Effect of process parameters on surface integrity in laser powder bed fusion of Ti-6Al-4V alloy Authors: Sabine Le Roux, Mehdi Salem, Mohammad Shabbar, Buchibabu Vicharapu, Nelson Muthu, Kamel Moussaoui, Anis Hor, Vincent Velay Published: 2025 ABSTRACT AbstractLaser powder bed fusion (LPBF) printing of Ti-6Al-4V often suffers from surface defects such as weld traces, unmelted particles, spatter, and porosity that degrade surface integrity and induce significant residual stresses due to a higher thermal gradient. This study aims to elucidate the evolution of surface topography and residual stresses as a function of build height and process energy input during LPBF fabrication. Ti-6Al-4V samples were produced at volumetric energy densities ranging from 21.8 to 65.4 J·mm⁻³ by varying laser power and scanning speed. Focus variation microscopy measurements show propagation of surface irregularities extending over several tens of layers. Higher energy densities resulted in stable melt pools and reduced surface roughness. The computed and measured residual stresses agreed well for the sound builds. In contrast, notable deviation was observed for the conditions that exhibited high porosity. Overall, the findings show that residual stress evolution in LPBF significantly influences surface morphology, which further supports the importance of process energy optimization to achieve a sound part.Keywords: Laser powder bed fusion, Surface integrity, Roughness, Residual stresses, Ti-6Al-4VSubject terms: Engineering, Materials science Introduction Advanced additive manufacturing (AM) techniques are valued for their ability to produce complex near-net-shaped metal parts without the need for expensive tools or molds. Powder Bed Fusion (PBF) techniques such as laser powder bed fusion (LPBF) and electron beam melting (EBM) in particular, have strong economic potential, since high-performance engineering materials can be used in high-value-added applications such as aerospace and biomedical1–3. Titanium alloys (mainly Ti–6Al–4V) are widely used due to their outstanding mechanical properties, corrosion resistance and biocompatibility3–6. LPBF is a layered material addition method that enables the production of solid 3D parts by selectively melting thin layers of metal powder using a high-power laser beam7. This process makes it possible to manufacture lightened structures by the integration of internal holes that are impossible to obtain using conventional manufacturing (e.g. casting, forging, turning, milling, etc.). This advantage, however, leads to a significant increase in the Surface-Volume ratio of LPBF parts, and it is known that these processes induce poor surface morphology, undesirable surface roughness, and high tensile residual stresses2–4,8–10. Such defects have a detrimental effect on the durability of structural components7,11, thus, surface integrity remains a major issue in the LPBF process. The surface integrity aspects considered in this study consist of the external topography of the surface and the residual stresses of the internal subsurface layer. The surface roughness of LPBF parts is influenced by the melt pool dynamics and wettability driven by various processing conditions. Many factors such as material and powder particle size, laser power, scanning speed, layer thickness, hatch distance, base-plate preheating, scanning strategy and laser re-scanning, by influencing the thermal history (cooling rates and temperature gradients) and the solidification behaviour of the material, contributes to variations in surface roughness4,8,12–16. Optimizing laser power and scanning speed, as well as hatch spacing and layer thickness, is crucial because of their impact on the input energy density at the surface of the melt pool3,17. Insufficient laser energy density - due to too low laser power or excessive scanning speed - causes incomplete melting of the powder or formation of porosity, both of which negatively impact the relative density of the solidified part3,18,19. A correlation between surface texture and relative density of LPBF parts was established in18,20, demonstrating that surface quality could even be used as an indicator of the amount of internal defects such as lack of fusion or porosities. Balling is another unwanted defect in LPBF, which may occur when the melt pool breaks off into small spheres under high surface tension between its center and its edges, resulting in increased roughness as these metal droplets solidify and remain stuck on the surface7,21–24. This phenomenon can be reduced by optimizing process parameters such as laser remelting25, repetition rate or scanning speed24. To better understand the mechanism of surface defect generation in LPBF, numerous studies have focused on single-melt track experiments26–29. However, these investigations do not consider the interaction between adjacent scanning tracks and superimposed layers, and therefore how defects accumulate during successive layer deposition. As shown in30, scanned tracks indeed cause partial remelting of pre-scanned tracks, resulting in a heat build-up effect. Correct measurement and analysis of the surface topography of AM metal surfaces is not trivial due to their high complexity and irregularity, with features at a wide range of scales, non-uniform optical properties, low aspect-ratio formations, high surface slopes, re-entrant features, undercuts and deep recesses2,9,16,31. Among the wide variety of tactile and optical instruments, focus-variation microscopy (FVM)32–34 has been shown to be particularly suitable for AM surface measurement, due to its good repeatability6,35,36 and ability to capture deep recesses and steep surface flanks2,5,6,31,35. X-ray computed tomography is also used in roughness measurements, although it is often used for the examination of internal defects37. Profile roughness parameters are still widely used to characterise the surface quality of LPBF surfaces7,8,13,14,23,25,38. However, areal field parameters defined in ISO 25178-2 are considered to provide more comprehensive information on texture and anisotropy and are statistically much more significant4,5,35,39–41. In addition, Grimm et al.16 clearly showed the absolute necessity of measuring an adequate surface to achieve a reliable analysis on AM surfaces. High level residual stresses resulting from the combination of high heat flux densities, steep temperature gradient and rapid heating and cooling cycles are also a key challenge in LPBF3,42. It has become widely accepted that these stresses - which are mainly thermal stresses - cause cracks and distortion of the parts43. As shown in44, the residual tensile stresses of as-built Ti-6Al-4V alloys exhibit significant anisotropy considering X and Y directions. The longitudinal stresses, which increase with the length of the scanning vector, are the principal component of stresses due to the large thermal gradient parallel to the scanning vector45. The process parameters