Correlative spatter and vapour depression dynamics during laser powder bed fusion of an Al-Fe-Zr alloy

Spatter during laser powder bed fusion (LPBF) can induce surface defects, impacting the fatigue performance of the fabricated components. Here, we reveal and explain the links between vapour depression shape and spatter dynamics during LPBF of an Al-Fe-Zr aluminium alloy using high-speed synchrotron x-ray imaging. We quantify the number, trajectory angle, velocity, and kinetic energy of the spatter as a function of vapour depression zone/keyhole morphology under industry-relevant processing conditions. The depression zone/keyhole morphology was found to influence the spatter ejection angle in keyhole versus conduction melting modes: (i) the vapour-pressure driven plume in conduction mode with a quasi-semi-circular depression zone leads to backward spatter whereas; and (ii) the keyhole rear wall redirects the gas/vapour flow to cause vertical spatter ejection and rear rim droplet spatter. Increasing the opening of the keyhole or vapour depression zone can reduce entrainment of solid spatter. We discover a spatter-induced cavity mechanism in which small spatter particles are accelerated towards the powder bed after laser-spatter interaction, inducing powder denudation and cavities on the printed surface. By quantifying these laser-spatter interactions, we suggest a printing strategy for minimising defects and improving the surface quality of LPBF parts.

Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

Introduction
Laser powder bed fusion (LPBF) is a prominent additive manufacturing (AM) technology that manufactures near-net shape metallic components with exceptional design freedom, minimal lead time, and no tooling cost, on a layer-by-layer basis [1][2][3][4].However, the adoption of LPBF for safety critical applications is hindered by the challenge of achieving defectlean, high-density metallic components that meet critical quality standards [5,6].
High-quality LPBF components can be superior to castings in defect levels and mechanical properties when appropriate process parameters are used [7]; however, currently they may not reach the surface quality and defect levels of components machined from wrought products.Aluminium LPBF is currently an attractive solution for the replacement of machined components only in a few niche applications, as machining of aluminium is relatively inexpensive for large numbers and waste can be recycled efficiently.Industrial adoption of aluminium LPBF becomes attractive for production of short lead time prototypes such as spare parts with non-machined surfaces, or for replacing complex systems of component assemblies with a single AM component [8,9].
Casting alloys such as AlSi10Mg [10] and high-strength alloys, e.g.6xxx and 7xxx series [11], may suffer from poor mechanical performance or poor processability in AM applications.To achieve superior properties (strength and conductivity) with good LPBF processability, Constellium Aheadd ® CP1 aluminium was designed specifically for LPBF based on the Al-Fe-Zr rapid solidification alloy system and registered with the Aluminium Association as AA8A61.50[12].The alloy design simplifies LPBF production.After printing, a precipitation hardening treatment, typically at 400 • C for 4 h, is used to reach a peak yield strength of ∼300 MPa and high levels of thermal (180 W•m −1 •K −1 ) and electrical (up to 30 mS•m −1 ) conductivity [13].The printed components of AA8A61.50 usually show high intrinsic defect tolerance and low residual stress, resulting in excellent fatigue performance even on non-machined samples with printing flaws.Reduced defects (or roughness) in the surface regions should bring further improvements on the fatigue performance of the LPBF components.These surface defects are often associated with spatter formation [1,[14][15][16][17][18][19] during LPBF and hence a better understanding of their evolution mechanisms is required to prevent them.
Four types of ejecta have been found in LPBF processes based on their formation mechanisms: (i) solid spatter, (ii) entrainment spatter, (iii) powder agglomeration, and (iv) liquid droplet spatter [20].Solid spatter forms when powder particles are directly ejected by the high-pressure metallic vapour plume at the laser-matter interaction zone [16,21].Entrainment spatter occurs when particles are entrained by a combination of inward shielding gas flow induced by the high-speed vapour plume and the wakefield generated by the scanning laser beam [22].Powder agglomeration occurs when powder particles exhibit modified surface chemistry and a high cohesion index owing to the presence of adventitious carbon, moisture, oxidation, or partial sintering [23].Liquid droplets mainly form from the front wall of the turbulent melt pool due to the vapourinduced recoil pressure (P r ) [22,24].
Liquid droplet and powder agglomeration spatter, usually exhibit a larger particle size than the feedstock, and are frequently identified as a major type of defect during LPBF [25].Oversized spatter can adhere to the surface of AM parts, increasing both surface defects and roughness [15]; they can also be trapped in the powder bed in subsequent build layers, leading to lack-of-fusion porosities [26].Spatter may undergo oxidation [27], which lowers the recyclability and reusability of powder [28].The surface oxides can inhibit particle fusion and promote the formation of porosity [23], decreasing the density of the LPBF parts.However, the role of small spatter on the final product properties remains unclear.
Much of the existing literature captures the spatter and vapour plume over the powder surface using in situ highspeed optical [14,22] or Schlieren imaging [29], which lack details on the changes of vapour depression or keyhole morphology during LPBF.Synchrotron x-ray sources can be used to capture both spatter and melt pool dynamics during LPBF at exceptional temporal resolution (up to 1 MHz) [30][31][32][33][34]. Combining synchrotron x-ray radiography and Schlieren imaging suggests that the vapour ejection angle is highly related to the morphology of the depression zone/keyhole [35]; however, the influence of keyhole morphology on spatter formation remains not well understood.
Here, we employed synchrotron radiography at a 40 kHz image acquisition rate to monitor and quantify the relationship between spatter ejection and vapour depression dynamics during LPBF of AA8A61.50 under varying industrially relevant process parameters.Our study emphasises that small spatter can induce unfavourable surface defects on LPBF parts, thereby highlighting the critical importance of understanding its behaviour to devise new printing strategies to avoid spatter and enhance LPBF quality.

