the escaping metal vapor Evap was added.Thus, the authors transformed the condition into an energy balance (see Equation (1)).
The flow velocity of the fluid element vfl corresponds to the upward melt flow velocity at the keyhole rear wall for welding speeds ≤6 m/min [17].This melt flow mainly contributes to the spa er detachment from the melt pool swellings.Various experimental studies have dealt with the measurement of melt flow velocities.Using tracer particles and x-ray imaging, the flow velocity at the keyhole opening was determined to be 1.65 m/s in the event of a spa er detachment at the keyhole front for welding pure titanium.A lower flow velocity of 1.37 m/s did not result in a detachment [18].For the joining process of aluminum with a tophat distribution, spa er velocities up to 10 m/s with an average spa er diameter of less than 0.11 mm were measured [19].Investigations on steel showed maximum spa er velocities of 4.1 m/s for low welding speeds (≤3 m/min) [20].For welding AISI 304 with welding speeds ≥8 m/min, the spa ers detached from the melt pool swellings with a velocity of 3.5 m/s [21].However, the flow around the keyhole increases with the welding speed [8] and should be taken into account when investigating spa er detachment at high welding speeds.When the critical energy for spa er detachment is exceeded, a melt column is formed and a spa er is detached by necking.The necking of a droplet from the melt column can be described by the Plateau-Rayleigh instability [22] which is determined by the surface tension.It describes how a liquid stream breaks up into drops due to a reduction in surface area [23].Due to the emission of the melt in the spectral range of 820 nm to 980 nm, the occurring spa ers can be observed by high-speed imaging using an appropriate optical filter.Subsequently, a suitable algorithm detects them in the recordings [24] and thus their trajectories can be tracked [25].
Various approaches to reduce the spa er formation for high welding speeds are already known.In addition to reducing the ambient pressure [9], the superposition of laser intensities [26], the change of the focal position of the laser beam [27], and laser beam oscillations [28], the use of a gas flow already showed a positive effect for welding with CO2 lasers in 1984 [29].In addition to suppressing the plasma by changing the flow direction of the metal vapor for CO2 lasers, the spa er reduction with solid-state lasers while using a gas flow was investigated in [30].For welding speeds ≤5 m/min the melt flow was stabilized due to the gas flow and spa er ejections were completely suppressed [31].Recent investigations obtained a loss of mass reduction of up to 91% with an argon gas flow under low flow rates (≤4.8L/min) [32].In addition, the geometry of the keyhole was affected for partial [33] and full penetration welding [34] and for the welding of lap joints [35].The influence of the gas flow on the pressure balance in the keyhole was estimated in [33] combining investigations in [36,37].Also, in [38] the gas flow was mentioned as a mechanism to stabilize the keyhole and to reduce spa er formation.Humps appeared at the weld seam due to the increased momentum transfer from gas flow to melt pool at flow rates ≥6.4 L/min.Using even higher flow rates of up to 32 L/min resulted in an extended formation of the weld seam.Therefore, these flow rates should be avoided due to the reduced weld seam quality [33].
However, the reduction of spa er formation cannot be a ributed solely to the dynamic pressure of the gas flow, also the influence of the surface tension must be considered [39].Accordingly, the surface tension must be high enough to prevent spa er detachment [22].The surface tension of the melt pool depends particularly on the oxygen level of the melt [40], its Sulphur content, and temperature [41].Furthermore, elements with limited solubility in liquid metals and considerably weaker intermolecular bonding forces are most likely highly surface active [40].Due to the high surface activity of oxygen, the surface tension is decreased also at low oxygen levels [42].Consequently, a weld process under the exclusion of oxygen showed an altered melt flow in previous investigations [43].
A reduction of weld seam imperfections was also achieved by increasing the laser spot size.Larger laser spot sizes result in an increase in the melt pool [44] and keyhole size [11], whereas an elongated melt pool suppresses the spa er formation [44] because of a calmer melt pool behavior [45].An increase in laser spot size at the same power and feed rate results in a steeper ejection angle of the spa ers.This is a result of an altered metal vapor flow due to a greater keyhole front inclination [11].In addition to the reduction of spa ers, the suppression of humping by means of larger laser spot sizes was already demonstrated in [45].Furthermore, the keyhole closing Laplace pressure decreases as the keyhole radius increases [36].
The keyhole stability is often related to spa er formation.However, keyhole oscillation does not seem to be the cause of spa er detachment [46].The earliest calculations of the keyhole oscillation by means of appropriate models showed a radial keyhole frequency of 500 Hz [47].Further models to describe the process dynamics determined oscillation frequencies of the keyhole depth between 2000 Hz and 5000 Hz [48].Measurements of the laser-induced plume by photodiodes, evaluated by means of Fast Fourier Transform (FFT) for a simplified calculation of the amplitude and frequency [49], revealed significant amplitudes for 1000 Hz which resulted from keyhole oscillations [50].Volpp calculated the oscillation over the keyhole depth, whereby the oscillation frequency was most pronounced in the keyhole center.While varying the laser power and the beam profile, oscillations above 100 kHz were achieved by welding an aluminum alloy.It appeared that the frequency behaves contrary to the oscillation amplitude.If the frequency decreases, the amplitude increases [46].Robertson et al. demonstrated that material ejections could be comparatively low even at high oscillation amplitudes [22].For the use of a local gas flow, a similar keyhole oscillation frequency but an increased keyhole oscillation amplitude were determined in comparison to the reference process without gas flow [38].
In addition to high-speed imaging for process visualization, the measurement of the keyhole depth by means of Optical Coherence Tomography (OCT) is a promising method to explore the keyhole dynamics [51].Weld penetration depths of up to 5 mm were measured reliably for steel [52] and aluminum [53] using data filtering.The filtering was carried out for example by determining local peaks in the OCT data frequency distribution which correlate to the position of the keyhole bo om [52].
Various approaches were used to reduce the spa er formation, especially the use of a local gas flow has proven to be uncomplicated to use.An increase in laser spot size also revealed a spa er reduction because of an elongated melt pool.Furthermore, the previous investigations indicated that there is no direct correlation between keyhole frequency and spa er detachment.Comparative studies for varying laser spot sizes and simultaneous use of shielding gas to examine the effect on spa er formation are pending.Consequently, the effect of the gas flow and a changed Laplace pressure due to a modified keyhole radius are investigated in the following.
This publication focuses on the effect of gas flow under varying laser spot sizes dspot and welding speeds vw for partial penetration welding of AISI 304.The effect mechanisms induced by varying gas flow rates V̇ were investigated after an initial description of the reference process using high-speed imaging.This was done by measuring the loss of mass after the welding process, in addition to analyzing high-speed images, weld seam surfaces, and cross sections.Image processing of the lateral high-speed imaging was used to determine spa er quantity.The spa er velocity vsp was also quantified to investigate the energy balance of spa er formation and its defining parameters (see Figure 1).In addition, the cross-sectional areas of the spa ers were examined to estimate spa er volume Vsp and surface area Ssp.measured keyhole length l out of the high-speed images to eliminate it as a defining parameter.At the end of the investigation, only the upward melt flow velocity vfl at the keyhole rear wall was unknown, assuming a constant density ρ and a higher surface tension σ using gas flow.The upward melt flow velocity was determined by a further evaluation of the lateral high-speed images.Consequently, the procedure described allowed the determination of the defining parameters of the gas flow for spa er detachment at high welding speeds.