like laser power, scanning speed, hatch spacing, and scanning strategy also have a significant effect on the residual stress formation in Ti-6Al-4V46–49. Xiao et al.46 observed that the impact of scanning speed is greater than that of laser power, while hatch spacing exhibited the least influence among the three parameters. Under the same energy density, reducing the laser power and increasing the laser exposure time can reduce the residual stress. Furthermore, residual stresses can be lowered by increasing the layer thickness49, as also shown by Vrancken50, who identifies this parameter as more influential than scanning speed and laser power. Balbaa et al.17 investigated the effects of various combinations of laser power, scanning speed and hatch spacing on relative density, surface roughness and residual stresses. However, the study showed shortcomings of using single-track depositions as the sole predictor of the quality of part, not considering how residual stresses evolve with the increase in the number of layers. Moussaoui et al.51 reported a ~ 30% underestimated measured residual stress value in Ti-6Al-4V using X-ray diffraction and attributed it to the poorly defined shape of the diffraction peaks, low intensity and background noise. To overcome such experimental constraints, a simulation-based approach is needed for more accurate residual stress prediction in LPBF-printed parts52. In the present study, Ti-6Al-4V samples were fabricated by LPBF with various combinations of laser powers and scanning speeds and an increasing number of layers, to investigate the evolution of the top surface integrity during layer-by-layer buildup. A multiscale analysis of surface defects were conducted from scanning electron microscope (SEM) observations and surface topographies acquired by focal variation microscopy (FVM). The evolution of surface defects as a function of the number of layers and process parameters was monitored by profile and areal field roughness parameters. Residual stresses were evaluated by X-ray diffraction, and a coupled finite element (FE) thermomechanical simulation was developed and completed by comparison with experimentally measured residual stresses. Experimental and simulation procedure Material and LPBF experiments Parallelepiped samples with a square base of 8 mm × 8 mm were produced by LPBF using a ProX DMP 200 machine (Phenix Systems, France). Titanium alloy, Ti–6Al–4 V was used as powder material (Fig. 1a-b). The powder was plasma-atomized, resulting in a particle size ranging from 15 μm to 70 μm and a median diameter d50 of about 35 μm, as measured by laser granulometry (Fig. 1c). The manufacturing was carried out in a protective argon atmosphere to prevent oxidation, using a non-preheated Ti–6Al–4 V base plate and without supporting structures. For all experiments, a zigzag scanning strategy along the recoating direction was used, without contour remelting. The fixed process parameters are reported in Table 1. Scanning electron microscope (SEM) observation of atomized titanium alloy powder used in LPBF experiments (a), detailed view (b), and particle size distribution of powders (c). Table 1Fixed process parameters used to manufacture the samples on the LPBF machine.Laser spot diameter d (µm)70Hatch spacing hs (µm)85Layer thickness tl (µm)60Scanning strategyzigzag (along θ = 0°) Fixed process parameters used to manufacture the samples on the LPBF machine. Various powers (P) and scanning speeds (s) were investigated (while the hatch spacing (hs) and the layer thickness (tl) were kept constant), resulting in three levels of volumetric energy density (Ev = P/(s.hs.tl)) ranging from 21.8 to 65.4 J.mm− 3 (Table 2). For each (P, s) condition, samples with various thicknesses (i.e., numbers of layers n = 1, 2, 3, 5, 10, 20, 50, 100 and 200) were manufactured. Table 2Pairs of laser power (P)/scan speed (s) conditions investigated in the experiments.Laser power P (W)100200300Scanning speed s (m.s− 1)0.60.61.80.91.82.7Volumetric energy density Ev (J.mm− 3)32.765.421.865.432.721.8 Pairs of laser power (P)/scan speed (s) conditions investigated in the experiments. Surface characterization Image acquisition and pre-processing After cleaning the as-built samples in an ultrasonic bath, their upper surface was observed with a scanning electron microscope (SEM). Topographic images were acquired at different scales using a secondary electron detector, in order to determine the morphology and nature of surface defects. Height maps of the top surface of the samples were also acquired using an Alicona Infinite Focus SL optical 3D measuring system. This FVM is based on the small variation of focusing depth with vertical scanning. The samples were mounted on a motorised stage and illuminated with ring light illumination. A 10× objective lens was used, giving a XY range of 9.4 mm × 9.4 mm constituted of (6 × 6) fields of view, including the base plate surface near the edges of the sample. The vertical resolution was fixed at 100 nm and the lateral resolution at 4 μm. Since the samples were still attached to the baseplate during the measurement, they can be considered distortion-free. The surface analysis software MountainsMap (DigitalSurf, France) was used to process the 3D surface maps acquired by the FVM. The first step consisted of applying a slight rotation (< 1°) to compensate for minor misalignments of the measured surfaces with respect to the X and Y axes. Then, the aligned raw height cards were cropped to (8 mm × 8 mm). The non-measured points (NMP) were filled by interpolation based on the values of their neighbourhood. It should be noted that these missing points represent, on an average, less than 1.5% of the measured area, considering the total sample area (and less than 0.7% in the useful central area considered to determine the filtered roughness parameters). However, this proportion may be higher for certain process conditions, especially low laser power or high scanning speed (Table 3) and for 10 or more layers (Table 4), without exceeding 8% for the worst condition. This percentage of NMP is actually related to the total amplitude in z and the presence of steep slopes on the surface, which increase as the number of layers increases. Table 3Statistics of the percentage of unmeasured points (NMP) on the height maps measured by focus variation microscopy (FVM) for the different sets of process parameters.Process parametersNMP (%) on total sample area (8 mm × 8 mm)NMP (%) on central area(4.8 mm × 4.8 mm)P (W)s (m.s− 1)Ev (J.mm− 3)MeanStandard dev.MeanStandard dev.1000.632.72.582.892.113.082000.665.40.370.360.190.212001.821.81.151.130.450.493000.965.40.280.240.080.093001.832.70.650.790.240.453002.721.83.233.010.730.67 Statistics of the percentage of unmeasured points (NMP) on the height maps measured by focus variation microscopy (FVM) for the different sets of process parameters. Table 4Statistics of the percentage of unmeasured points (NMP) on height maps measured