In situ synchrotron x-ray imaging
The LPBF processing parameters (table 1) were selected to cover a range of linear energy densities (E l ) ranging from 210 J•m −1 to 840 J•m −1 .E l is calculated using: where P is laser power, v is the scan speed and D is the duty cycle.When the laser operates in continuous wave (CW) mode, its duty cycle (D) is 100%.For pulse width modulation mode, we multiply a duty cycle of 80% to normalise the laser power.The duty cycle is defined by the ratio of laser exposure time, t e at each of the semi-discrete points forming a scan vector with a delay time, t d between consecutive points: We used high E l conditions to replicate keyhole mode melting for contour scans to reduce surface roughness.Despite sub-surface keyhole porosity, such conditions can potentially improve the fatigue performance of AA8A61.50 parts.In contrast, we used low E l conditions to replicate conduction mode melting for maximising LPBF productivity.However, a low E l can lead to the formation of lack of fusion defects and increase scatter in mechanical properties.Laser welding on AA8A61.50 was also performed under the same processing conditions for comparison.The laser spot size is 80 µm.
The abovementioned experiments were performed using bespoke AM machine, called the Quad-laser in situ and operando process replicator (Quad-ISOPR).The Quad-ISOPR comprises four lasers and the scan head of a RenAM 500Q (Renishaw plc., UK) system combined with a custom-built environmental chamber.A substrate with a 1 mm throughthickness and 15 mm height is mounted in the chamber onto which a thin layer of the powder is automatically deposited, see figure 1(a).The substrates were produced by LPBF using the same AA8A61.50powder feedstock to mimic processing of a thin powder layer on top of a 15 mm tall AM part.Each substrate was mounted in a configuration depicted in [36].Argon shielding gas was used during LPBF and laser welding.
The in situ experiments were carried out at the European Synchrotron Radiation Facility's (ESRF) high-speed imaging beamline ID19 [37].This beamline employs two U32 undulators to produce a polychromatic hard x-ray beam with a mean energy of ∼30 keV.Once the incident beam passes through the sample and glassy carbon windows on either side of the enclosure, the attenuated x-ray is converted by a LuAG: Ce scintillator and emits visible light.The visible light image is magnified by a 5× objective and then captured by a high-speed camera (Photron FASTCAM SA-Z 2100K, Photron Ltd, Japan) at a framerate of 40 kHz.The sample thickness was chosen to achieve ∼40% x-ray attenuation through the substrate and glassy carbon stack according to the Beer-Lambert law.The field of view (FoV) was 1024 px (width) × 512 px (height) with an isotropic pixel size of 4.3 µm, giving a FoV of 4.4 mm × 2.2 mm.Each scan track was 4 mm in length and was centred within the FoV of the x-ray image to capture the process dynamics during the onset, steady state, and end of laser scanning.The laser trigger, x-ray shutter opening, and image acquisition were synchronised using a PandABox (Quantum Detectors, Oxford, UK).

