Fluctuations of Tracks and Layers during Aluminium Laser Powder-Bed Fusion

: Laser Powder-Bed Fusion (LPBF) is one method in Additive Manufacturing where layer-wise complex structures can be built. However, although the LPBF machines produce promising parts, the efﬁciency and process speed are typically still low, which can make the process expensive and uneconomical. Recent research showed that volume elements in the parts can be melted several times, while only a small material volume is added, which indicates a high loss of energy. In order to understand the process better, in this work, theoretical modeling and smart powder-bed experiments were designed to explain the impact on the track dimensions based on the previously built tracks and layers. It was found that the powder availability varies for each track and has an alternating character within and between layers. The comparison of the simulation and experimental results indicates that the powder pick-up from neighboring powder volumes is the main reason for the variations of the powder availability.


Introduction
Laser Powder Additive Manufacturing contains two main process variants. Processing blown powder is named Direct Energy Deposition, Direct Metal Deposition or Laser Metal Deposition, while building structures with pre-placed powder is named Selective Laser Melting, Laser Metal Fusion or Laser Powder-Bed Fusion (LPBF). The LPBF process uses laser beams that locally melt the pre-placed powder of the powder bed. Layer by layer, complex shaped parts are possible to produce. The LPBF process is industrially used for applications where usually high-cost material and complex structures are necessary to be produced. The LPBF process enables the production of high-quality parts using several powder materials, including Ni [1,2], Fe- [3] and Ti-alloys [4]. In addition, certain aluminium alloys are more widely used in LPBF, showing advantages in part performances due to lightweight design possibilities [5]. For the typically used AlSi10Mg alloy, studies showed that single and overlapping tracks can be produced using different scan speeds, showing a homogeneous distribution of alloying elements and no porosity, while, however, a higher layer thickness is not recommended to use due to the increased risk of defect formation [6]. Almost fully dense structures could be created with the alloys AlMg3, AlMgSi0.5, AlSi12 and AlSi10Mg [7]. Due to the higher thermal conductivity of aluminium, higher laser power and/or slower processing speeds are required compared to other metals like many steel grades [8].
Defects can occur that can influence the resulting part properties [9]. Therefore, a minimum of defects is usually required [10]. Typical defects during LPBF are, e.g., solidification cracks [11] or pores due to incomplete densification [12,13]. Further possible origins of the pores and voids were identified to be the space in the bulk powder and trapped gases [14]. The part density is defined as the percentage of fully molten and connected material without interruptions of defects, while typically a density of 100% is desired. The processing and partially overlapped laser spots in a sealed chamber under Argon atmosphere, which is the typical industrial setup. No preheating was applied to the build plate. Multiple layers were produced by keeping the scanning direction (90 • ) constant after each layer completion. Although conventional LPBF machines usually scan different layers in different directions to reduce material anisotropy, for this investigation the tracks and layers were built in one direction in order to investigate the full cross section of the single tracks. The commonly used AlSi10Mg powder with particle size between 20 and 63 µm was produced by IMR Metal Powder Technologies GmbH ( Table 1). Optimisation of LPBF parameters was performed in order to evaluate the effect of scanning speed and hatch distance on the material density. Table 2 reports the set of process parameters that led to the highest density with hatch distance of 100 µm (99.57% ± 0.06). Not all results of process parameters optimization are reported in this work. The process parameters used for the experiments are shown in Table 2.
During the experiments, the hatch distance h was varied between 80 µm (A), 100 µm (B) and 120 µm (C). In order to be able to investigate the tracks and layers before they get remelted by subsequent laser scans, single track, multiple tracks and multiple layers were progressively built onto the top of an additively manufactured AlSi10Mg substrate, as shown in the schematics of Figure 1.