by focus variation microscopy (FVM), for the different numbers of layers (all process parameters combined).Number of layersNMP (%) on total sample area(8 mm × 8 mm)NMP (%) on central area(4.8 mm × 4.8 mm)(n)MeanStandard dev.MeanStandard dev.10.0030.0060.0000.00120.0810.1940.0140.02730.0520.0470.0220.03250.340.340.140.16101.400.850.930.64202.602.471.151.55501.981.980.770.761002.802.590.470.532003.143.362.193.61 Statistics of the percentage of unmeasured points (NMP) on height maps measured by focus variation microscopy (FVM), for the different numbers of layers (all process parameters combined). Profile analysis For each height map, 239 profiles of 4.8 mm length were extracted in the central area, along both X and Y directions (respectively parallel and perpendicular to the laser scan orientation, θ). The profiles were first levelled by least squares adjustment, before being filtered to extract roughness. For comparison purposes and due to the limited sample size, a Gaussian filter with a cut-off (λc) of 0.8 mm was systematically used, although this cut-off does not necessarily comply with the ISO 4287 standard53. Indeed, for samples with a high number of layers, a higher cut-off value (2.5 mm or even 8 mm) would have been prescribed. Then, amplitude parameters (Ra, Rq, Rz, Rt, Rp, Rv, …), were calculated per ISO 428854. The asymmetry of the height distribution of the profile can be evaluated by the skewness parameter RSk. A value near zero reflects a Gaussian surface (with a symmetrical shape for the height distribution), RSk being either negative or positive depending on whether the bulk of the material is above or below the mean line55. Spatial (root mean square deviation of the profile RSm) and hybrid parameters (root mean square slope of the assessed profile Rdq) were also determined. The initial roughness of the base plate was also determined using a Gaussian filter λc of 0.8 mm. In addition, the total height of the sample was measured on a mean profile extracted in the middle of the top surface (perpendicularly to the scanning direction) and after straightening with respect to the two side surfaces of the plate, as shown in Fig. 2. Method for determining the total height of the LPBF sample on a profile extracted in the middle (here for 10 layers). The extreme borders of the sample are excluded for the measurement. Determination of areal field parameters On the raw surface on which NMP were interpolated, an area of interest of 8 mm × 8 mm was first extracted by excluding the base plate area as well as the extreme edges of the sample, as shown in Fig. 3a-b. A Gaussian S-filter nesting index (low-pass filter) of 8 μm was then applied to remove very high spatial frequencies. This value was limited by the lateral resolution of the measurement device. For optical instruments, the ratio between the S-filter index and the measurement sampling distance must be at least 3:1, per ISO 25178-356. The actual sampling distance of 1 μm gave a ratio of 8:1 that meets the above requirement. Then the surface was straightened by subtracting the mean least squares plane, resulting in the so-called SF surface. After that, a reduced area of 5 mm × 5 mm was extracted in the middle of the SF surface, before applying a Gaussian L-filter (high-pass filter) using a nesting index of 0.8 mm to remove long-scale components. The resulting so-called SL surface (Fig. 3c) can be considered as the filtered roughness surface on which the contribution of waviness (Fig. 3d) has been removed. Processing applied to height maps acquired by focus variation microscopy (FVM): (a) rough height map of the sample top surface (black areas represent unmeasured points of the base plate); (b) SF surface obtained after extraction of a (8 mm × 8 mm) area of interest, filling of unmeasured points, removing of very high spatial frequencies and shape correction; (c) SL (roughness) surface after removing of low frequency in a (4.8 mm × 4.8 mm) area of interest; (d) waviness surface (4.8 mm × 4.8 mm) after removing of high frequency [Example for P = 300 W, s = 0.9 m.s− 1, n = 50 layers]. SF surfaces were analyzed to determine the total z amplitude of the sample in order to get an idea of the lack of straightness that may be for instance linked to the convex shape of the melt pool, while areal field parameters were calculated according to ISO 25178-257 on scale-limited SL surfaces. Height parameters (Sa, Sq, Sz, SSk, etc.) were used to characterize the amplitude and distribution of surface defects. Spatial parameters, such as the texture aspect ratio Str (a unitless parameter between 0 and 1) allowed us to measure the isotropy degree using the surface autocorrelation function55. High values (Str > 0.5) indicate uniform texture in all directions, while smaller values (Str < 0.3) indicate an increasingly strong directional structure. The representation of the polar spectrum of the Fourier transform makes it further possible to determine the directions of surface structures and - if applicable - the privileged texture direction Std (corresponding to the angle with the largest power spectrum). It should be noted that the parameter Std is really significant for anisotropic surfaces (Str < 0.6)55 and should not be considered for lower values of Str. Hybrid parameters were used to describe the complexity of the surface, which depends on both texture amplitude and spacing. The developed interfacial area ratio Sdr expresses the percentage of additional surface area contributed by the texture (curvilinear surface) as compared to the support surface (ideal plane), while Sdq calculates the Root Mean Square (RMS) Gradient and is a general measurement of the slopes that comprise the surface over all directions. Sdq allows differentiating surfaces of similar amplitudes and average roughness. Typically, its value increases with the spatial intricacy of the texture, whether the average roughness Sa changes or not. In the same way, a finer spaced texture will have a higher Sdq value than a surface with the same Sa, but wider spaced features55. Feature parameters (such as Spd, Spc, Sha, etc.) were also considered to characterize the density and dimension (area and mean volume) of surface motifs - either hills/peaks or dales/pits - on the SL surface. Such parameters are derived from image segmentation methods, based upon the application of watershed algorithm and Wolf pruning (with default minimum size criterion, i.e., hills/dales above 5% of Sz, to select only significant motifs)55. Residual stress evaluation Experimental procedure X-ray diffraction was used to estimate the magnitude of macro-residual stresses on the middle of the surface. Residual stresses were evaluated using the conventional sin2() along both longitudinal (sxx) and transverse (syy) directions. Measurements were performed using 6-axis XRayBot diffractometer (MRX Company, France), equipped with a goniometer with a pure Si solid state detector and copper radiation source (Cu-K l = 1.54 