Surface roughness measurement
The surface profiles of all the build samples were imaged by a high-resolution optical microscopy (Keyence VHX-7000, Keyence, Japan).The surface roughness (R q ) was calculated through the measured height profiles [38] (see supplementary figure S2) via: R q = √ ∑ n , where h x is the measure heights along the building track, h m is the mean height and n is the measured position number.

Image processing and quantification
The acquired radiographs were processed with ImageJ [39] and MATLAB© using a processing pipeline illustrated in figures 1(b)-(d).The radiographs were first corrected using FFC = I 0 /Flat ave [40], where FFC is the flat-field corrected image, I 0 is the raw radiograph, and Flat ave is the average of 100 flat-field images.The dark signal is already subtracted by the camera software.A custom background subtraction (BS) was applied to remove stationary objects.The solid spatter and liquid droplet particles were segmented by applying a Gaussian filter and manual threshold value.The particle movement was tracked and quantified using the TrackMate plugin [41] in ImageJ by applying a mask detector and advanced Kalman Tracker (see examples in supplementary videos V1 and V2).The maximum number of frames over which a particle identified per track was set at 3. To capture the steady state keyhole or depression zone morphology, the BS images were re-framed to focus on the melt pool region in an Eulerian frame of reference using a bespoke Python script.

Identification of solid spatter, powder agglomeration and liquid spatter
We have classified five types of ejecta during LPBF and laser welding of AA8A61.50 using in situ x-ray imaging according to their formation mechanisms: ejected solid spatter, entrained solid spatter, powder agglomeration, jet spatter and large liquid droplet, see x-ray radiographs in figures 2(a) and (b) and the corresponding schematic diagram in figure 2(c).
Two types of solid spatter are observed during LPBF: (1) the un-melted powder surrounding the melt pool that is ejected directly by the vapour plume under high vapour pressureterms ejected solid spatter; (2) entrained solid spatter occurs when powder particles are entrained by a vortex flow of argon gas induced by the vapour plume and the wakefield of the scanning laser beam.The solid spatter accounts for >82% of the total ejecta.Their size and circularity (as shown in figures 2(h) and (i)) match the powder feedstock size distribution (see supplementary figure S1).
The powder particles close to the melt pool boundary can be partially melted or sintered by the scanning laser beam, forming powder agglomerations.This agglomerate spatter exhibits a larger particle size and a lower circularity (<0.88) compared to the solid spatter, see figures 2(h) and (i), respectively.Most of these agglomerates were carried away from the laser-matter interaction zone and then entrained by the argon gas flow.
The formation of liquid spatter can be caused by the complex interplay among the melt flow, the high vapour pressure/recoil pressure generated by metal vaporisation, and the surface tension of the liquid metal.The liquid spatter generated The localised recoil pressure on the front keyhole wall first generates a small melt protrusion.When the protrusion reaches the rim of the front keyhole wall by the melt flow, it pinches off causing jet spatter.Our observed jet spatter mechanism is different to the bulk-explosion event from [24].Such jet spatter observed during laser welding without powder is smaller or equivalent to the solid spatter, however, it only accounts for ∼2% of the total number of spatter instances under the conditions studied.
Apart from the above small jet spatter, we observed a new mechanism by which large liquid droplets form at the rear rim of the keyhole.Under keyhole melting conditions, the Marangoni flow [42] pushes the liquid metal near the keyhole bottom towards the top melt pool surface.The intense vapour flow (blue solid arrow) induces vapour pressure towards the rear keyhole wall (figure 2(d)).Previous work estimates that the vapour jet velocities can reach from 20 m•s −1 [43] to 300 m•s −1 [29] whereas the melt flow can only reach up to 5 m•s −1 [43].The differences in velocities and densities between metal vapour and liquid metal induce a perturbation at the gas/vapour-liquid interface, a.k.a.Kelvin-Helmholtz instability [44,45], which also imposes a shear force along the gas-liquid interface (orange dotted arrow).The combination of the Kelvin-Helmholtz instability and Marangoni convection (red dotted arrow) promotes the formation of a large liquid protrusion at the rear wall (figure 2(e)).If the liquid protrusion is sufficiently large, it may amalgamate into the melt pool again under surface tension [23].As the vapour jet continues to interact with the protrusion, it superheats the liquid metal and decreases its surface tension, forming a neck region.Once the vapour pressure overcomes the surface tension of the neck region, a large liquid droplet is detached (figure 2(f)) and then ejected by the vapour plume (figure 2(g)).
The average diameter of liquid droplets ejected during laser welding ((43 ± 12) µm) is smaller than those formed during LPBF ((68 ± 21) µm).This confirms the powderdroplet/droplet-droplet coalescence events (see supplementary figure S3) are another formation mechanism of liquid droplet during LPBF, matching the observation reported in [25].The size of the liquid droplets in LPBF is comparable to the powder agglomerations (figure 2(h)) yet can be distinguished by their higher circularity (figure 2(i)).
Figure 2(j) shows no correlation between spatter size distribution and linear energy density (E l ) as the equivalent diameters of solid spatter/liquid droplet are of similar magnitudes across the E l range.However, it shows a strong positive correlation between the number of solid spatter (N solid ) or liquid droplet (N droplet ) and E l .Therefore, we have deduced the following equations to predict the number of spatters during LPBF of AA8A61.50: for solid spatter ( R 2 = 0.99 ) and for liquid droplets ( R 2 = 0.92 ) , where E l ranges from 210 J•m −1 to 840 J m −1 .Both equations can be used for future model validation and to devise new printing strategies for minimising spatter.