Experimental Setup
In order to gain insight into the track occurrence during LPBF, AlSi10Mg single/multiple tracks and layers were built in a Renishaw AM250 system, equipped with the Reduced Built Volume device. The printer operates with a pulsed 200 W single-mode fiber laser with an estimated beam diameter at focal point of 75 μm. Laser melting was performed by discrete and partially overlapped laser spots in a sealed chamber under Argon atmosphere, which is the typical industrial setup. No preheating was applied to the build plate. Multiple layers were produced by keeping the scanning direction (90°) constant after each layer completion. Although conventional LPBF machines usually scan different layers in different directions to reduce material anisotropy, for this investigation the tracks and layers were built in one direction in order to investigate the full cross section of the single tracks. The commonly used AlSi10Mg powder with particle size between 20 and 63 μm was produced by IMR Metal Powder Technologies GmbH ( Table 1). Optimisation of LPBF parameters was performed in order to evaluate the effect of scanning speed and hatch distance on the material density. Table 2 reports the set of process parameters that led to the highest density with hatch distance of 100 μm (99.57% ± 0.06). Not all results of process parameters optimization are reported in this work. The process parameters used for the experiments are shown in Table 2. During the experiments, the hatch distance h was varied between 80 μm (A), 100 μm (B) and 120 μm (C). In order to be able to investigate the tracks and layers before they get remelted by subsequent laser scans, single track, multiple tracks and multiple layers were progressively built onto the top of an additively manufactured AlSi10Mg substrate, as shown in the schematics of Figure 1.

Microstructure Analysis
Samples were cut for transversal cross-section analysis. Microstructural observations were conducted with a Nikon Eclipse LV150NL light optical microscope and with Zeiss Sigma 500 field emission scanning electron microscope (FE-SEM), equipped with electron backscattered diffraction detector (EBSD). Samples were mounted, grinded, polished and etched with Keller's reagent to evaluate the geometrical and microstructural features of the built structures. EBSD analyses were conducted to investigate grain size and orientation. Specimens were prepared for EBSD analysis with emery paper up to 2500 grit and diamond pastes. A final polishing step was performed by silica in a Struers Tegramin-25 machine. Results of microstructure analysis were compared for the different hatch distances.

Simulation
In order to simulate the tracks and layer appearances, a model was established in MATLAB (Version R2017a). The model is based on previous work of the authors [19,20]. The model calculates the molten cross-sectional area based on a moving Gaussian heat source, while the resulting temperature field was used to find the melting line and describe the lower part of the melt pool shape. Material parameters, such as the absorptivity, thermal diffusivity, melting temperature and the latent heat of melting were used as temperatureindependent variables [19]. In the model, it is assumed that the laser beam illuminates the theoretical available powder area ( Figure 2a) and integrates 30% of the neighbor areas in addition [25]. For the following tracks (Figure 2b), varying powder amounts are available to build the tracks. This leads to different track occurrences. The track shapes were calculated based on the heat input assuming an absorption of 15% on the aluminium flat melt pool surface (based on [26]). The upper parts of the tracks were simulated as circle sections. Due to the powder densification during processing, a shrinkage of 60% was assumed [27]. A higher powder availability leads therefore to higher tracks.

Microstructure Analysis
Samples were cut for transversal cross-section analysis. Microstructural observations were conducted with a Nikon Eclipse LV150NL light optical microscope and with Zeiss Sigma 500 field emission scanning electron microscope (FE-SEM), equipped with electron backscattered diffraction detector (EBSD). Samples were mounted, grinded, polished and etched with Keller's reagent to evaluate the geometrical and microstructural features of the built structures. EBSD analyses were conducted to investigate grain size and orientation. Specimens were prepared for EBSD analysis with emery paper up to 2500 grit and diamond pastes. A final polishing step was performed by silica in a Struers Tegramin-25 machine. Results of microstructure analysis were compared for the different hatch distances.

