Printing parameters for titanium powder in a Truprint 1000 to improve density and surface finish

3D printer machine settings need to be adjusted for each material. End-users (companies and individuals) consider it too risky and time-consuming to experiment with changes of their hardware and/or powder feeds. However, those end-users also recognise the untapped opportunities that different materials and processes may bring to their product innovation cycle. Little guidance on machine-material parameter selection is available. The International AM Community has been working on lowering entry barriers for non-experts, democratising use and accelerating adoption of AM technologies for some time. Parameterisation of printing settings to optimise for density, surface finish and materials performance is crucial in those efforts. Open decentralised platforms and databases such as SENVOL (2024) and AddiMap (2024) are making a significant contribution to this endeavour as well as enabling the realisation of Industry 4.0. Real-time application and monitoring technologies require analytics that deal with design parameters and the prediction of properties, and software that generates material properties (Choppara et al., 2023). But these digital tools require validation to provide robust models and realistic predictions. Experimentation via case studies, benchmark artifacts and demonstrator designs (Gruber et al., 2020) continues to be not only relevant, but much needed to capture the intrinsic aspects of any AM technology, beyond geometry and cost and to include functional properties stemming directly from printing parameters from the hardware and properties arising from the processing of raw materials.

Selective Laser Melting (SLM) is a type of Laser Powder Bed Fusion (LPBF) technology able to create metal complex parts with intricate designs, esp. lattices with customised topologies. However, SLM components can exhibit manufacturing-related defects, including the presence of pores and keyholes. These pores are typically caused by meltpool thermokinetics, i.e. insufficient thermal energy being supplied to the powderbed during melting, leaving un-melted powder within the component (Salem et al., 2019; Montalbano et al., 2021). Some studies have shown the presence of pores due to excess thermal energy being supplied to the powderbed. This can cause meltpool turbulence, leading to inclusions of the inert gas in the powderbed within the component (Salem et al., 2019). The presence of voids is known to affect mechanical properties negatively, leading to a reduction of elastic modulus, yield strength and ultimate tensile strength (Joshi et al., 2023; Li et al., 2023; Pal et al., 2020). In addition to meltpool dynamics, the fabrication of lattices via SLM is also affected by another mechanism of void formation, i.e. hatch porosity. This is observed when the infill hatching and border contour do not have sufficient overlap (Yin et al., 2023). Parts with greater surface area per volume, and therefore a larger border contour per layer, are more prone to this defect, rendering lattice structures particularly vulnerable.

From the suite of LPBF techniques, SLM creates parts with better surface finish compared to Electron Beam Melting or Selective Laser Sintering, but secondary roughness (Greitemeier et al., 2016), caused by powder particles partially sintered onto the component walls, is still present. Secondary roughness differs from material roughness in that the latter refers to the uneven surface created by the solidified metal that underwent complete melting by laser exposure, while the former refers to the coarseness created by the partly molten particles adhered to the printed walls or struts. Secondary roughness can cause component weakening due to the presence of stress raisers and crack initiation sites (Koutiri et al., 2018; Walker et al., 2017) and can also cause significant geometric inaccuracy due to the addition of raw powder particles extending beyond the design intent of the component, particularly seen within small cavities (Ma et al., 2020). With a medical device application in mind, the risk of the powder particles detaching from the component once implanted is intolerable (Wang et al., 2020). It is possible to diminish or remove secondary roughness by post-processing methods including shot-peening, electropolishing or chemical etching (Bai et al., 2021; Mohammadian et al., 2018; Torres-Sanchez et al., 2023; Wysocki et al., 2019; Yang et al., 2019). However, these methods can be hazardous requiring strong chemicals, expensive capital investment and are heavily compromised for 3D lattices, where line-of-sight to affect internal surfaces, required for e.g. sand-blasting and electropolishing, does not exist. As a premise, it is desirable to minimise secondary roughness formation (i.e., partly attached particles present) during the manufacturing process itself, but this cannot be at the expense of essential properties such as density.