Materials and Methods
The experimental investigation was carried out on austenitic stainless steel sheets (AISI 304, 1.4301, X5CrNi18-10) with a thickness of 2 mm as bead-on-plate welds.The rectangular specimens had the dimensions 35 mm × 100 mm with a weld seam length of 75 mm.The chemical composition of the material is given in the following Table 1 and the selected material properties are in Table 2.
Table 1.Chemical composition of high-alloy steel AISI 304 in wt% [54].A disc laser (TruDisk 12002, Trumpf, Di ingen, Germany) with a wavelength of 1030 nm in combination with stationary welding optics (YW52, Precitec, Gaggenau-Bad Rotenfels, Germany) under an inclination angle γ of 85° in welding direction was applied for the welding experiments.The aspect ratio of the welding optics of 3:2 combined with a fiber core diameter of 200 µm resulted in a laser spot size of 300 µm.A laser spot size of 600 µm was utilized for comparative trials by changing the focal position towards the processing head.The different focal positions and the changed beam caustic were considered in the following analysis.The specifications of the laser system are shown in Table 3.The study was carried out for the welding speeds 8 m/min, 12 m/min, and 16 m/min based on the state of the art [33].Therefore, various process regimes with different process behavior were investigated [14].The laser power was adapted to achieve a comparable penetration depth of approx.1.5 mm for all experiments (see Table 4).Therefore, the welding depth was determined by means of OCT in preliminary tests.Figure 2 shows the experimental setup.Two high-speed cameras recorded each experiment.The first camera (1) was directed towards the top side of the sample at an angle of 50°.The second camera (2) was positioned parallel to the sample surface to obtain a side view of the spa er detachment.The keyhole and melt pool behavior on the top side were recorded by means of a high-speed camera (SA-X2, Photron, Tokyo, Japan) with a zoom lens (12xZoomLens, Navitar, Rochester, NY, USA).A high-power LED system (λ = 630 nm) illuminated the surface during the welding process.A bandpass filter for 630 nm with a half-power width of 40 nm was implemented in the optical path of the camera.The second high-speed camera (Phantom VEO 410L, Vision Research, Wayne, NJ, USA) filmed the spa er formation from the side view using a bandpass filter for 810 nm and a halfpower width of 12 nm.Due to the spa er glow, no additional illumination was necessary.Both cameras recorded the experiments with a framerate of 10 kHz.