Quantification of solid spatter and liquid droplet motion
To understand the spatter dynamics across different scan velocities and laser powers, we have tracked and quantified the trajectory angle, fraction, and velocity of solid spatter particles and liquid droplets during LPBF, see figure 3. The respective velocity and angle distributions are detailed in supplementary figures S4 and S5.The spatter is further divided into backward and forward spatter types based on their trajectory angles relative to the laser beam position (0 • ), see inset in figure 3(a).The backward spatter has a trajectory angle spanning from −90 • to 0 • , with the rest (0 • -90 • ) considered as forward spatter.Possible quantification errors could be due to the overlapping particles and particles that move in and out of the page of the radiographs.
We also compare the kinetic energy of spatter, E k , with the input laser energy, E i , to estimate the percentage of energy consumed by the spatter generation process, , where L is the track length, and m and v m are the mass and velocity of individual spatter, respectively.The mass of spatter is given as m = 4  3 π r 3 ρ, where r is the equivalent radius of spatter, ρ is the density of spatter which is taken as 2710 and 2385 kg m −3 for solid and liquid aluminium, respectively [46].
Figure 3(a) shows a negligible effect of increasing the laser power from 350 W to 500 W on the average trajectory angle at a constant scan speed of 800 mm•s −1 , which remains from −10 • to −25 • across all laser powers.Under these conditions, most ejecta are backward spatter (70%-80%).
Figure 3(b) shows that most ejecta remain as backward spatter (ejected angle <−18 • ) at a scan speed between 1000 mm•s −1 and 2000 mm•s −1 and a constant laser power of 420 W. Given that the vapour depression zone is four times wider than the typical keyhole aperture observed in our study, the high-velocity vapour plume is more likely to interact with spatter, inducing more backward liquid droplets (>90%).The average ejection angle of solid spatter/liquid droplets reduces to ∼0 • when the scan speed decreases to 500 mm•s −1 , which is closely linked to the keyhole morphology, see explanation later.
Figure 3(c) demonstrates a strong positive correlation between laser power and spatter velocities/kinetic energy, owing to that higher laser power (or high E l ) would induce more metal vaporisation and spatter during LPBF.The spatter velocity magnitude of AA8A61.50 is comparable to other Al systems, such as Al10SiMg [20,33], however, it is 2 ∼ 3 times higher than that of Ni alloys [23].
Figure 3(d) also shows an increased trend of kinetic energy with decreased scan speed (or increased E l ).However, an  inverse particle velocity-scan speed relationship between solid spatter and liquid droplets can be observed, which is due to the increase of powder entrainment, see explanation in section 3.3.2.
Figure 3(e) shows an increase in keyhole depth with increasing laser power, but there are negligible changes in keyhole morphology and spatter angle, see figure 3(a); this remains true when the front keyhole wall angle (θ) is < 82 • .A reduction in scan speed from 2000 mm•s −1 to 1000 mm•s −1 at 420 W laser power continues to increase the surface area of the depression zone, depth, and θ, see figure 3(f).As the scan speed further reduces 500 mm•s −1 , the vapour depression zone transforms into a deep 'I'-shaped keyhole, promoting forward spatter.Based on our analysis in figures 2-3, the scan speed has a stronger influence on the spatter trajectory angle than laser power, therefore we would recommend that faster scan velocities or low E l can reduce θ and spatter during LPBF of AA8A61.50.