Simulation
In order to simulate the tracks and layer appearances, a model was established in MATLAB (Version R2017a). The model is based on previous work of the authors [19,20]. The model calculates the molten cross-sectional area based on a moving Gaussian heat source, while the resulting temperature field was used to find the melting line and describe the lower part of the melt pool shape. Material parameters, such as the absorptivity, thermal diffusivity, melting temperature and the latent heat of melting were used as temperature-independent variables [19]. In the model, it is assumed that the laser beam illuminates the theoretical available powder area ( Figure 2a) and integrates 30% of the neighbor areas in addition [25]. For the following tracks (Figure 2b), varying powder amounts are available to build the tracks. This leads to different track occurrences. The track shapes were calculated based on the heat input assuming an absorption of 15% on the aluminium flat melt pool surface (based on [26]). The upper parts of the tracks were simulated as circle sections. Due to the powder densification during processing, a shrinkage of 60% was assumed [27]. A higher powder availability leads therefore to higher tracks. In addition, several layers were modeled considering the actual powder availability. For the first layer, the powder layer height is defined by the user (Figure 2c). When applying, e.g., the second powder layer, the volume above the previous first layer is larger Appl. Sci. 2021, 11, 3132 5 of 15 than theoretically assumed due to the shrinkage of the material. Calculations of the track position and heights were conducted according to the actual powder availability for each track in 5 layers.
For validation of the model, the resulting track dimensions were evaluated and compared to the experimentally derived track geometry of a single track. Figure 3 shows the simulated tracks. Table 3 shows the comparison between the simulated and experimentally derived values. In addition, several layers were modeled considering the actual powder availability. For the first layer, the powder layer height is defined by the user (Figure 2c). When applying, e.g., the second powder layer, the volume above the previous first layer is larger than theoretically assumed due to the shrinkage of the material. Calculations of the track position and heights were conducted according to the actual powder availability for each track in 5 layers.
For validation of the model, the resulting track dimensions were evaluated and compared to the experimentally derived track geometry of a single track. Figure 3 shows the simulated tracks. Table 3 shows the comparison between the simulated and experimentally derived values.  The depth of the track is slightly underestimated due to neglecting effects of the structure being heated from previously built tracks that occur in the experiments that lead to a processing at higher temperatures of the previously built structures. However, the evaluation in the current work is focused on relative fluctuations of the tracks, which are possible to derive from the simulation results. Modelled track geometries and fluctuations were compared to the experimental observations in order to conclude about the origin and mechanisms of the observed track fluctuations.

Simulation Results
For the different investigated hatch distances, the simulated tracks are given in Figure 4, while all parameters show no risk of voids due to the sufficient overlap of tracks and layers, as also observed in other works, where a hatch distance >70% of the laser beam diameter was necessary to produce voids [7].  The depth of the track is slightly underestimated due to neglecting effects of the structure being heated from previously built tracks that occur in the experiments that lead to a processing at higher temperatures of the previously built structures. However, the evaluation in the current work is focused on relative fluctuations of the tracks, which are possible to derive from the simulation results. Modelled track geometries and fluctuations were compared to the experimental observations in order to conclude about the origin and mechanisms of the observed track fluctuations.

Simulation Results
For the different investigated hatch distances, the simulated tracks are given in Figure 4, while all parameters show no risk of voids due to the sufficient overlap of tracks and layers, as also observed in other works, where a hatch distance >70% of the laser beam diameter was necessary to produce voids [7].
Based on the simulated available powder for each track, the maximum heights of the single tracks were derived and plotted for each layer. The heights of the tracks in each layer show fluctuations, while the characteristics in each layer vary ( Figure 5).   In order to derive how much space there is above the already built structure to the height where the new powder layer is applied, this calculated space was plotted in Figure  6. Although there is a powder layer height programmed of 40 μm, the actual applied pow-   In order to derive how much space there is above the already built structure to the height where the new powder layer is applied, this calculated space was plotted in Figure  6. Although there is a powder layer height programmed of 40 μm, the actual applied powder layers have around 45 μm to 60 μm powder layer thicknesses depending on the actual In order to derive how much space there is above the already built structure to the height where the new powder layer is applied, this calculated space was plotted in Figure 6. Although there is a powder layer height programmed of 40 µm, the actual applied powder layers have around 45 µm to 60 µm powder layer thicknesses depending on the actual  The simulations enabled to calculate the accumulated material available for each built structure (Figure 7), which is the sum of all available powder material for all tracks built for each application strategies. For all application strategies, the available amount of powder increases at larger hatch distances.   (Figure 8b); it is likely that their growth was hindered by grains with more favorable orientation to the thermal gradients. Figure 8c shows a <001> texture both along the scanning direction and building direction (z-direction and y-direction in the EBSD reference system, respectively). Indeed, epitaxial and The simulations enabled to calculate the accumulated material available for each built structure (Figure 7), which is the sum of all available powder material for all tracks built for each application strategies. For all application strategies, the available amount of powder increases at larger hatch distances. track height. The increased space can be related to reduced amounts of available p and densification effects. The simulations enabled to calculate the accumulated material available f built structure (Figure 7), which is the sum of all available powder material for al built for each application strategies. For all application strategies, the available am powder increases at larger hatch distances.   (Figure 8b); it is likely that their grow hindered by grains with more favorable orientation to the thermal gradients. Fi shows a <001> texture both along the scanning direction and building direction (z tion and y-direction in the EBSD reference system, respectively). Indeed, epitax   Figure 8b,c, respectively. Grains slightly elongated towards the center of the melt pool are visible. Fine grains are also noticeable at the boundary of the solidified melt pools (Figure 8b); it is likely that their growth was hindered by grains with more favorable orientation to the thermal gradients. Figure 8c shows a <001> texture both along the scanning direction and building direction (z-direction and y-direction in the EBSD reference system, respectively). Indeed, epitaxial and competitive growth of grains occurs during the solidification of the melt pool, leading to the formation of a strong <001> texture along the maximum heat flow directions.