Efforts to optimise process parameters to improve surface quality and minimise defects (Oliveira et al., 2020) are leveraging statistical and machine learning tools, although this space is still in its infancy and it is only the most common commercial alloy types [i.e., stainless steel (Qin et al., 2024; Tran and Lo, 2019) and Ti64 (Park et al., 2022)] that are under most frequent study. Validation is necessary and this requires experimental data. Several studies (Table 1) have established optimum parameter sets for SLM processing of commercially pure Titanium (cpTi) or its alloys on various machines. A range of properties, including density, microstructure and mechanical performance, typically reported against volumetric energy density (VED), were tabulated. Despite the recognised detrimental impact of secondary roughness, this is typically not considered, or not reported, when optimising the process parameters.

This study presents experimental data from an optimisation exercise that uses grade 1 cpTi powder as feedstock and a Trumpf Truprint 1000 as the hardware. No such study on that powder/hardware pair has been reported to date. The aim is to identify SLM printing parameters to maximise material density and minimise secondary roughness of printed parts, with a verification case study for the realisation of lattices with intricate complex internal architectures. Because no theoretical model that allows the prediction between printing parameters, VED and relative density or roughness, empirical studies are necessary. This work contributes to the AM community’s endeavour to populate a framework that includes the material-machine pairing to enable digital tools and create robust predictive models.

Cylindrical coupons (6 mm diameter, 12 mm height) were created, positioned and sliced with Materialise Magics v22.01 (Materialise, Belgium) to be printed on their axial direction (OZ), and placed directly onto the buildplate. A Trumpf Truprint 1000 (Trumpf, Germany) equipped with a 55 µm laser beam spot size was used with grade 1 cpTi powder of 15–45 µm size (D10 = 20 µm, D50 = 34 µm, D90 = 45 µm, <15 µm 3% vol.) produced by gas atomisation (AP&C, Canada). The chemical composition (measured as per ASTM E1941 (C), 1409 (O, N), E1447 (H), E2371 (Fe and others), Table 2), flow rate (ASTM B213, 28 s/50g) and apparent density (ASTM B212, 2.59 g/cm³) characterised the powder.

The design of experiments included a systematic study across the main printing parameters (laser power, speed and layer height, keeping hatch distance constant) to achieve different values of VED (Figure 1 and Table 3). VED is defined as the Joules of energy imparted onto the powder by the laser (J/mm³), and calculated using equation (1), where P is the laser power (Watts, W), v. is the laser speed (mm/s), H is the hatch distance (mm) and L is the layer height (mm):

The values used are reported in Table 3 and the rationale of the experimental plan is as follows:

Experiment 1:VED broad search. Over a range of 30–230 J/mm³, VED was achieved by varying speed between 184 and 1414 mm/s whilst holding the laser power at 140 W, the layer height at 30 µm and the hatch distance at 110 µm. This allowed a general trend between coupon density and VED to be established.

Experiment 2:VED narrow search. Informed by the results from experiment 1, a narrower VED range of 50–70 J/mm³ was explored. This was achieved by changing the laser power between 126 and 174 W, reducing the layer height to 20 µm and holding the laser speed at 1136 mm/s and the hatch distance at 110 µm.

Experiment 3: Fixed VED with varying speed and laser power. Informed by experiments 1 and 2, the target VED (55 J/mm³) was held whilst laser speed and power were changed within ranges 686–1383 mm/s and 83–168 W, respectively. The layer height (20 µm) and the hatch distance (110 µm) were kept constant. Based on these results, parameter set 4 (laser power 108 W and speed 889 mm/s) was adopted for the subsequent experiments.