Å). The X-ray was generated with a tension of 25 kV and a current of 1 mA. The spot was passed through a 3 mm collimator and the Cu-K l = 1.387 Å was removed using a Ni filter. The three-dimensional scans involved 38 positions, with 19 tilt angles (−37° < < 39°) and 2 rotation angles (φ = 0 for sxx and φ = 90° for syy). Time exposition was fixed at 240 s for each position resulting in a total measurement time of 2.5 h for the two stress directions. The scattered X-rays were recorded at each position for an angular range between 124° < 2< 156° which corresponds to the width of the linear detector. Four diffraction peaks are recorded over this range, and the diffraction peaks were smoothed by a Pseudo-Voigt function, and the background noise was assumed to be linear. The X-ray data treatment and the computation of the residual stresses were done by StressDiff software based on the equation ε = sin2(). The strains were related to the residual stresses by the sin2 law according to continuum mechanics as: where the X-ray elastic constants of Ti-6Al-4V: ½ S2{213} = 11.89 10− 6 MPa− 1 and -S1{213} = 2.64 10− 6 MPa− 1, sφ is the principal stress considering the material under plane stress (s33 = 0), and τφ the shear stress tensor. An elliptic regression of the strains as a function of sin2ψ (ε = f (sin2)) was implemented, and the residual stress data were deduced considering the terms of the equation: ½ S2 sφ as the ellipse slope, ½ S2 τφ sin(2) as the ellipse spacing and 2 S1 sφ as the ellipse origin. Simulation procedure For the numerical simulation, a geometric model made in SolidWorks was imported to the simulation software, then an appropriate finite element (FE) mesh was generated tailored for accurate thermomechanical analysis. Part-scale level residual stress simulation in LPBF processes is highly computationally intensive58, particularly when using general purpose finite element (FE) tools. As part of the analysis, Simufact Additive software was chosen to carryout part-scale simulations, that utilises inherent strain method and voxel-based meshing to produce accurate stress calculations with significantly lower computation time. For numerical analysis, solid-state phase transformation strain was neglected. Changes in the composition were ignored during the laser-material interaction. In addition, all of the stress, strain, and thermal expansion coefficients were taken to be zero in the powder and liquid states of the material. The FE simulation of the LPBF process is governed by Fourier’s 3D heat conduction equation, which captures heat transfer within the deposited material, powder, and substrate. The equation accounts for the transient thermal fields due to the moving laser heat source providing a 3D temperature distribution, given by Eq. 2: where k is the thermal conductivity of the material (in W.m− 1.K− 1), Q is the volumetric heat generation rate (in W.m− 3.s− 1) that accounts for laser scanning parameters by calculating the input energy per unit volume based on scanning speed, hatch spacing, and layer height, ρ is the density (in kg.m− 3) and cp is the specific heat capacity (in J.kg− 1.K− 1). To account for the heat loss to the surroundings, boundary conditions of convection and radiation were applied at the surfaces of the part and the baseplate exposed to the surroundings, as given below: where n is the direction vector normal to the surface, h is the heat transfer coefficient at the surface in W.m− 2.K− 1, TS is the temperature at the surface in K, To is the temperature of the surroundings in K, ∈ is the emissivity, and σ is the Stefan-Boltzmann constant (in W.m− 2.K− 4). The model considers a Gaussian distribution for the laser heat source as shown in Fig. 4a, and it is written following the reference59, as follows: Schematic of the Gaussian heat source for the laser power distribution (a), and voxel mesh for the sample with 100 layers (b). where I is the laser intensity (in W.m− 2), P is the laser power in W, A is the laser absorption coefficient, r is the radial distance from the centre of the laser beam (in m), and w is the laser spot size (in m). The residual stress calculations are performed on the macroscale using the inherent strain method as outlined in57. Total strain accumulation, including elastic, plastic, and thermal strains, is calculated as follows: The inherent strain component of the above equation can be written as: The residual stresses can be calculated from the inherent strain using: where [K] is the elastic stiffness matrix, [u] is the nodal displacement vector, [f *] denotes the nodal force vector due to the inherent strain, and [D] is the stiffness matrix. The total strain for the isotropic material according to the Cartesian coordinate system was calculated as: The residual stress components can then be calculated as: The FE model simulates the deposition of multiple blocks of (8 × 8) mm2 cross-section layer by layer on a substrate. The different blocks are built incrementally with 10, 20, 50, 100, and 200 layers, each having a uniform layer thickness of 60 μm. The experimental and simulation parameters are identical for comparison purposes, as provided in Tables 1 and 2. A moving laser heat source is defined on the selected process parameters, ensuring precise energy input. At the same time, boundary conditions, such as thermal and mechanical constraints, are applied, reflecting operational conditions. The thermophysical properties of the Ti-6Al-4V alloy used in the model were taken from the work of Parry et al.45. The material’s mechanical properties, including the flow curves, are provided directly from the Simufact Additive material library. The emissivity of the material was assumed to be ~ 0.3560. For printed blocks with 10 and 20 layers, the voxel height is equal to the height of a single layer, which is 60 μm. For blocks with more than 20 layers, two layers are combined into a single voxel with a height of 120 μm. This is illustrated in Fig. 4b, which presents the voxel mesh representation for a sample with 100 layers, comprising a total of 50 voxel elements along the height. The other voxel dimensions, the surface and volume element sizes, are uniformly set at 0.4 mm, ensuring consistency and balancing computational efficiency with model accuracy. It is to be noted that the simulations in Simufact Additive are tailored for part-scale models; therefore, the simulations are performed for samples with 5 or more layers. Results and discussion Roughness analysis Surface defects at micro and meso-scale The evolution as a function of the number of layers (n) of the arithmetic mean roughness (Ra), measured perpendicular and parallel to the laser scanning direction, is shown in Figs. 5a-b, respectively for high (300 W) and lower (100 and 200 W) laser powers. The initial roughness of the base plate (measured across the polishing grooves) is indicated for comparison. All curves exhibit a roughly sigmoidal evolution, indicating an increase in