Spatter trajectory angles.
Here, we have selected four processing conditions to explain how the front keyhole wall angle and the keyhole morphology change the spatter trajectory angle.Figures 4(a) and S4 reveal the formation of a small, cone-shaped, near-vertical keyhole under a laser power of 350 W and a medium scan speed of 800 mm•s −1 .Figure 4(b) shows an inclined 'J'-shaped keyhole as the laser power increases to 500 W. The dominant vapour flow (orange solid arrow) is generated at the hottest region (or the front wall) of the vapour depression zone/keyhole [22].Once the initial vapour plume reaches the rear keyhole wall, it deflects upward (orange dotted arrow), see illustration in inset 1.The resultant spatter has a steeper ejection angle (red solid arrow), distinct from the direction of the initial metal vaporisation (orange solid arrow).
Figure 4(c) shows that a high scan speed (>1000 mm•s −1 ) promotes the formation of a shallow quasi-semi-circularshaped depression zone, with a front wall angle of <72 • (figure 3(f)).The vapour pressure generated from the shallow front wall of the quasi-semi-circular depression zone promotes spatter mostly in a negative spatter ejection angle in the same direction as the vapour plume (see inset 2). Figure 4(d) illustrates a more deeply penetrating 'I'-shaped keyhole is formed at the slow scan speed of 500 mm•s −1 .The vapour plume impinges on the rear keyhole wall and further redirects along it, driving the spatter upwards.
In summary, the vapour pressure is the main driving force attributed to the spatter ejection direction during LPBF which is correlated to different keyhole geometries as quantified in figures 3 and 4. Increasing the laser power from 350 W to 500 W with a scan speed of 800 mm•s −1 leads to a twofold increase in keyhole depth, but this has minimal change in keyhole morphology and spatter trajectory.However, decreasing the scan speed from 2000 mm•s −1 to 500 mm•s −1 with 420 W laser power changes the depression zone morphology from vapour depression to keyhole.The increase in both depression depth and front wall angle causes a shift from shallow-angle backward spatter to predominantly vertical spatter.Similar observations (correlation of E l , depression zone morphology and vapour trajectory) have been reported in [35], however, it mainly focused on the vapour plume behaviour and did not correlate spatter dynamics, e.g.velocities and trajectory angles, with other features such as front wall angle, keyhole geometry, or processing parameters.