Cr
A gas nozzle specially designed for the application supplied the shielding gas (argon) locally to the keyhole.The nozzle tip had a length of 15 mm and an inner diameter of 1.4 mm.Three linear guides and a rotary axis enabled a flexible positioning of the gas nozzle with an accuracy of 50 µm.Before the experiments, the gas nozzle was positioned to a spot weld of the laser beam on the sample surface using a coaxial pilot laser.Based on be er welding results under a trailing orientation of the gas nozzle in comparison to a leading orientation, the gas nozzle unit was mounted on the welding optics in the direction of the keyhole rear wall [39].The angle of incidence α between the nozzle tip and sample surface was set to 48° and the distance to the welding spot h to 5 mm, similar to previous investigations [34].Based also on the previous investigations, gas flow rates of 1.5 L/min and 5.0 L/min were used.The gas flow rate of 1.5 L/min was applied to significantly reduce the spa er formation.The second flow rate of 5.0 L/min was set at the transition area between spa er reduction and the formation of humps.
The loss of mass due to the welding process was determined to quantify the influence of the gas flow on spa er formation.A weighing of each sample before and after the welding process was carried out by a high-precision balance (PLJ 2000-3A, Kern & Sohn, Balingen, Germany).To obtain a parameter that was independent of the weld seam length, the measured loss was divided by the weld seam length.All experiments were performed three times.The data points depicted in the following graphs are mean values for the loss of mass and the error bars represent the standard deviation.The weld seam cross sections were inspected using microscope images to investigate the effect of the gas flow on the weld seam.The samples were therefore hot mounted, ground, and polished after cu ing.The microscope camera (Axiocam 208 color, Zeiss, Oberkochen, Germany) of an incident light microscope (Axio Scope.A1, Zeiss, Oberkochen, Germany) captured the cross sections.
The evaluation of the lateral high-speed images was carried out by means of two different algorithms for image processing.The first algorithm evaluated the upward velocity of the melt pool swellings from the lateral high-speed recordings.Therefore, the algorithm extracted a region of interest (ROI) from the original image (see Figure 3a) and located the melt pool swelling based on its maximum (see Figure 3b).Subsequently, it tracked the growth of the melt pool swelling directly behind the keyhole and determined the velocity based on the first two frames that showed the swelling.Thus, the velocity was assumed to be representative of the upward melt flow velocity at the keyhole rear wall in combination with the melt flow around the keyhole.The second algorithm tracked spa ers over multiple frames to determine their number and direction (see Figure 3c, more details are provided in [21]).The average distance for a detected spa er trajectory divided by the time defined by the spa er velocity.In addition, the program measured the visible area of each spa er.The multiple spa er areas of one trajectory were averaged to calculate the cross-sectional area of each spa er.The following evaluation of the spa er volumes and surfaces also considered a different deformation of the spa ers depending on the welding parameters.Due to the small distance of the spa ers between two frames at a frame rate of 10 kHz, only every fourth frame was considered in the investigation.The following depictions of the image superposition utilized a maximum intensity projection.The analysis of the melt pool velocity and the spa er detachment took place within the static area of the welding process that is described below.The keyhole behavior was described for varying parameters by means of high-speed imaging of the sample top side (see Figure 4a).The evaluation of the image data was done by a third image processing algorithm for this purpose.The first step was to binarize the image and select the ROI (see Figure 4b).Afterward, the morphological operation "closing" closed defects in the detected keyhole geometry.Finally, the algorithm measured the length and width of the keyhole opening.The centroid of the area was considered the center of the keyhole geometry through which the measurement lines had to pass.One-thousand frames were measured for each sample to evaluate the keyhole geometry on the specimen's top side.The keyhole behavior was studied based on the generated data of keyhole length by analyzing frequencies and amplitudes of the keyhole oscillation.The data of the 1000 frames was pre-processed by a median filter (3 × 1) to limit measurement irregularities.A FFT algorithm determined the normalized amplitudes and frequencies.In order to exclude temporal changes in the keyhole oscillation over the observation time, time-dependent changes in the keyhole oscillation were investigated.For this purpose, several calculations were performed for the 1000 frames with an area of 100 frames in each case.The first calculation was thus performed for the frame range one to 100.Subsequently, the range was shifted by one frame to the area two to 101.This was carried out 901 times to regard all 1000 frames.All calculated results were compiled in one three-dimensional diagram.Two exemplary time-wise windowed FFTs of the keyhole length without gas flow and welding speed of 12 m/min are depicted in Figure 5.No significant variations of amplitudes or frequencies occurred for both laser spot sizes during the investigation period.Consequently, FFTs of the entire range of 1000 frames were sufficient to evaluate the keyhole dynamic.
The previously described evaluations of the high-speed imaging were carried out within the area of a consistent penetration depth.This ensured the comparability of the experimental results with different process parameters.Reasons for a deviation of penetration depth were the acceleration and deceleration of the axis and also a necessary time to reach the steady state after starting the weld.For this purpose, the measurement values of an OCT were used.The beam paths of the measuring OCT beam and the laser beam were similar.The applied system (PRECITEC IDM) was mounted at the welding optics by means of a beam spli er module.The wavelength was 1.55 µm with a sampling rate of 70 kHz.To reduce the measurement noise, values with a difference greater than 0.1 mm to the previous value were filtered.Figure 6 shows three exemplary measurements of the OCT system for the two laser spot sizes and welding speed of 12 m/min with intervals for keyhole measurement and spa er evaluation.The OCT signal for the welding speed of 12 m/min and both laser spot sizes showed fluctuations at the beginning since the process was not yet in a steady state.For this reason, the start of the evaluation was set to 0.15 s.At the end of the welding process, an increase in welding depth was observed due to a deceleration of the axis.Thus, the end of the steady state was set to 0.35 s.For the keyhole measurement and oscillation analysis 1000 frames beginning at the start of the steady state were analyzed.The spa er and upward melt flow velocity analysis requires a constant weld seam length for the evaluation.Considering other welding speeds, a welding speed of 16 m/min had the shortest steady state in terms of process time and weld seam length, so the evaluation distance was set to 28.33 mm.No difference was found regarding the OCT signal comparing the reference experiment and the two flow rates.Thus, the analysis areas could be adopted also for the experiments with gas flow.