Cross-Section Evaluation
ppl. Sci. 2021, 11, x FOR PEER REVIEW competitive growth of grains occurs during the solidification of the melt p the formation of a strong <001> texture along the maximum heat flow direc  Example track occurrence at different hatch distances are shown in dotted red line surrounds the added material by the tracks.     Example track occurrence at different hatch distances are shown in Figure 10. The dotted red line surrounds the added material by the tracks.
The area of these zones was measured for all experiments. Figure 11 shows the evaluation of used material (as cross-sectional area) for the different applied structures in the cross sections compared to the simulated material additions. In general, the simulated results are in good agreement with the measured values. In particular, for hatch distances of 100 µm and 120 µm, the simulation shows the tendency to predict larger material addition when applying more layers. Example track occurrence at different hatch distances are shown in Figure 10. The dotted red line surrounds the added material by the tracks. The area of these zones was measured for all experiments. Figure 11 shows the evaluation of used material (as cross-sectional area) for the different applied structures in the cross sections compared to the simulated material additions. In general, the simulated results are in good agreement with the measured values. In particular, for hatch distances Appl. Sci. 2021, 11, x FOR PEER REVIEW 9 of 14 of 100 μm and 120 μm, the simulation shows the tendency to predict larger material addition when applying more layers. An exemplary comparison of simulated and experimentally derived tracks shows that the model gives a surface bulging that represents the additionally applied material, while the experimentally created tracks partly show no rounded caps ( Figure 12). When applying more layers, round surfaces occur. The bulging occurs with a low structure going into a maximum and a minimum in the center and back to a maximum leading into a low structure at the end ( Figure 13). An exemplary comparison of simulated and experimentally derived tracks shows that the model gives a surface bulging that represents the additionally applied material, while the experimentally created tracks partly show no rounded caps ( Figure 12). of 100 μm and 120 μm, the simulation shows the tendenc dition when applying more layers. An exemplary comparison of simulated and experi that the model gives a surface bulging that represents the while the experimentally created tracks partly show no ro When applying more layers, round surfaces occur. structure going into a maximum and a minimum in the c leading into a low structure at the end ( Figure 13). When applying more layers, round surfaces occur. The bulging occurs with a low structure going into a maximum and a minimum in the center and back to a maximum leading into a low structure at the end ( Figure 13). When applying more layers, round surfaces occur. The bulging occurs with a low structure going into a maximum and a minimum in the center and back to a maximum leading into a low structure at the end ( Figure 13).

Discussion
A comparison of experimental and simulated results of different tracks and layer reveals new insights into LPBF processes. The simulation and experimental results show in a wide range comparable tendencies. Therefore, the simple model is considered feasibl to explain certain effects occurring during powder-bed processing.