Experiment 4: Scan strategy. This experiment was designed to optimise print paths for the best performing parameter combination from experiment 3. Six strategies were considered (Figure 2). The default scan strategy (used in experiments 1–3, Strategy 1 and therefore considered as the reference) consisted of one border and a hatched infill, with the infill rotated 90° each subsequent layer and a 55 µm offset between the border and the infill. Strategy 2 used those same settings but with a border-infill offset of 0 µm resulting in overlapping of infill and border scan paths at the boundary surface. Strategies 3 and 4 were built with 5 borders, while the rest of the settings remained identical to the reference strategy (Strategy 1), but Strategy 3 started at the inner border and worked outwards, whilst Strategy 4 started at the outer border and worked inwards. Strategy 5 was built with a conformal infill style, achieved by projecting the border contour inwards as many times as required to fill within the perimeter walls. Strategy 6 was also built with a conformal infill style, but also included a “double pass”, where the laser passes over the powder substrate twice. This was achieved by halving the hatch distance and overlapping the melt pool by 50% with each pass. Strategies 5 and 6 were manufactured starting from the inside towards the periphery.

Lattice trabecular structures were chosen for the validation study. The trabecular structure was designed in nTop v.4.6.2 (nTopology Inc. USA). Seed points were placed randomly in a 3D design space of 10×10×11 mm³ cuboid at a point spacing distance of 1.75 mm. The Voronoi Tessellation method (Rokicki and Gawell, 2016) was used to divide the design space into cells corresponding to each seed point. A thickness value of 0.4 mm was applied to the boundaries of these cells resulting in a scaffold with cylindrical struts. The values of point spacing and thickness were selected to achieve an intended porosity of 76%.

Porosity $(φ_vol)$ of the lattice designs was obtained from the CAD models and applying equation:

where V_s and V_b are the scaffold actual and bulk volume values respectively in cm³.

Two lattice structures were manufactured:

1. using the optimum settings stemming from experiment 4, strategy 6; and

2. using a sub-optimal parameter setting (experiment 1, set 1 using strategy 1), for comparative purposes.

Once manufactured, coupons and lattices were removed from their build plates with a high-speed saw (Buehler, USA) and washed in an ultrasonic bath (Grant Instruments, UK) with distilled water for 30 min to ensure the removal of any loose powder. They underwent no further treatment.

Density was determined using Archimedes’ method (based on ASTM D792-20), whereby the sample is weighed dry (⁠$mdry$⁠), then weighed when submerged in acetone (⁠$msub)$⁠. Acetone was chosen due to its very low surface tension enabling it to penetrate the sample fully, critical when measuring porous structures. The bulk material density (⁠$ρmaterial$⁠) was calculated using equation (4) entering acetone density value at the corresponding temperature (⁠$ρacetone)$⁠:

The relative density (⁠$ρr$⁠) of the coupons is reported here as %. This was calculated by dividing $ρmaterial$ by the density of pure solid Titanium (4.506 g/cm³, Royal Society of Chemistry Periodic Table). The porosity $(φ)$ of the manufactured lattices is defined as the volume fraction of air present in the bulk volume and was calculating using equation (5):

The secondary roughness of the coupons was measured using a 4 K digital microscope (VHX-7000, Keyence, Japan) at x80 magnification. Areal surface roughness parameters Sa (arithmetical mean peak height) and Sz (maximum peak height) (ISO 25178–2:2021) were extracted by measuring a 3 mm × 2.5 mm region of interest (ROI) [Figure 4(b)] on the coupon’s sidewall [Figure 4(a)]. An appropriate shape correction value was applied to the topographies to flatten the curved sidewall of coupons for roughness measurement [Figures 4(c) and (d)]. No additional filtering was performed on the surface image. The average value of three samples for each parameter set were used for analysis.

The visual inspection of the solid coupons was undertaken with the same microscope at x100 magnification. The thickness (t) of the secondary roughness layer was measured as the thickness of the ring created by subtracting the radius of the solid material (i.e. core, post-application of the known shrinkage values of SLM processed cpTi) from the measured total sample radius.