Reference Process
The loss of mass was utilized as a describing parameter for the weld quality of the reference process without applying local shielding gas. Figure 7a depicts the loss of mass over the welding speed for the laser spot size of 300 µm.The lowest loss of mass of 0.04 ± 0.02 mg/mm was measured at a welding speed of 8 m/min.The high-speed images show melt pool swellings at the rear wall of the keyhole opening.The melt pool swellings occurred periodically and subsequently merged with the melt pool.No spa ers detached from the melt pool swellings at this welding speed which is typical for the "Rosenthal" regime.Contrary to the state of the art, no spa ers were emi ed from the keyhole rim [12].This is a ributed to the lower energy input in the current investigation compared to the study in [12].However, the beginning formation of melt pool swellings already indicates a transition to the "single-wave" welding regime as described in [12].The characteristic and distinct melt pool swellings of the "single-wave" regime steadily formed behind the keyhole at a welding speed of 12 m/min and spa ers were constantly detached.The spa er detachment reduced the sample mass by 1.10 ± 0.05 mg/mm.Melt columns formed, which dissipated into spa ers due to the Plateau-Rayleigh instability condition [22].Consequently, the increase of the welding speed from 8 m/min to 12 m/min changed the mechanism of spa er detachment resulting in an enhanced spa er quantity.A further increase in welding speed to 16 m/min led to a minor reduction in loss of mass to 1.06 ± 0.02 mg/mm.The process behavior in this case was assigned to the "elongated keyhole" regime due to the observed elongation of the keyhole with increasing welding speed [12].Spa ers detached from melt pool swellings by necking the melt columns as already described in [22] similar to the process at 12 m/min.
The welding speed of 8 m/min also showed the lowest loss of mass with 0.07 ± 0.03 mg/mm using the laser spot size of 600 µm.However, the melt pool swellings were less pronounced.Therefore, the process behavior was assigned to the "Rosenthal" regime.There was also a larger melt pool in front of the keyhole compared to the smaller laser spot size.This was a ributed to a changed beam caustic due to the defocusing to increase the laser spot size.The loss of mass rose to 0.85 ± 0.05 mg/mm as the welding speed was increased to 12 m/min for the larger laser spot size.However, compared to the smaller laser spot size the loss of mass was slightly lower at this welding speed.A further increase of welding speed to 16 m/min led to a significant decrease in loss of mass to 0.55 ± 0.05 mg/mm.This effect was a ributed to the keyhole elongation with increasing welding speed typically known as the "elongated keyhole" regime [12].A keyhole elongation with increasing welding speed from 12 m/min to 16 m/min of 37% (0.50 mm to 0.69 mm; median values for 1000 measured keyhole lengths) was determined for the laser spot size of 300 µm while the laser spot size of 600 µm revealed an elongation of 46% from 0.59 mm to 0.86 mm.An extended keyhole length possibly decreased the interaction between escaping metal vapor and melt pool swellings.
As a result, despite the increased flow velocity around the keyhole due to the higher welding speed [8], fewer spa ers detached from the melt pool swellings, and the loss of mass decreased.A detailed discussion of the changing keyhole size is described in detail in Section 3.4.The welding process for the process regimes with spa er detachment from the melt pool swellings exhibited a lower loss of mass for the larger laser spot size in general.Similar behavior with reduced spa er formation due to an elongated melt pool has also been observed in investigations with larger laser spot sizes of up to 560 µm [44].

Effect of a Local Gas Flow on Loss of Mass
Figure 8 shows the loss of mass over welding speed and the resulting keyhole and melt pool behavior for both laser spot sizes depending on different flow rates of argon.The initial loss of mass of 0.04 ± 0.02 mg/mm for the reference process with a laser spot size of 300 µm at a welding speed of 8 m/min was not significantly reduced at flow rates of 1.5 L/min and 5.0 L/min due to the already negligible number of spa ers in the reference measurement (see Figure 8a).Furthermore, the spa ers detached from inside the keyhole.There was no spa er detachment from the melt pool swellings.A flow rate of 1.5 L/min enabled a 77% reduction of the loss of mass from 1.10 ± 0.05 mg/mm down to 0.25 ± 0.02 mg/mm for the welding regime with the highest loss of mass of all reference experiments for the welding speed 12 m/min.The further increase of the gas flow rate up to 5.0 L/min showed also a reduction down to 0.34 ± 0.05 mg/mm which, however, is a slightly decreased reduction of loss of mass compared to a flow rate of 1.5 L/min.The high-speed images (see Figure 8a) show a change in melt pool behavior under the influence of the gas flow.Similarly to the reference process without gas flow, melt pool swellings behind the keyhole occurred.However, significantly fewer spa ers detached from the melt pool swellings.If a spa er detached, it often fell back into the melt pool and was reabsorbed.Reabsorption was favored by an elongation of the melt pool due to the gas flow.Furthermore, the spa er trajectories were influenced by the gas flow and the changed flow of the escaping metal vapor.The spa ers also tended to be reabsorbed into the melt pool if they had low kinetic energy.The influence on the melt pool swellings and spa er formation was similar for both flow rates and will be explained in the following investigations.The subsequent rise of the welding speed to 16 m/min showed a decrease in loss of mass from 1.06 ± 0.02 mg/mm to 0.60 ± 0.02 mg/mm for 1.5 L/min while an increased gas flow rate of 5.0 L/min resulted in a loss of mass of 0.53 ± 0.02 mg/mm.Thus, the use of a local gas flow resulted in a positive effect on loss of mass for welding speeds of 12 m/min and 16 m/min.
A laser spot size of 600 µm revealed a similar loss of mass for the reference process at 8 m/min.An increase in welding speed up to 12 m/min resulted in a reduced loss of mass to 0.10 ± 0.04 mg/mm by applying a gas flow of 1.5 L/min.A similar reduction was achieved for a gas flow rate of 5.0 L/min with a reduction of 92% to 0.07 ± 0.02 mg/mm.The percentage of loss of mass was higher for this laser spot size because the value of the reference experiment without gas flow was already lower than for the smaller laser spot size.The effect of the gas flow on the melt pool behavior was similar to experiments with the smaller laser spot size.Spa ers were only detached occasionally.A welding speed of 16 m/min caused an initial loss of mass of 0.55 ± 0.05 mg/mm for the reference process while the local gas flow reduced it down to 0.15 ± 0.02 mg/mm for both gas flow rates.
At 8 m/min welding speed, the gas flow had a negligible effect on the loss of mass for both laser spot sizes.The highest reduction of loss of mass was obtained for a welding speed of 12 m/min for both laser spot sizes.However, a similar but slightly lower effect was noticed for a welding speed of 16 m/min and both laser spot sizes.The behavior can be explained by a shift of the process regime due to a changed keyhole and melt pool behavior depending on welding speed for the laser spot size of 300 µm.A greater keyhole front inclination due to a higher welding speed affected the interaction between metal vapor and the keyhole rear wall [56].The upward melt flow velocity at the keyhole rear wall was increased.
According to [10] the increase in flow velocity around the keyhole is nonlinear and therefore higher than the welding speed increase.Thus, higher welding speeds decrease the effect of the gas flow for the smaller laser spot size.In summary, the use of gas flow for 12 m/min and 16 m/min reduced the loss of mass for both laser spot sizes.The gas flow was particularly beneficial at a welding speed of 12 m/min.This welding speed was investigated further in the following.
The reduced loss of mass was also evident in the form of less spa er adhesions on the sample top sides for the welding speed of 12 m/min (see Figure 9).Consequently, the possibility that spa er adhesions on the top of the specimen distorted the loss of mass results can be excluded.In general, all specimen top sides revealed few spa er adhesions for the welding speed of 12 m/min suggesting a high kinetic energy of the spa ers due to a high velocity.Figure 10 shows typical weld seam cross sections for samples welded at 12 m/min, but different laser spot sizes and different gas flows.The cross sections show a distinct underfill due to the loss of melt volume for the experiments without gas flow, a laser spot size of 300 µm, and a welding speed of 12 m/min (see Figure 10a), which reflects the determined loss of mass.Using a gas flow rate of 1.5 L/min, sca ered spa er adhesions were observed and no underfill was visible due to the reduced loss of mass (see Figure 9a and Figure 10a).By increasing the gas flow rate to 5.0 L/min, the process behavior shifted into the "pre-humping" regime.Melt accumulations from displaced melt occurred at the weld seam surface due to an enhanced momentum transfer from the gas flow to the melt during the welding process.The penetration depth in the experiments using gas flow was comparable to the reference experiment.Increasing the laser spot size to 600 µm resulted in a wider weld seam and sca ered spa er adhesions on the sample surface for the reference experiment without gas (see Figure 9b and Figure 10b).The gas flow with a flow rate of 1.5 L/min enlarged the seam width compared to the reference experiment due to a widening of the keyhole top side (see Section 3.4).An increase in the gas flow rate to 5.0 L/min resulted in an excess weld reinforcement (see Figure 9).This indicated a transition to the "pre-humping" regime.Due to a larger melt pool behind the keyhole in comparison to the experiments with a smaller laser spot size, the momentum transfer of the gas flow did not result in a switch to the "pre-humping" regime.
In summary, the loss of mass determination is not distorted by adhering spa ers using the gas flow and is considered suitable for assessing the welding process with gas flow.To what extent the loss of mass reflects the reduction of the spa er formation is investigated in the following by evaluation of the lateral high-speed images.