roughness during the construction of the first layers, followed by stabilization (or even a slight decrease) between a few dozen and a hundred layers. Figure 5a also indicates that for high laser power (P = 300 W), increasing the scanning speed promotes the roughness amplitude of surface defects, particularly for the initial few layers. Although not all results are presented here, it should be specified that very similar evolutions as a function of n have been observed for other amplitude parameters, with, of course, higher values in particular for Rz and Rt, which can reach 150–200 μm depending on the process conditions. Evolution of the arithmetic average roughness (Ra) versus the number of layers (n), for samples built with (a) a high laser power (P = 300 W); (b) a lower laser power (P = 100–200 W) [The abscissa is represented with a logarithmic scale for a better visualization of the first layers]. Three categories of surfaces can be distinguished, according to the stabilized Ra value (asymptotic part in Figs. 5a-b): those whose Ra is ≥ 30 μm; those whose Ra is ≥ 30 μm; those whose Ra is between 15 and 25 μm; those whose Ra is between 15 and 25 μm; those whose Ra is ≤ 30 μm, particularly in the laser scanning direction. those whose Ra is ≤ 30 μm, particularly in the laser scanning direction. The smoothest surfaces (last case) result from process parameters leading to high volumetric energy densities (Ev = 65.4 J.mm− 3). It can be noted that for these surfaces, the roughness of the very first layers in the scanning direction is even lower (Ra < 2 μm) than that of the base plate (Table 5). Moreover, the roughness measured in the two orthogonal directions is significantly different for the high energy densities, whereas it is similar in the other cases. Table 5Initial roughness of the base plate measured across the Polishing grooves (along X direction) according to ISO 4288 (evaluation length = 4.0 mm, Gaussian filter λc = 0.8 mm).Roughness parameterRa (µm)Rq (µm)Rz (µm)Rt (µm)Rp (µm)Rv (µm)Mean1.932.5113.218.86.506.69Standard deviation0.210.311.882.430.901.01 Initial roughness of the base plate measured across the Polishing grooves (along X direction) according to ISO 4288 (evaluation length = 4.0 mm, Gaussian filter λc = 0.8 mm). To better highlight the surface anisotropy suggested by the difference observed between the Ra values parallel and perpendicular to the laser scanning direction, it may be useful to consider areal field parameters and, more specifically, spatial parameters. As shown in Fig. 6a, the overall increasing evolution of the texture aspect ratio Str as a function of n, especially for the first layers, indicates that all surfaces are strongly anisotropic at the beginning of the construction (Str < 0.3), reflecting the laser scan imprint. For the least rough surfaces (produced with Ev = 65.4 J.mm− 3), the anisotropy remains marked up to several tens or even hundreds of layers (Fig. 6a). Evolution of the texture aspect ratio (Str) versus the number of layers (n), for all investigated process conditions (a); polar spectrum of samples built with P = 200 W and s = 1.8 m.s− 1 and 1 layer (b), 3 layers (c), 10 layers (d) and 200 layers (e), indicating a disappearance of the initial anisotropy with increasing n. It can also be noted that the preferential direction of the anisotropic surfaces corresponds to the orientation θ of the laser scan (close to 0°). However, for the roughest surfaces (produced with lower Ev), the anisotropy quickly disappears (Str > 0.5) by increasing n, as shown by the polar spectra indicating the privileged texture directions (Figs. 6b-d). This is probably due to numerous spherical defects (such as unmelted particles, spatters and porosities) that “pollute” the surface, as revealed by SEM observations in Fig. 7. These micrographs further reveal irregular melt tracks with non-uniform width (necking) and a bumpy surface (Fig. 7a-d). In addition, periodic ripples forming “fish scale” patterns are clearly visible in the 2-layer sample produced with low laser power and scanning speed (Fig. 7a). This kind of unwanted defects can be attributed to an instability of the melt pool. It has also been observed in 316 L single-tracks, and linked to violent fluctuations of the melt pool caused by the incorporation of agglomeration spatter (i.e. large projections originated at the front of the melt pool due to clusters of spatter particles)61. SEM observations of the top surface of the roughest samples produced with P = 100 W, s = 0.6 m.s− 1 (a, c, e) and P = 300 W, s = 2.7 m.s− 1 (b, d, f), indicating an increase in surface defects (such as unmelted spherical powders, spatters and large porosities) by increasing the number of layers: n = 2 (a, b), n = 10 (c, d) and n = 200 layers (e, f). Figure 8a further reveals that for samples consisting of 10 or more layers, the surface is strongly isotropic for low volumetric energy densities (as also shown in the relative SL height map, Fig. 8b), and the anisotropy is stronger as Ev increases. For a median Ev (32.7 J.mm− 3), a significant increase in laser power (from 100 to 300 W) also appears to promote the anisotropic character of the surface (Figs. 8c-d). This can be explained by a more efficient melting of the powders, better revealing the laser scanning trace which can be clearly distinguished on the height map of samples produced at 65.4 J.mm− 3 (Fig. 8e). Effect of the volumetric energy density (Ev) on the texture aspect ratio (Str) for samples consisting of 10 layers or more (a), and associated SL roughness height maps for 10 layers and Ev = 21.8 J.mm− 3 (b), 32.7 J.mm− 3 (P = 300 W and s = 1.8 m.s− 1) (c), and 32.7 J.mm− 3 (P = 100 W and s = 0.6 m.s− 1) (d), and 65.4 J.mm− 3 (e). In Fig. 9, plotting the average roughness Sa (integrated over the entire surface) versus n, the same orders of magnitude and overall trend as for equivalent profile parameter Ra (see Figs. 5a-b) are observed. Evolution of the arithmetic mean height (Sa) versus the number of layers (n), for all investigated process conditions. In Fig. 10, detailed SEM observations of 1-layer samples show that optimal energy conditions (Ev = 65.4 J.mm− 3) lead to the formation of a very flat surface with continuous and smooth melting tracks, which reflects good stability of the melt pool. In contrast, surfaces produced with low energy density (21.8 J.mm− 3) are disturbed by marked bumps along two directions X and Y. The undulation along the scan direction consists of a discontinuity of the scanning tracks when the scanning speed is very high (s = 2.7 m.s− 1). This is associated with a balling phenomenon, which reveals a highly irregular and unstable melt pool. This type of defect is in fact often observed when the scanning speed is excessively high, resulting in insufficient penetration depth and energy input3,17,19,26,28,62. Balling has been widely described in the literature as being due to a lack of melting of the substrate and the resulting spheroidization