Discussion
A comparison of experimental and simulated results of different tracks and layers reveals new insights into LPBF processes. The simulation and experimental results show in a wide range comparable tendencies. Therefore, the simple model is considered feasible to explain certain effects occurring during powder-bed processing.

Fluctuations within One Layer
The cross-sectional views of the built structures ( Figure 9) confirmed that the single tracks show large depths compared to the added material, which shows that the energy input is large compared to the desired volume to melt, which is a typical trend for LPBF processes to use conservative process parameters to avoid defects [27]. Similar trends were seen in [18], where volume elements were seen to be melted several times, which requires a lot of energy but creates no additional value, since material is not added.
It was observed that fluctuations occur in track appearance within one layer, as was also documented in [20] or [28], where variations of the track depths were observed. The simulation results showed that variations of available powder occur due to the varying powder available for the single tracks. An alternating characteristic was found ( Figure 5) in the observed modelled tracks within one layer. Since the model is based on assumptions about powder availability and powder pick-up from neighboring sections, the varying availability of powder can explain the observed variations. This confirms that the powder availability model predicts the material availability well.
At higher layers, the simulation overestimates the material addition compared to the experimentally measured ones for all hatch distances (Figure 11), while when building only a few tracks, the simulation prediction and experiments are similar or even the experimental material usage is slightly higher. The deviations are possibly related to effects that were not considered in the simple model: there is, e.g., a potential powder particle loss during the processing by induced gas flows [29] from the vapor plume [22]. Additionally, spattering can happen, which is induced by the melt pool dynamics [30] or resonant melt pool waves [31], denoting material losses. High-speed images could show that non-melted particles and agglomerations of powder particles are ejected from the processing zone [20]. This effect, in combination with the powder pick-up from neighboring sections, leads to the denudation zone and reduces the amount of available powder for the subsequent tracks.
The amplitudes of the fluctuations vary within one layer. When less space above the previous track was available, the layer became smoother. This observation is another hint that the denudation effect is different for each track. The first layer shows quite different boundary conditions compared to the subsequent layers. In the first layer, the first track sees a comparably high amount of powder, since no denudation zone was created before the processing ( Figure 5). However, the track heights within the layer show large fluctuations in the simulations ( Figure 5) and experiments ( Figure 13) compared to higher layers. The denudation effects seem to have a larger impact on the powder availability and lead to more variations. In higher layers, the track height variations within one layer were reduced ( Figure 5). The environment consisting of powder seems to support the powder to remain closer to the processing zone. Also, the observation that, at higher hatch distances, the powder availability is in general higher (Figure 7), indicates that the effect of particles that are blown away by induced gas flows is the minor effect compared to the powder pick-up from neighboring sections.