Solid spatter/liquid droplet velocity. Figures 3(c) and
(d) show that the overall kinetic energy of solid spatter and liquid droplets increases with increasing E l through either increasing P or decreasing v.An increase in E l will lead to a high melt pool temperature and more metal vaporisation [47], resulting in higher recoil and vapour pressure, and increasing spatter (figure 2(e)).To confirm this hypothesis, we have quantified the recoil pressure, P r induced by the metal vaporisation process, and calculated the saturated vapour pressure, P s [48] over a melt pool surface.This is given by [47]: (5) where P s is the vapour pressure, P 0 is the atmospheric pressure (1 atm), ∆H V is the latent heat of vaporisation, which is 291 KJ•mol −1 for aluminium [47], T b is the boiling temperature of Al, which is 2773 K [47], and R is the gas constant, which is 8.
The peak temperature, T, at the melt pool surface processed by a stationary Gaussian beam can be estimated by [48]: where A is the absorptivity (0.58 [49]), σ is laser spot size of 80 µm, κ is the thermal diffusivity, taken as 89.3 W•mK −1 [50] for liquid Al, λ is thermal conductivity of molten Al, which is 3.3 × 10 −5 m 2 •s −1 [50], and I is the laser intensity calculated by: I = P/2π σ 2 .Combining equations ( 5)-( 7), the calculated recoil pressure increases from 3.1 atm to 83.2 atm as P increases from 350 W to 500 W at 800 mm•s −1 and increases from 0.6 atm to 62.1 atm with decreased v from 2000 to 500 mm•s −1 at 420 W. The velocity of recoil/vapour pressure driven spatter increases with both increased P and decreased v (or increased E l ).However, a drop in solid spatter velocity with decreasing v is evident for the solid particles, as shown in figure 3(d).The increased proportion of entrained solid spatter with a lower spatter velocity compared to the ejected solid spatter, is hypothesised to be the underlying reason for this.The inset 3 in figure 4 illustrates a drag force, F g , is exerted on the entrained solid spatter by an outward argon vortex flow and estimated by [51]: where ρ g is the Argon density, C x is the drag coefficient, S is the particle's projected area and U is the Argon flow velocity.The criterion of particle entrainment is defined as: where F G is the weight of the entrained particle and α is the contact angle between two particles.A pressure drop occurs inside the vapour plume as a result of the velocity increase when vapour flows out from the constriction at the keyhole/depression zone opening according to the Bernoulli Effect [52].This pressure difference accelerates the surrounding argon towards the vapour depression zone, which in turn causes the formation of an argon vortex flow.Powder entrainment occurs when the F g exceeds the critical value according to equation (7).Decreasing the scan speed reduces the keyhole/depression zone opening volume, and hence creates a stronger argon vortex flow under an enhanced Bernoulli Effect.Therefore, we expect a high proportion of entrained solid spatter as the F g increases, according to equation (6).The calculated outward argon flow velocity above the powder bed is up to ∼7 m•s −1 [51] and two orders of magnitude slower than the vapour plume velocity.As a result, the velocity of the entrained solid spatter is expected to be slower than that of the ejected solid spatter, similar to observations reported in [33].The increased proportion of entrained solid spatter with decreased scan speed explains the corresponding decrease in average solid spatter velocity.In summary, both the spatter velocity and the kinetic energy generally increase with higher E l (increased P or decreased v) due to the higher vapour pressure and recoil pressure.Additionally, the depression zone morphology changes from a shallow depression zone to deep keyhole as a result of decreased v, reducing the overall average solid spatter velocity by inducing increased powder entrainment.