effects, which severely impede interlayer connection21,29,63. In the direction perpendicular to the scanning, undulations correspond to the separation of weld beads. SEM observations showing the different morphologies of the melt pool after melting of the first layer (n = 1), for all investigated process conditions. The first and third columns focus on the end of the scan at the bottom edges of the sample, allowing to determine the dimensions of the weld bead (W) and the crater left by the laser at the end of the scanning (a,b), while the second column focus on the center of the top surface, with a magnified view showing the apparent width of scanning (Wa). The curves for the two samples produced with the same median Ev (32.7 J.mm− 3)​​ show very different trends versus n (Fig. 9): a sharp increase in roughness (Sa > 30 μm) when the laser power and scanning speed are low (P = 100 W, s = 0.6 m.s− 1), and a much smaller increase (Sa < 20 μm) when the laser power and speed are higher (P = 300 W, s = 1.8 m.s− 1). It can be assumed that in the first case, insufficient output laser energy per unit time did not allow optimal melting of the powders, as previously shown in SEM surface micrographs (Figs. 7a, c and e). The same explanation can be given for the sample produced with the highest scanning speed (2.7 m.s− 1), also characterized by the presence of numerous partially melted powders and spatters on the surface (Figs. 7b, d and f). Figure 9 further indicates that for samples produced with the same scanning speed (e.g. 0.6–1.8 m.s− 1), increasing the laser power (from 100 to 200 W or from 200 to 300 W) results in a decrease in the average roughness, in accordance with the results reported in64. To better highlight the dynamics of roughness construction, the rate of average roughness evolution between two steps (∆Sa/∆n) was calculated and is presented as a function of n (Fig. 11a). The resulting curves confirm the asymptotic evolution of roughness beyond a few dozen layers, as previously noted. The mean evolution of the roughness between 0 and 10 layers was also calculated and plotted against the volumetric energy density (Fig. 11b). It indicates that the lower the Ev, the greater the increase in roughness in the early stages of layer construction. High Ev (especially with high laser power P and low speed s), therefore allows for minimizing the growth of surface defects. Evolution of the rate of the arithmetic mean height (Sa) variation as a function of number of layers (n), for all investigated process conditions (a); and effect of the volumetric energy density (Ev) on the average rate of roughness (Ra) increase from 1 to 10 layers (b). Figure 12a shows, for the same laser power (300 W), the effect of the laser scanning speed s on the mean, maximum, and total amplitude of roughness defects (Ra, Rz, and Rt, respectively) averaged in the asymptotic part (i.e., from 10 to 200 layers). A quasi-linear increase with a fairly good coefficient of determination is highlighted. Similar increase in roughness with scanning speed was also reported in64, but over a narrower range of laser powers (100–200 W) and scanning speeds (0.6–1.6 m.s− 1) compared to the present study. Figure 12b, considering the relative distribution over the surface of peaks and valleys, indicates that the samples produced at high laser power (300 W) and low scanning speed (0.9 m.s− 1) are mainly disturbed by protrusions (Rp > Rv and skewness RSk > > 0). This is consistent with SEM observations of the samples (Figs. 13a-b), which reveal that these surface defects are mainly composed of attached particles, such as unmelted or partially melted powders and spatters. However, the amount of unmelted powders remains relatively lower compared to the other conditions (Figs. 13c-f). When the scanning speed is maximum (i.e. 2.7 m.s− 1), a quasi-symmetrical distribution is observed (RSk close to 0) associated with a greater amplitude of peaks and valleys (Rp and Rv close to 70 μm), reflecting the wavy appearance of the surface (Figs. 13e-f). The defect amplitudes and their symmetry for the highest speed scans resemble the undulations associated with the weld bead, with amplitudes in the order of 35 μm, which corresponds to the median diameter (d50) of the powders at 0.9 m/s. Effect of the scanning speed (s) on: (a) the average, maximum and total amplitude of roughness defects (evaluated by Ra, Rz and Rt, respectively), and (b) the peak/valley amplitude (Rp, Rv) and skewness (RSk), for samples manufactured with a high laser power (P = 300 W). SEM micrographs showing unmelted and partially melted particles on samples manufactured with a laser power (P) of 300 W and various scanning speeds: s = 0.9 m.s− 1 (a, b), s = 1.8 m.s− 1 (c, d), s = 2.7 m.s− 1 (e, f), and number of layers: n = 10 (a, c,e) and 100 layers (b,d,f). Figure 13 also reveals that the morphology of the melt pool, observed after solidification at the beginning of the layer build-up (for n = 1, in Fig. 10), is preserved after the addition of successive layers: namely, a “flat” and homogeneous melt pool (presenting very small undulations, notably in the direction of the laser scan) for conditions with high volume energy density Ev (Fig. 13a), and a more disturbed melt pool (showing many bulging and humping) for lower Ev (Fig. 13c and e). Again, this may be related to the good stability (in the first case)/strong instability (in the second case) of the melt pool during the manufacturing process. Figures 14a-b, reporting the analysis of areal feature parameters, support previous observations. Figure 14a reveals an increase in peak density (Spd) with increasing scan speed (s), followed by stabilization above 1.8 m/s. This reflects a very complex topography at high speeds due to the presence of different surface defects, caused by incomplete fusion of the powder bed. On the other hand, Fig. 14b shows that the mean area (Sda) and volume (Shv) of dales are also dependent on the volumetric energy density (Ev), which favors their growth particularly above 32.8 J mm-3. But the laser power plays a greater role at high Ev (65.4 J.mm− 3), by increasing the volume of dales by a factor of 2 from 200 to 300 W. Effect of the scanning speed s on the peak density (Spd) (a), and of the volumetric energy density Ev on the mean dales area (Sda) and volume (Sdv) (b). The evolution versus number of layers (n) of the surface complexity, represented in Fig. 15 by the developed interfacial area ratio Sdr, indicates that significantly wider spaced texture (Sdr < 25%) is observed for optimal process conditions (i.e. Ev = 65.4 J.mm− 3). In contrast, other conditions lead to a finer-spaced texture (Sdr close to or higher than 50%). Similarly, the profile parameter Rdq (root mean square slope of the profile) shows an increasing evolution of the slope of the asperities with n for optimal process conditions (Figs. 16a-b). A constant difference between the two measurement directions (along and across the