Spa er Quantity, Area, and Velocity for Both Laser Spot Sizes
The spa ers were quantified by lateral high-speed images to investigate if the loss of mass correlated with the spa er quantity.A change in spa er size could result in differences between sample loss of mass and spa er quantity.The results were obtained for the seam parts with constant welding depth (see Section 2).The determined number of spa ers was divided by the welding seam length to obtain the number of spa ers per millimeter seam to provide a be er comparability to the loss of mass. Figure 11 shows the spa er quantities per millimeter for the welding speed of 12 m/min and different local gas flow rates in combination with typical lateral high-speed images.A spa er reduction of 76% was reached for the laser spot size of 300 µm and a gas flow rate of 1.5 L/min.A further increase in flow rate to 5.0 L/min reduced the number of spa ers per millimeter up to 80% which correlated well with the measured loss of mass (see Figure 11a).Figure 11a also shows a superposition of the lateral high-speed images.A visible reduction in spa er formation and a reduced angle of spa er detachment occurred with increasing gas flow.Using the laser spot size of 600 µm the gas flow even reduced the spa er quantity by 95% for 1.5 L/min and by 98% for 5.0 L/min.Again, these values correlated well with the measured loss of mass (see Figure 11b).The already observed low spa er quantity under the use of a gas flow is particularly evident in the superposition of the high-speed images.Also, the decrease of the spa er detachment angle with increasing gas flow rate can be seen nicely.In addition, the already described spa er reabsorption is discernible in the superposition images.In summary, the determination of the spa er quantities from the lateral high-speed images showed a good correlation with the loss of mass.Therefore, the use of a local gas flow results in a significant reduction in spa er formation during laser beam welding with different laser spot sizes.
Figure 12 shows the measured cross-sectional spa er areas Asp for different gas flow rates and laser spot sizes.The spa er areas were related to their volume, however, due to the irregular shape of the spa ers only their measured areas were regarded.In addition to the median and mean values in Figure 12, the gray boxes represent 50% of the measured values.Whiskers are limited to 1.5 times the interquartile range (IQR) in the figure.Furthermore, the outliers are displayed as individual points.For a laser spot size of 300 µm, similar median values were measured for the reference experiment without gas flow and for a gas flow rate of 1.5 L/min.For the flow rate of 5.0 L/min the median value of the spa er area was slightly increased by 0.08 mm 2 .However, the mean values were in the same range.Considering the measurement method and comparing the gray boxes with 50% of the measured values, the spa er area was assumed to be about constant for the experiments with a laser spot size of 300 µm.The experiments with the larger laser spot size of 600 µm showed similar median values to the experiments with the 300 µm laser spot size.Additionally, both gas flow rates did not exhibit a change in the spa er area.In summary, the spa er area and thus the spa er volume and surface were neither influenced by the gas flow rate nor by the laser spot size.In addition to the spa er size, the spa er velocity was of interest.Figure 13 depicts the determined spa er velocities for a welding speed of 12 m/min.At a laser spot size of 300 µm without gas flow (reference experiment) the median value of the spa er velocity was 2.9 m/s.The use of the flow rate of 1.5 L/min decreased the velocity to 2.3 m/s.Increasing the flow rate to 5.0 L/min led to a spa er velocity of 2.8 m/s, which was similar to the determined spa er velocity in the reference experiment without gas flow.For the experiment with the larger laser spot size of 600 µm, the spa er velocity was increased by 24% to 3.6 m/s for the reference experiment.Using a gas flow with 1.5 L/min decreased the velocity to 2.9 m/s.With a higher flow rate, the median value increased again to 3.9 m/s, which was also comparable to the experiment without gas flow.
In general, for both laser spot sizes, a velocity decrease was discernible for the flow rate of 1.5 L/min.In the Introduction, we already mentioned that oxygen is known to reduce the surface tension of high-alloy steels.Therefore, it was assumed that the gas flow increased the surface tension of the melt due to oxygen shielding.Considering the energy balance (see Equation ( 1)), the spa er velocity should decrease.An even higher flow rate of 5.0 L/min increased the spa er velocity again to the initial value of the experiment without gas flow.Probably, the higher momentum transfer of the gas flow accelerated the metal vapor flow.A modified flow of the metal vapor was already shown in [38].The increased metal vapor flow outweighed the increase of surface tension due to the protective gas and accelerated the spa ers.The spa er velocity was enhanced as a result.This general behavior was observed for both laser spot sizes.