Interaction between laser beam and solid spatter
Previous studies [15,25,26] show that large liquid droplets can be trapped between layers inducing impurities and lack-of-fusion porosity, detrimental to the properties of LPBF products.Figure 5 illustrates a new type of defect, namely a spatter-induced cavity, that is induced by the solid powder spatter interacting with the laser beam during flight.A forward spatter particle (∼36 µm) is first ejected into the laser scanning path (figure 5(a)) and then changes its spattering direction towards the powder bed after the laser-spatter interaction under a laser radiation force, F rad (figure 5(b)), which can be estimated using the following equation [53]: where q is the laser reflection coefficient, which is taken as 0.8 for Al particles with oxidation [54], P is the laser power and c is the speed of light.At a laser power of 450 W, the calculated F rad is 2.4 × 10 −6 N. The spatter changes its trajectory from upward to downward and accelerates from 0.57 m•s −1 along its initial path to 3.39 m•s −1 over 350 µs, prior to colliding with the powder bed.The critical force required by a spatter to maintain its original trajectory can be estimated by F c = ma, which is 1.0 × 10 −6 N < F rad .Upon the collision (figure 5(c)), this induces further powder spatter at the powder bed (figures 5(d) and (e)), forming a localised denudation zone ahead of the laser beam and a surface cavity in the final deposited layer after laser melting (figure 5(f)).Such spatter-induced surface cavity could act as a crack initiation site for fatigue failure [55].
We also observed multiple laser-spatter interactions during LPBF.In the first instance, a small spatter agglomerate interacts with the laser beam (supplementary figure S6) and then transforms into a liquid droplet (47 µm).Upon further laser-droplet interaction, the droplet alters its trajectory and accelerates towards the powder layer thereby causing a spatter-induced surface cavity.This phenomenon occurs when F c < F rad satisfies, in which F c is 1.4 × 10 −6 N and F rad is 2.2 × 10 −6 N. In the second instance, a larger liquid droplet (∼130 µm in size) is generated from a similar laser-powder agglomerate interaction (supplementary figure S7), the F rad (2.2 × 10 −6 N) is insufficient to change the trajectory of the large droplet as it has a much larger F c D Guo et al (9.5 × 10 −6 N), i.e.F c > F rad in this case.To minimise spatter-induced surface cavities, the end-users should minimise F rad , avoid small powder size fractions and use uniform powder size distribution.
Figure 5(g) demonstrates the correlated front wall angles of the keyhole/depression zone, forward-spatter fraction, frequency of laser-spatter interaction (F I ), and surface roughness as a function of processing parameters, divided into three distinct levels of interaction frequency: low (orange), medium (green), and high (blue).A shallow keyhole/depression zone front wall angle (such as samples S2-S4 and S6-S8 in figure 5(g)) creates minimal forward spatter.Reducing forward spatter (which may interact with the laser) should lead to decreased laser-spatter interaction events, hence reducing surface cavities and improving the surface quality (i.e.low surface roughness).Figure 5(h) further interprets the influence of keyhole/depression zone morphology on the laser-spatter interaction frequency.Quasi-semi-circular depression zones or 'J'shaped keyholes with relatively low front wall angles (∼70 • ) tend to result in a lower fraction (< 25%) of forward spatter and hence low frequency of spatter-laser interaction events.As the keyhole front wall angle becomes steeper (at an angle of 82 • ), a corresponding increase in the frequency of laser-spatter interaction was observed with the associated higher possibility of surface cavity formation and decreased surface quality (i.e.high surface roughness).Our results confirm that a high keyhole front wall angle (87 • ) is the key factor leading to a high frequency of laser-spatter interaction and spatter-induced cavities during LPBF.

Conclusions
The interrelated dynamics of spatter and keyhole formation during LPBF of an aluminium alloy, AA8A61.50(Aheadd ® CP1) were characterised using in situ synchrotron x-ray imaging.
Five distinct spatter particle types are classified based on their differing size and circularity characteristics: ejected solid spatter, entrained solid spatter, jet spatter, liquid droplets, and powder agglomerations.Additionally, we discovered two new spatter mechanisms during LPBF: (i) Rear rim droplet spatter-occurs when the metal vapour jet overcomes the surface tension of the protrusion induced by the combination of Kelvin-Helmholtz instability and the Marangoni convection at the rear rim of the keyhole.The observed droplet formation mechanism in high-speed welding is indicative of that in LPBF under the same conditions, due to the observed similarity in keyhole shape and size when the powder layer is less than 60 µm [35,56].Therefore, the effect of powder on the vapour dynamics is expected to be similarly minimal.(ii) Spatter-induced surface cavity-occurs when the laser radiation force exceeds the critical force acting on the particle during laser-spatter interaction.A high keyhole front wall angle (87 • ) leads to a high frequency of laserspatter interaction and spatter-induced cavities during LPBF.
We quantified the dynamics of solid spatter and liquid droplets during LPBF, e.g.size distribution, trajectory angle, spatter velocity and kinetic energy, and formulated two equations to predict spatter.Additionally, we correlated these measurements with the keyhole/depression zone morphologies as a function of linear energy density.
Our results reveal that the scan speed has a stronger influence on the spatter trajectory angle than laser power.We would recommend that faster scan speeds (> 1 000 mm•s −1 with laser power of 420 W), or lower E l (in an appropriate range that avoids the formation of lack-of-fusion defect), should be used to reduce the front wall angle and increase the vapour depression zone opening area during LPBF of AA8A61.50.