Fluctuations between Layers
In theory, the machine movement providing a constant stepwise movement of a defined distance before applying the next powder layer provides always the same powder layer thickness. However, in reality, the powder layer thicknesses vary in different layers. Only the first layer sees the programmed value, while due to denudation effects and densification, the subsequent powder layer thicknesses show larger heights ( Figure 6). This might be a reason why LPBF machines use quite conservative parameters to ensure the melting of the tracks to the structure even when higher amounts of powder must be processed [27]. The knowledge about the process behavior can help to reduce the energy input where, e.g., less powder is available and the energy input would just mainly remelt the previous structure.
The observed layer fluctuations ( Figure 5) show higher track additions in the uneven layers compared to the even-numbered layers, which was seen in the simulation and experimental results. Possible reasons can be the varying powder amount above the previous layer or surface tension effects that locally deform the structure compared to the expected structure.
A comparison between simulated and experimentally measured structure heights in different layers ( Figure 14) shows that the model predictions of the layer material additions are quite close to the experimental values. At large hatch distances (h = 100 µm and h = 120 µm), the simulation predicts slightly larger structure heights, which can be explained by the previous discussion about additional effects that occur in reality and were not considered in the model. Considering those effects would make the model more precise. However, the quantitative values of some of these effects are not yet known, e.g., the amount of material loss due to spattering. An implementation of those effects requires additional assumptions, which might reduce the possibility of finding possible explanations for the observed effects when the single impacts of too many parameters are present.
The large variations of the structure height are reflected as large error bars in Figure 14. When building higher layers, the surface tension effect leads to an accumulation of material around the edges after the beginning and before the end of the layers. This leads to larger heights and a minimum in the center. However, also the very first track (Figure 15a) sees a large amount of powder, which is available for incorporation into the structure, which increases the already extensively grown structure size. Therefore, this geometrical effect can alter the powder availability for the tracks and layers in addition to the denudation effects discussed before. Figure 15b shows that for the second (even) numbered tracks, less material is available and mainly the previous structure is remelted, which can support the formation of the surface tension-driven round structure, while only a minimum of material is available. Less powder will be available at the high-structure regions and more powder will be available in the structure center. There, the risk of incomplete melting can occur since the laser energy might not be sufficient anymore to melt the available powder to the structure. Interestingly, however, the lower surface in the center of the structure indicates that there was not much powder material available for building those tracks. In Figure 15c, slightly more powder is available compared to Figure 15b. However, the denudation effects reduce the amount of powder already before processing. Therefore, the powder availability is limited and the structure will remain low in the center region even though the material was available in the beginning of the layer processing. An additional problem to consider is that the laser beam is not focused as expected when hitting the actual powder surface.
Typically, the laser beam focus is set on the top of the powder layer to guarantee high intensities, while the intensity is reduced in defocused position at larger laser spot sizes, which can alter the process. h = 120 μm), the simulation predicts slightly larger structure he plained by the previous discussion about additional effects that o not considered in the model. Considering those effects would ma cise. However, the quantitative values of some of these effects are amount of material loss due to spattering. An implementation o additional assumptions, which might reduce the possibility of fin tions for the observed effects when the single impacts of too many The large variations of the structure height are reflected as la 14. When building higher layers, the surface tension effect leads material around the edges after the beginning and before the end to larger heights and a minimum in the center. However, also the 15a) sees a large amount of powder, which is available for incorpo which increases the already extensively grown structure size. The effect can alter the powder availability for the tracks and layers i dation effects discussed before. Figure 15b shows that for the se tracks, less material is available and mainly the previous structure support the formation of the surface tension-driven round struct mum of material is available. Less powder will be available at the It is indicated that, besides denudation effects, induced gas flows and geometrical characteristics of the built structure can alter the powder availability of tracks, which leads to variations of track and layer dimensions. the powder availability is limited and the structure will remain low in the even though the material was available in the beginning of the layer process tional problem to consider is that the laser beam is not focused as expected the actual powder surface. Typically, the laser beam focus is set on the top o layer to guarantee high intensities, while the intensity is reduced in defocus larger laser spot sizes, which can alter the process. It is indicated that, besides denudation effects, induced gas flows and characteristics of the built structure can alter the powder availability of tracks to variations of track and layer dimensions.

Conclusions
The developed simple analytical model predicts the tracks and layers A comparison between simulation and experimental results led to the follo sions: • Applied powder layer heights vary for each track and each layer. Abo powder layer thicknesses larger than the programmed machine increm expected due to shrinking effects.

Conclusions
The developed simple analytical model predicts the tracks and layers in LPBF well. A comparison between simulation and experimental results led to the following conclusions: • Applied powder layer heights vary for each track and each layer. Above the layers, powder layer thicknesses larger than the programmed machine increment must be expected due to shrinking effects.

•
The powder availability varies from each track and between layers. A larger hatch distance can increase the powder usage efficiency due to the reduced powder pick-up from the surrounding and powder volumes denoted for subsequent tracks. The powder availability between layers fluctuates as seen in the simulations, which was shown to lead to similar tendencies of different distances between the layers in the experiments.

•
The first track in one layer always (except for the first layer) has the most powder available since no powder was taken from other tracks. Afterwards, gas flows during processing and powder pick-up led to the reduction of available powder and denudation.
• Additional effects can alter the powder availability. It was seen that the effect of surface tension forms rounded structures in higher layers.