Effects of the Local Gas Flow on the Keyhole Length
The following results describe the effect of the gas flow on the keyhole length.Figure 14 shows the measured keyhole length values for each parameter and the red line represents their median value.The gas flow of 1.5 L/min resulted in a decrease in keyhole length of 26%, from 0.50 mm to 0.37 mm at a laser spot size of 300 µm.Increasing the gas flow rate to 5.0 L/min led to a small length increase to 0.39 mm. Figure 14 also shows a decrease in the deviation of the measurement values for the flow rate of 1.5 L/min compared to the other two flow rates.It seems the deviation of keyhole lengths around the median was reduced by a comparably low gas flow.The median absolute deviation was reduced by 40% in comparison to the experiment without gas flow.Using the larger laser spot size of 600 µm, the average keyhole length was 0.59 mm for the reference experiment.That means the keyhole length was smaller than the laser spot size due to the changed beam caustic caused by defocusing.Applying a gas flow expanded the keyhole slightly to 0.62 mm for 1.5 L/min and to 0.64 mm for 5.0 L/min.Consequently, the gas flow had only a minor influence on keyhole length.Especially for 5.0 L/min the deviation in the measurement values increased.The subsequent frequency analysis will consequently show higher amplitudes.
The gas flow also changed the shape of the keyhole.Figure 15 shows the ratio of keyhole width to length of each measurement point.A ratio of one corresponds to a circular keyhole shape.If the ratio is less than one, the keyhole is elongated in the welding direction.It should be noted that the inclination angle of the welding optics resulted in an elliptical intensity distribution on the sample surface which favors an elliptical keyhole opening.Further information regarding the measured keyhole width is given in Appendix A. In general, the use of a local gas flow changed the keyhole shape for both laser spot sizes.The keyhole shifted from an elongated to a more circular shape using the flow rate of 1.5 L/min.This shift of the keyhole geometry was also described in [33].The change in keyhole geometry was explained by the pressures affecting the keyhole during the welding process.Therefore, the keyhole closing Laplace pressure is calculated on the basis of surface tension and keyhole geometry [36].The surface tension and the pressure of the melt flow mainly determine the keyhole pressure at high welding speeds [57].Consequently, the increase in surface tension resulted in an equalization of the pressures and the keyhole geometry changed to a more circular shape.The higher flow rate of 5.0 L/min resulted in a similar ratio as in the experiment without gas flow for both laser spot sizes.The momentum transfer of the gas flow to the keyhole rear wall outweighed the effect of the increased surface tension.Therefore, the keyhole was elongated in the direction of the gas flow.The amplitudes and frequencies of the keyhole length are depicted in Figure 16 for an analysis of the keyhole oscillation at a welding speed of 12 m/min.Therefore, the frequencies and amplitudes were determined from the measured values shown above, and the maximum amplitudes are marked in the diagrams.The figure depicts the relative oscillation amplitudes over the frequencies.Relative values were calculated for comparability of the oscillation amplitudes in relation to keyhole size.The respective amplitude was divided by the median value of the 1000 measured values resulting in a dimensionless value that enabled comparability between the different laser spot sizes.
The frequency of 1140 Hz had the maximum amplitude for a laser spot size of 300 µm and the reference experiment.A flow rate of 1.5 L/min decreased the frequency of the maximum amplitude by 40% to 680 Hz.The amplitudes were similar in both cases.The flow rate of 5.0 L/min had a maximum amplitude for the frequency of 70 Hz.The depicted weld seam in Figure 9 had periodic melt accumulations for this flow rate.Since these have not yet formed into complete humps, this process behavior was assigned to the "prehumping" regime.Counting of the melt accumulation resulted in an average frequency of 72 Hz for the three welded specimens.Thus, there is a distinct correlation between the keyhole oscillation in the welding direction and the occurrence of melt accumulations.No melt accumulations were discernible for the lower flow rates.
Using the larger laser spot size of 600 µm the flow rate of 1.5 L/min reduced the frequency of the maximum amplitude by 21% from 920 Hz to 760 Hz.The increase of the flow rate to 5.0 L/min enhanced the oscillation amplitude, as was the case for the smaller laser spot size.In addition, the frequency of the maximum amplitude was reduced to 540 Hz.The calculated oscillation frequency did not indicate any melt accumulations, which are also not visible in Figure 9.In general, the flow rate of 1.5 L/min reduced the oscillation frequencies of the keyhole for both laser spot sizes.Increasing the gas flow to 5.0 L/min influenced the oscillation of the keyhole length.The oscillation amplitudes were increased and the frequencies of the maximum amplitude were reduced.An inverse correlation between the oscillation frequency and amplitude was evident.The amplitudes were increased with a decreasing frequency and vice versa.This has already been investigated by means of calculations in [46].The keyhole oscillation is influenced by the changed Laplace pressure due to the shielding of oxygen using a gas flow.Furthermore, an excessive flow rate (in this case 5.0 L/min) increased the oscillation amplitude due to a higher momentum transfer to metal vapor, melt pool, and keyhole rear wall.However, the oscillation frequency does not seem to be the cause of the significant spa er reduction.The oscillation frequency was still distinct, but the spa er quantity was significantly reduced for a flow rate of 1.5 L/min.Further considerations are required to analyze the occurring effect mechanisms using a local gas flow.