Figure 1 .
Figure 1.Evolution of a single melt track during LPBF of AA8A61.50 (Sample S2, P = 400 W, v = 800 mm•s −1 ).(a) Schematic of in situ x-ray imaging set up during LPBF.(b) Sequential images of melt pool and spatter dynamics with flat-field correction and background subtraction.(c) Particle (solid spatter) tracking using TrackMate plugin in ImageJ, after segmentation by applying a Gaussian filter and manual threshold.The colour of the track indicates the spatter velocity.(d) Re-framed images focusing on the melt pool region.

Figure 2 .
Figure 2. Identification of solid spatter, powder agglomeration, and liquid droplet during LPBF of AA8A61.50 (Sample S2, P 400 W, v = 800 mm•s −1 ).X-ray image during LPBF with (a) powder and (b) laser welding without powder showing 5 spatter particle types distinguished with different colours.(c) Schematic of the corresponding melt pool and spatter behaviour; where orange, yellow, green and red solid arrows refer to the moving trend of ejected powder, entrained powder, liquid droplet and jet spatter motion.The pink solid arrow refers to the moving trend of powder into droplet.The black solid arrow indicates the recoil pressure, blue dot arrow indicates the vapour flow, generating the shear force (orange dot arrow) along rear keyhole wall, and red dot arrow indicates the Marangoni flow.(d)-(g) Imaging sequence indicating the formation mechanism of large liquid droplet.(h) Distribution of diameter and (i) circularity for solid spatter, liquid droplet and agglomeration.(j) Quantification of solid spatter/liquid droplet amount and diameter as a function of linear energy density and number (N)-Linear energy density (E l ) curve fits.

D
Guo et al from melt pool is more visible during laser welding compared to LPBF under same conditions, see figures 2(b) and (d)-(g).

Figure 3 .
Figure 3. of solid spatter/liquid droplet dynamics during LPBF: (a) average trajectory angle (line graph) and backward spatter fraction (bar chart) as a function of laser power (quantified from sample S1-S4) and (b) as a function of scan speed (quantified from sample S5-S8).(c) Average spatter velocity and P E as a function of laser power and (d) as a function of scan speed.(e) Keyhole/depression zone depth and front wall angle (θ) as a function of laser power and (f) as a function of scan speed.The black and blue dotted lines depict keyhole and depression zone, respectively.

Figure 4 .
Figure 4. Schematic depicting the correlation between solid spatter/liquid droplet motion with keyhole/depression zone morphology, where (a) and (b) present the effect of increased laser power, and (c) and (d) present the effect of decreased scan speed.The red and green solid arrows indicate the motions of ejected solid and entrained solid spatter, respectively.The orange solid arrow indicates the vapour flow caused by metal vaporisation in the front depression zone/keyhole wall, and the orange dotted arrow indicates the reflection of vapour plume under keyhole mode.Insets 1 and 2 illustrate different vapour plume direction (blue solid arrow) under keyhole and conduction modes.Inset 3 illustrates particle entrainment induced by the Argon vortex flow (gray solid arrow) under drag force, Fg.

D Guo et alFigure 5 .
Figure 5. Laser-spatter interaction and its impact on the track surface.(a)-(e) Time-resolved x-ray images showing ejected spatter impacting the powder bed after interaction with the moving laser beam for sample S3 (P = 450 W, v = 800 mm•s −1 ).Red arrows indicate the velocity vectors of the spatter particle, and the blue arrows indicate those of the impacted powder particles.(f) Optical image of the surface defect induced by the laser-spatter interaction.(g) The correlated keyhole/depression zone front wall angle, forward-spatter fraction, frequency of laser-spatter interaction (F I ), and surface roughness as a function of processing parameters.(h) Schematic depicting the relationship between laser-spatter interaction frequency, ejection angle, and the morphology of the keyhole/depression zone.