laser path) is maintained for optimal process conditions, which can be easily explained by the presence of hatching furrows and small ripples (Fig. 10). For the less efficient process conditions (Ev = 21.8 J.mm− 3), a different trend versus n can be observed, with in particular a high value (Rdq close to 50°). This could be related to discontinuity of the melting tracks, especially for the sample produced at high laser power and speed (P = 300 W, s = 2.7 m.s− 1). These results indicate that surface defects are primarily associated with the presence of unmelted powders and ripples under optimal conditions. In contrast, they are related to undulations and porosities for non-optimal conditions. Evolution of the developed interfacial area ratio (Sdr) as a function of the number of layers n, for all investigated process conditions. Evolution of the root mean square slope of the profile (Rdq) as a function of the number of layers (n), for samples built with: (a) a high laser power (P = 300 W), (b) a lower laser power (P = 100–200 W). The width of the weld beads can be evaluated by the spatial parameter RSm (measured on profiles perpendicular to the laser scanning), which provides a rough measurement of the mean spacing on surfaces having periodic or pseudo-periodic motifs. Figure 17a shows an increasing trend of RSm as a function of n, then a stabilisation from a dozen of layers. Two different cases can be highlighted by considering the first step of the layer-by-layer construction: Evolution versus the number of layers n of the mean spacing (RSm) measured perpendicular to the laser path, for all investigated process conditions. surfaces with RSm (weld bead with) close to or < 100 μm (i.e. close to the hatch spacing and the laser spot diameter), which correspond to unfavourable energy conditions (Ev ≤ 32.7 J.mm− 3) for which the laser back and forth path gives two distinct traces. surfaces with RSm (weld bead with) close to or < 100 μm (i.e. close to the hatch spacing and the laser spot diameter), which correspond to unfavourable energy conditions (Ev ≤ 32.7 J.mm− 3) for which the laser back and forth path gives two distinct traces. surfaces with RSm (weld bead with) close to or > 140 μm (i.e. about twice the laser spot diameter), which correspond to favourable energy conditions (Ev≥ 32.7 J.mm− 3) for which the laser back and forth path is almost the same. This is also confirmed by SEM observations of the first layer, as shown in Fig. 10. surfaces with RSm (weld bead with) close to or > 140 μm (i.e. about twice the laser spot diameter), which correspond to favourable energy conditions (Ev≥ 32.7 J.mm− 3) for which the laser back and forth path is almost the same. This is also confirmed by SEM observations of the first layer, as shown in Fig. 10. RSm values remain constant for high numbers of layers, reflecting the persistence of wave forms on the surface for low Ev. However, RSm values are less relevant, even aberrant (> 400 μm), for optimal energy conditions. This is undoubtedly related to the reduction in the anisotropic character of the surface texture with an increase in the number of layers (Fig. 6), due to surface pollution by spherical defects that mask the weld tracks. The analysis of spatial roughness parameters, such as RSm, is supported by interactive measurements carried out from SEM observations of 1-layer samples (Fig. 10) and reported in Table 6. The actual width of the weld bead (W) was evaluated on the last weld tracks, which corresponds to the last laser path (images in the first column of Fig. 10), and the apparent scan width (Wa) at the center of the sample (enlarged images in the middle column). The results indicate that increasing Ev leads to increase W, while Wa remains roughly stable around 90 μm (± 10 μm), which is logically close to the hatch spacing (85 μm) prescribed in the LPBF experiments. At equivalent laser power, reducing the scanning speed also leads to wider weld beads. The ratio Wa/W can further be used to estimate the overlap of the laser scanning tracks (zero overlap corresponding to a ratio of 1). The results reveal that process conditions leading to low volumetric energy density (Ev = 21.8 J.mm− 3) produces zero overlap, while this overlap becomes more significant as Ev increases (Wa/W < 0.5 for 65.4 J.mm− 3), indicating a more efficient melting of the powder bed leading to an optimized melt pool. Focusing on the end of the laser scanning (third column in Fig. 10), a crater corresponding to the mark left by the laser can be distinguished, whose dimensions (width a and length b) were also estimated (Table 6). For a same Ev, the higher the laser power P, the greater the crater width a; and for a same P, the lower the scanning speed s, the greater the crater length b. This phenomenon can be explained by the Gaussian distribution of the heat flow induced by the laser beam (Fig. 4), which leaves a trail of varying length depending on the scanning speed. Furthermore, different crater shapes (described by the ratio a/b) are highlighted depending on the input energy density: high Ev produce elliptical craters (a/b < 0.4), while pseudo-circular craters (a/b close to 1) are observed for lower Ev. Table 6Width of weld beads (W), apparent width of laser scanning track (Wa) and dimensions of the elliptical crater (a and b) left by the laser at the end of the scan, estimated from SEM micrographs (see Fig. 8) for all process conditions.P (W)s (m.s− 1)Ev (J.mm− 3)W (µm)Wa (µm)Wa/Wa (µm)b (µm)a/b3002.721.880801110951.162001.81001001100851.253001.832.71701000.591601750.911000.61051000.951001500.673000.965.4285850.301809500.192000.6215900.421705250.32 Width of weld beads (W), apparent width of laser scanning track (Wa) and dimensions of the elliptical crater (a and b) left by the laser at the end of the scan, estimated from SEM micrographs (see Fig. 8) for all process conditions. The investigated roughness parameters, especially amplitude parameters, indicate a cumulative effect of defects up to about ten layers. Many parameters, such as the amplitude and distribution of defects (peaks and valleys) on the surface, as well as the spatial parameters characterizing the texture anisotropy (such as Str and RSm), are highly sensitive to the process parameters, especially the volumetric energy density (Ev) and the laser scanning speed. The fusion quality of the first layers, optimized with process parameters leading to a high Ev, strongly influences the flatness of the melt pool and the overlap of the scan trace. From the first layers, process parameters identified as unfavorable (i.e., low Ev induced by high laser scanning speeds or low laser powers) produce disrupted weld beads with poor overlap of the scan traces. This impacts the surface quality of the final part, as well as its density. The accumulation of defects leads to the formation of porosity or cavities during the deposition of successive layers. All these phenomena observed at the micro and meso-scales seem to