The manufactured lattices were scanned using a v|tome|x M (Waygate Technologies, US) micro computed tomography (µCT) system. An X-ray voltage of 180 kV, 50 µA current and scan resolution of 10 µm in XYZ were used with an 0.5 mm Cu filter. The scan was reconstructed using Datos|x software from the CT system manufacturer with correction for any minor movements using the “scan optimizer” tool. A beam hardening correction value of 8 was applied before exporting the data as a 3D volume in 16-bit data format. After reconstruction, the 3D volumes were imported to ORS-Dragonfly v.2022.2 (Comet Technologies Canada Inc.) for morphological characterization. The images were segmented into solid-void phases. The solid phase represented the material and void phase represented the connected and closed pores in the scaffolds. The volume of the labelled voxels in solid phase was obtained in cm³ and equation (2) was used to calculate lattice open porosity. The labelled voxels of closed pores were then isolated from the void phase and their volume values were used to calculate closed porosity.

Coupons were manufactured in triplicate for each printing parameter set. Values are reported as averages and with standard deviation (SD), depicted in the plots as error bars. Statistical analyses were performed using Prism 10 (GraphPad LLC.). Pearson’s correlation was used to identify positive (+1), negative (−1) or null correlations (0) in pairwise comparisons. Significant differences were detected using a two-tailed paired parametric t-test. Normality (Gaussian) of residuals was tested using Shapiro-Wilk. Statistically significant values were considered when p-value was < 0.05, and t-stat was compared to t-critical to test the nullity of this hypothesis.

Secondary roughness is inevitable and inherent to LPBF technologies. Efforts to modify the thermo-kinetics of the meltpool can mitigate this phenomenon, but it is usually at the expense of the material density. Therefore, there is a compromise between secondary roughness and relative density, and the Trumpf TruPrint 1000 machine is not immune to this. High relative density relies on ample thermal energy being supplied to the powder bed by the travelling laser beam, but by increasing the supplied thermal energy, secondary roughness probability is increased. The results of VED optimisation by modifying laser speed (experiment 1) or laser power (experiment 2) are presented in Figures 3(a)–(b). A VED range 30–230 J/mm3 was explored, delimited by the lowest VED used in our lab that produces parts of sufficient integrity, and the VED ceiling that the machine can deliver. The relative density of the coupons correlates positively with VED until 90 J/mm³ and then plateaus. Concurrently, the secondary roughness values (thickness t, Sa and Sz) also rise as the VED increases. This phenomenon has been observed in other studies (Greitemeier et al., 2016; Montalbano et al., 2021; Salem et al., 2019). The density for VEDs below 50 J/mm³ is too low to be viable and the secondary roughness features deteriorate too. A deterioration in relative density has been reported as the VED increases due to gas entrapment within the meltpool (Salem et al., 2019), but this phenomenon was not manifested at the largest VED used in this study (230 J/mm³).

Results from experiments 1–2 informed that a VED of 55 J/mm³ can deliver a large relative density (98.7 ± 0.1%, only 1.1% lower than the maximum 99.8 ± 0.1% achieved at 230 J/mm³), while maintaining low values of secondary roughness features [ring thickness (t), Sa and Sz] [Figure 3(b)]. Secondary roughness layer thickness was included in our measurements as it introduces an additional feature to characterise surface finish and is rarely reported in surface roughness studies. Layer height was decreased from 30 to 20 µm after experiment 1 because the latter produced specimens of smoother surface finish [see comparison between results from experiment 1 sets 2 and 3 in Figure 3(a) and experiment 2 sets 1 and 5 in Figure 3(b)]. Experiment 3, where VED values were fixed at 55 J/mm³ by varying laser speed and power [Figure 3(c)], shows relative density values spanning 96.0%–98.5%, and it peaks at speed is 889 mm/s and power 108 W. At these conditions the secondary roughness is at its optimum because its thickness, Sa and Sz are all minima.

The results from the Scan Strategy study (experiment 4) are displayed in Figure 3(d). Strategy 2 displays an increase in secondary roughness values (t, Sa, Sz) compared to the reference strategy (Strategy 1). It is hypothesised to be caused by an artefact inherent to Strategy 2, whereby the infill overlaps with the sidewall, causing remelting and increased secondary roughness [vertical stripes in Figure 4(e)]. This surface finish was prevented when strategy 6 was adopted [Figure 4(f)].