Velocity of the Arising Melt Pool Swellings
The energy balance in Equation ( 1) includes the upward melt flow velocity in addition to the already described spa er velocity and size.The upward melt flow velocity was determined by measurements of the rising melt pool swellings behind the keyhole.Figure 17 shows the measured values for the rise of melt pool swellings at a welding speed of 12 m/min, according to spa er size, spa er velocity, and keyhole width-to-length ratio.
For the reference experiment, the median upward melt flow velocity is 2.18 m/s for the smaller laser spot size of 300 µm.Using the gas flow 1.5 L/min and 5.0 L/min did not influence the median value of the upward melt flow velocities.Only the high-velocity outliers were reduced by means of a flow rate of 1.5 L/min.So, the average value was the lowest for this flow rate.In general, there was no significant influence of the gas flow on the upward melt flow velocity due to the equal medians and the similar distributions of the measured values in the gray boxes.Increasing the laser spot size to 600 µm the upward melt flow velocities were increased by 50% to 3.26 m/s compared to the smaller laser spot size.A flow rate of 1.5 L/min resulted in a significant reduction of the median value by 56% to 1.45 m/s.With further rise of the flow rate to 5.0 L/min the value increased slightly to 1.81 m/s.The mean values display the same trend but were distorted by the outliers for the evaluation.Consequently, the velocity of the melt pool swellings resulting from the upward melt flow at the keyhole rear wall and the melt flow around the keyhole were significantly reduced by both gas flow rates.The generally slightly lower flow velocities compared to the spa er velocities for both laser spot sizes were a ributed to two causes.On the one hand, spa ers did not detach from every melt pool swelling (see Figure 8).However, each melt pool swelling was considered to determine the upward melt flow velocity.On the other hand, there was an additional increase in the kinetic energy of the spa ers due to the escaping metal vapor that occurred for every spa er detachment.
In general, the gas flow had only a slight influence on the upward melt flow velocities for the laser spot size of 300 µm.A significant deceleration of the melt flow was evident The gas flow primarily reduced the keyhole length for the laser spot size of 300 µm and increased the keyhole width for the laser spot size of 600 µm.Especially the keyhole widening for the larger laser spot size (see Figure A1) influenced the flow conditions for melt flow in the melt pool and the flow of metal vapor inside the keyhole.Therefore, the melt flow upwards the keyhole rear wall slowed down due to a wider cross-sectional area.However, since there is a reduction of loss of mass and the spa er quantity for both laser spot sizes, the interaction of the effect mechanisms had to change relating to keyhole geometry, upward melt flow velocity, and surface tension.

Discussion of the Experimental Results
The conclusive consideration of the effect mechanisms was done with regard to the energy balance (Equation ( 1)). Figure 18a shows the energy balance of spa er detachment and how the defining parameters were affected by the local gas flow and laser spot size.The momentum transfer based on the escaping metal vapor Evap was supposed to be equal for the reference and gas flow experiments during the investigation.The energy of the fluid Ekin,fl, which is composed of the melt density ρ, the spa er volume Vsp, and the upward melt flow velocity vfl, is mainly determined by the quadratic influence of the upward melt flow velocity.In particular, the gas flow reduced the upward melt flow velocity for the larger laser spot size of 600 µm.It was assumed, that the density of the melt was not affected by the shielding gas.Furthermore, the effect of the gas flow on the cross-sectional area of the spa ers (see Figure 12) and thus on the spa er volume Vsp were minor.The kinetic energy of the spa ers is mainly determined by the quadratic influence of the spa er velocity.The velocity decreased using gas flow (especially 1.5 L/min) for both laser spot sizes and was the result of the influence of the keyhole and melt pool behavior.The inert gas also changed the surface energy Esur required for spa er detachment, which is determined by surface tension σ and spa er surface Ssp.The spa er surface is related to the cross-sectional area of the spa ers and was therefore supposed to be constant, as described before.The surface tension is increased by the shielding of oxygen as already explained.Thus, the use of an inert shielding gas reduces the oxygen content in the atmosphere surrounding the melt pool and increases the energy required for spa er detachment.There were significantly fewer spa ers which detached from the melt pool swellings, and the kinetic energy of the few spa ers decreased.

V ̇
In summary, the gas flow diminishes the spa er formation and spa er velocity vsp by reducing the upward melt flow velocity and increasing the surface tension of the melt.A cooling of the melt pool by the gas flow and an additional heat input to the melt pool by the distraction of the metal vapor flow is considered negligible.Due to the minor influence on the upward melt flow velocity for the laser spot size of 300 µm, the primary reason for the spa er reduction is assumed to be the oxygen shielding.Consequently, the surface tension of the melt pool swellings increases and they are more pronounced if using the gas flow for both laser spot sizes (see Figure 18b).In addition to the increased surface tension, the upward melt flow velocity is decreased by the gas flow as a result of a keyhole widening for the laser spot size of 600 µm.The spa er reduction is therefore not only influenced by the surface tension but also by a changed upward melt flow velocity due to a manipulated keyhole geometry.The following two equations of the energy balance for spa er detachment summarize the investigation results considering the energy of the escaping metal vapor Evap, the fluid flow Ekinf,fl, the spa ers Ekin,sp, and the surface Esur (see Equation ( 2) and ( 3)).dspot = 300 µm, V ̇ = 1.5 L/min: E + E , = E , ↓ +E ↑ (2) dspot = 600 µm, V ̇ = 1.5 L/min: E + E , ↓= E , ↓ +E ↑ (3)