Strategies 3, 4 and 5 produced coupons of similar values of secondary roughness thickness and Sa, and the relative density correlated negatively with Sz. Strategy 6 delivered the highest averaged relative density values (99.6 ± 0.1%, a 1.09% increase from the reference strategy). The secondary roughness characteristics achieved using this strategy were markedly lower than the reference strategy’s, with an overall smooth finish of the surfaces [Figure 4(f); thickness 75.3 ± 2.1, Sa 2.85 ± 0.22, Sz 30.13 ± 2.79 µm]. The overlapping of the meltpools exerted by the “double pass” printing path strategy led to VED doubling in the path axis, as per equation (1). Remelting has been reported as a process to optimise the surfaces in demanding applications such as tool making for injection moulding (Simoni et al., 2021), and our results are in agreement. The maximisation of the relative density is guaranteed with this “double pass”, as the outer half-meltpool was not overlapped, leading to a desirable combination of properties, with the secondary roughness thickness being equivalent to those of the VED 55 J/mm³ specimens and the relative density equivalent to the 110 J/mm³ specimens.

Exemplars of coupons surface examined under microscopy are presented in Figure 5. Different secondary roughness thickness [Figures 5(a), (c), (e)] and hatch porosity can be observed [Figure 5(g)] in the top views of the coupons. The side views provide observations along the build direction: voids and flaws [Figure 5(b)], and coarse [Figure 5(d)] or fine finishes [Figure 5(f)], depending on the printing parameters. Optimum surface features were achieved in Experiment 4, printing strategy 6 with a “double pass” approach where the thickness was minimal [Figure 5(e)] and the surface presented no voids [Figure 5(h)].

The statistical analysis performed on the results from experiments 1–3 reveals strong correlations and statistical significance between the dependent and independent variables. The pictorial representation of Pearson’s correlations, listing the correlation coefficients and the corresponding p-values are depicted in Figure 6. Table 4 displays the results from the statistical analysis of each pair, including p-value, t-stat and t-critical. VED, a dependant variable on laser speed, presents a strong negative correlation with it, as expected from eq.1 [Figure 6(a)]. On the contrary, while VED is also dependant on laser power, their correlation is much weaker. This could be rooted in the smaller range of power values explored in this study. This is also mirrored in the correlations between laser power and density or secondary roughness features (thickness, Sa and Sz). In other words, within the power values explored, it was difficult to control the output features of the printed coupons. The correlations between the output features and VED are strongly positive; the correlations with the laser speed are strongly negative, positioning laser speed as an important factor in the control of resultant density and surface finish of the coupons. Furthermore, the p-value of every laser speed pairing is statistically significant in all instances. The pairings that are statistically insignificant (p > 0.05) are VED with roughness (Sz) and relative density with thickness [Figure 6(b)]. Despite being positively correlated, this lack of statistical significance could indicate that these can be controlled independently. This is an important practical implication that has not been reported to date. Pairs such as VED vs density, power vs thickness and density vs roughness (Sz), are marginally correlated. But their |t-stat| vs t-critical values are in close proximity, so it is unclear whether the null hypothesis should be rejected. In two instances there is non-normality of residuals either. Considering this, laser power is a poor predictor of thickness and relative density, because their pairings are neither correlated nor significant. As a practical implication for the end-user, laser power should not be used as a variable to control relative density nor surface finish, and laser speed should be used instead.