Conclusions
The present study describes the effect of a local gas flow of inert gas on spa er detachment during laser partial penetration welding of high-alloy steel AISI 304 (1.4301, X5CrNi18-10) for the welding speeds 8 m/min, 12 m/min, and 16 m/min.A gas nozzle provided an argon flow with flow rates of 1.5 L/min and 5.0 L/min.The investigations were carried out for laser spot sizes of 300 µm and 600 µm.The gas flow reduced the sample loss of mass by up to 92% for both laser spot sizes and a welding speed of 12 m/min.In addition to the determination of the loss of mass due to spa er detachment, high-speed imaging from the top and lateral views was used to analyze the spa er formation and keyhole behavior.Therefore, a similar reduction of spa er formation was also found in the analysis of the spa er quantities.The cross-sectional spa er areas and velocities were investigated to determine the cause of the spa er-reducing effect of the gas flow.In particular, the gas flow rate of 1.5 L/min caused a velocity reduction for both laser spot sizes with similar spa er sizes.A measurement of the keyhole geometry was carried out for further causal research.The gas flow resulted in a keyhole length reduction for the laser spot size of 300 µm and reduced the oscillation frequencies of the keyhole for both laser spot sizes.However, the reduction of the oscillation frequency was not considered to be the main cause of the reduced spa er detachment due to the unchanged high oscillation frequencies.The velocity of the occurring melt pool swellings was determined in a further investigation.The gas flow had only an insignificant influence on the upward melt flow velocity for the smaller laser spot size of 300 µm.A significant velocity reduction of up to 56% was evident using the gas flow for the laser spot size of 600 µm.
The effect of the gas flow Is a ributed to an increase in surface tension of the melt pool due to oxygen shielding for the smaller laser spot size.The kinetic energy of the melt required spa er detachment is increased and fewer spa ers detach from the melt pool swellings.Increasing the laser spot size to 600 µm affects the keyhole geometry in such a way that the melt flow at the keyhole rear wall slows down and the fluid energy decreases.Combined with the increased surface tension of the melt pool, fewer spa ers are detached from the melt pool swellings.If the local gas flow is too high, the increased momentum transfer from the gas flow to the melt pool results in the formation of melt accumulations especially for the smaller laser spot size.
In summary, the study extends the understanding of the effect of a local gas flow on laser beam welding and spa er formation using different laser spot sizes.A reduced spa er formation allows higher welding speeds in industrial production with the same weld seam quality as at lower welding speeds.This increases the profitability of manufacturing companies by using a local gas flow during laser beam welding.Further investigations will focus on the effect of the gas flow on keyhole geometry.High-speed synchrotron X-ray imaging will be used to measure the keyhole depth, inclination angle, and keyhole length at different keyhole positions.Furthermore, the melt flow at the keyhole rear wall can be determined directly.

Figure 1 .
Figure 1.Approach of the investigation: (a) schematic illustration of spa er formation; (b) defining parameters of the energy balance for spa er detachment.

Figure 2 .
Figure 2. Schematic illustration of the experimental setup for high-speed imaging for the top side (1) and side view (2) under the use of a local gas flow.

Figure 3 .
Figure 3. Image processing of spa er formation: (a) single frame from the high-speed recording; (b) determination of the upward melt flow velocity at the keyhole rear wall; (c) superposition of the evaluated frames with spa er trajectories.

Figure 4 .
Figure 4. Image processing of the keyhole geometry to determine its length and width: (a) highspeed imaging raw image; (b) identified geometry of the keyhole opening.

Figure 5 .
Figure 5. Windowed FFT of the keyhole length for 12 m/min.

Figure 6 .
Figure 6.Evaluated weld seam parts for spa er and keyhole investigation with static process behavior for the welding speed of 12 m/min.

Figure 7 .
Figure 7. Loss of mass, keyhole and melt pool behavior of the reference welding process for various welding speeds: (a) laser spot size of 300 µm; (b) laser spot size of 600 µm.

Figure 8 .
Figure 8. Loss of mass for various welding speeds, keyhole and melt pool behavior under the use of local gas flow at a welding speed of 12 m/min: (a) laser spot size of 300 µm; (b) laser spot size of 600 µm.

Figure 9 .
Figure 9. Weld seam surfaces for a welding speed of 12 m/min for the reference experiment and using a gas flow: (a) Laser spot size of 300 µm; (b) Laser spot size of 600 µm.

Figure 10 .
Figure 10.Exemplary cross sections from the center of the sample for a welding speed of 12 m/min of the reference and gas flow experiment: (a) Laser spot size of 300 µm; (b) Laser spot size of 600 µm.

Figure 11 .
Figure 11.Spa er quantity and superposition of the lateral high-speed images for a welding speed of 12 m/min: (a) laser spot size of 300 µm; (b) laser spot size of 600 µm.

Figure 12 .
Figure 12.Cross-sectional areas of the spa ers in laser spot size dependence and gas flow rate for the welding speed of 12 m/min.

Figure 13 .
Figure 13.Analyzed spa er velocities in laser spot size dependence and local gas flow for the welding speed of 12 m/min.

Figure 14 .
Figure 14.Measured keyhole length for different laser spot sizes and gas flow rates at a welding speed of 12 m/min.

Figure 15 .
Figure 15.Width-to-length ratio of the keyhole at different gas flow rates and laser spot sizes for the welding speed 12 m/min.

Figure 16 .
Figure 16.Oscillation of the keyhole length for different laser spot sizes and gas flow rates at the welding speed of 12 m/min.

Figure 17 .
Figure 17.Upward melt flow velocities determined from the melt pool swellings for 12 m/min.
using the gas flow for the larger laser spot size of 600 µm.The changed keyhole geometry by the gas flow was regarded as the cause for the different upward melt flow velocities.

Figure 18 .
Figure 18.Effect mechanisms: (a) energy balance with the effect of local gas flow and laser spot size; (b) schematic illustration using gas flow for various laser spot sizes.

Figure A1 .
Figure A1.Measured keyhole width for the welding speed 12 m/min.

Table 2 .
Selected material properties of high-alloy steel AISI 304.

Table 3 .
Laser properties and se ings.

Table 4 .
Laser spot size dependence on laser power and welding speed.