For the validation exercise lattice trabecular structures were selected. They are highly stochastic and typically feature very small pores and thin struts with a variety of angles and lengths, making them very challenging to manufacture and prone to defects if the printing parameters are not optimal. Lattice 1 was manufactured using parameters set 1 from experiment 1 and the reference scan strategy (strategy 1) from which a sub-optimal structure was expected, yielding a material relative density of only 97.9 ± 0.2% and a low volume porosity of 69.7 ± 0.2%. Lattice 2 was manufactured using printing parameters from strategy 6 in experiment 4, and its material relative density resulted in 99.1 ± 0.5% with a volume porosity of 72.7 ± 0.2% (intended design volume porosity was 76%). Their results are presented in Table 5. Details of the cross-section from the two trabecular structures are presented in Figure 7 for comparative purposes. The lattice [Figures 7(a), (d)] printed at sub-optimal conditions [Figures 7(b), (e)] displays an evident occurrence of process-induced flaws, discontinuous struts, incomplete fusion instantiated as broken line profiles, “hatch porosity” and closed pores that are not observed in the lattice printed using the optimum parameter set and strategy [Figures 7(c), (f)].

Based on that it could hypothesised that lattice 1 volume porosity should be larger. However, visual inspection of the µCT scans provides an assessment of the quality of the structures and provides an explanation to the lower porosity value measured in lattice 1. Incomplete hatching creates a small-scale network within the strut creating open porosity. When the experimental measurements of the porosity are conducted, this discontinuous network allows the ingress of acetone into the pores and confounds the density measurement, giving an artificially low volume porosity. While hatch porosity has been reported when there is insufficient overlap between the infill and border scans (Yin et al., 2023), this study shows [Figure 5(g)] that a detrimental effect also appears when the overlap is too large (i.e., strategy 2 in experiment 4). Strategies 5 and 6 in experiment 4 were adopted to eliminate overlaps via a conformal scan style, and this resulted in an improvement in material density [Figure 5(h)]. Furthermore, the optimum strategy identified in this study (Strategy 6) directs no rotation of infill between layers which results in an increase in density when compared to other strategies with infill rotation. This result disagrees with other studies that have reported that parallel infill scans with no rotation between layers can lead to greater thermal gradients and residual stresses in the components, leading to higher quantities of voids (Rehman et al., 2023). Because no sign of thermal gradient has been captured in the bulk alloy [Figure 5(f)], it is hypothesised that the reduction of hatch porosity outweighs any possible increase in porosity caused by parallel infill scans.

Every machine/material/application combination requires an assessment of the trade-off between material density and secondary roughness when generating optimum printing parameters to improve surface finish as well as ensuring a material continuum. To contribute to the international efforts to characterise Additive Manufacturing hardware when processing metal powders, this study has established optimum printing parameters and path scanning strategy for cp Titanium powder being processed in a Trumpf Truprint 1000, as well as identifying the more relevant variables influencing the manufacturing process. From this study we conclude that the effect of laser speed is the most impactful on material density and secondary roughness, with statistical significance, while the effect of laser power is only moderate. Based on that, the following set of parameters produced optimum results: 108 W laser power, 889 mm/s laser speed, 55 µm hatch distance and 20 µm of layer height to generate a total VED 110 J/mm³, delivered via a “double pass”, each of VED 55 J/mm³, in a conformal infill fashion. The solid coupons had a relative density 99.6 ± 0.1%, secondary roughness thickness 75.3 ± 2.1 µm, roughness Sa 2.85 ± 0.22 µm and Sz 30.13 ± 2.79 µm. A conformal infill style in the scan strategy gives greater material density than a conventional border/crosshatch style by eliminating the opportunity for hatch porosity to appear and creating a smoother surface finish. From the statistical analysis performed we can extract that while VED vs roughness Sz and density vs secondary roughness layer thickness are positively correlated, it is without statistical significance, which means that they could be controlled independently. In the validation step, lattices manufactured using the optimal parameters and scan strategy delivered higher quality struts and structures, with continuous infills and fewer instances of closed pores, and their as-manufactured porosity was closer to the as-designed intent, an indication of a more successful print.

The authors are grateful to the Engineering and Physical Sciences Research Council (EPSRC, grants EP/L014998/1 and EP/V007335/1) and the Wolfson School Bursaries for financial support.