suggest a problem during powder spreading, induced by the topography of the melted surface, with process parameters identified as unfavourable (in particular, a high laser scanning speed or a low laser power). Indeed, the convex shape of the weld beads observed on samples produced, especially at high speeds, is likely to reduce the volume of powder deposited by the recoater. Combined with insufficient fusion energy, this could result in a deficit of deposited material. Based on selective laser melting experiments on the same material with higher powers and speeds than ours, Qiu et al.18 have indeed shown that increased surface roughness could cause inhomogeneous powder spreading and distribution, and thus inconsistent melt flow and melt pool in the subsequent layers. The irregularities of the topography, amplified by the cumulative effect during the layer-by-layer construction, could then lead to an increase in porosity and thus a decrease in the density of the parts, as shown by Yonehara et al.20 highlighting a correlation between roughness and density of LPBF parts. A macroscopic analysis was conducted on SF surfaces (including the sample edges) to examine defects of higher order than roughness. Surface defects at macro-scale For all the investigated process conditions, the evolution of the measured sample height (Hm) is presented as a function of the number of layers (n) in Fig. 18a, which also shows for comparison the estimated theoretical height (Ht). A gap between the calculated and measured height is observed, which varies depending on the number of layers and the process parameters. It should be noted, however, that this gap is overestimated regarding the macroscopic topography and the evaluation method based on the lowest point of the surface (as set out in Fig. 2). Figure 18b, better highlighting this difference (Hm - Ht) up to 50 layers, reveals a significant material deficit for the first layer and up to 20 layers, with an effect of process parameters which becomes significant from 5 layers. From 50 layers, the actual height then exceeds the theoretical height. It is interesting to note that the sample exhibiting the maximum height deficit (for n = 20 layers) and the one with the highest height excess (for n = 200 layers) both result from process conditions inducing the highest volumetric energy density. Evolution versus the number of layers (n) of: (a) the measured sample height Hm (the theoretical height Ht, calculated as the product of the number n and thickness t of layers, is shown in dotted line for comparison); (b) the height difference (Hm - Ht), indicating a deficit of material at the beginning of the construction of layers (up to 20 layers), for all investigated process conditions. Raw height maps show, particularly for samples with a large number of layers, that the sample edges are raised (Fig. 3a). This issue, which can be amplified by powder spreading issues before layer melting, has elsewhere been highlighted in64–67. The resulting shape deviation can be quantified by the total flatness deviation parameter (FLTt) on the unfiltered SF surface, which was plotted as a function of the number of layers n in Fig. 19, for all process conditions. The curves indicate an overall increase, followed by stabilization at a high number of layers. FLTt does not exceed 100 μm up to five layers, and beyond, it increases strongly depending on the process parameters. Asymptotic evolution of the total flatness deviation (FLTt) versus the number of layers (n), for all investigated process conditions. FLTt was averaged in the asymptotic part (i.e. for samples with 10 layers or more) and represented as a function of the scanning speed (Fig. 20a) and the volumetric energy densities Ev (Fig. 20b). As can be seen on these two figures, FLTt increases as the scanning speed decreases and Ev increases, especially when the laser power is low. The maximum FLTt (> 500 μm) is observed for the process conditions with P = 200 W and s = 0.6 m.s− 1, although - as shown by previous roughness results on the SL surface which does not take into account the edges of the sample - the melt pool exhibits a flat surface in the center. In fact, for such optimal energy conditions (Ev = 65.4 J.mm− 3), the height of the solidified melt pool differs between its center and its edges, as revealed by the raw surface height maps of both samples (Figs. 21a-b). Previous SEM observations (Fig. 10) confirmed that the deceleration and change of direction of the laser scanning disturb the shape of the melt pool on the edges. Effect of the scanning speed s (a) and the volumetric energy density Ev (b) on the total flatness deviation FLTt, for samples with 10 to 200 layers. Comparative analysis of material surface (based on profile extraction from the surface maps) of samples fabricated with high volumetric energy density (Ev = 65.4 J.mm− 3) and 10 layers: (a) P = 300 W, s = 0.9 m.s− 1 and (b) P = 200 W, s = 0.6 m.s− 1. For both sets of process parameters (P = 300 W, s = 0.9 m.s− 1 and P = 200 W, s = 0.6 m.s− 1) resulting in a high volumetric energy density (Ev = 65.4 J.mm− 3), relatively flat melt pools were previously highlighted considering the scale-limited SL surface. This may allow a larger amount of powder to be deposited compared to disturbed surface topographies produced with less favourable energy conditions. A comparative analysis, carried out to quantify the amount of material constituting the two samples (represented by the green surfaces measured from extracted profiles in Figs. 21a-b), clearly highlights a significant difference (> 20%). Figure 21b further shows a height difference of up to several hundred micrometres between the edges and the center of the sample. Hence manufacturers’ recommendations to create internal and external contours (remelting strategies) to achieve a more regular surface. Adjusting the process parameters, particularly by decreasing the laser power or increasing the scanning speed (as previously shown in Fig. 20a and also reported in67 or optimizing the blade and speed of the recoater68, can also help reduce such edge effects. Considering all surface defect scales (from micro- to macro-scale), the optimal process condition emerging from the different combination of process parameters investigated thus corresponds to 300 W laser power and 0.9 m.s− 1 scanning speed (leading to a volumetric energy density Ev of 65.4 J.mm− 3). This is in agreement with several studies aimed at determining the optimal process window for the same material (Ti–6Al–4 V alloy)69,70. However, in order to complete the analysis of surface integrity and to confirm the previous assumptions, residual stresses should also be investigated. Effect of process parameters on residual stresses The evolution with the number of layers n of the measured residual stress sxx (along the scanning direction) and syy (perpendicular to the scanning direction) is plotted respectively in Fig. 22a-b, for various combinations of laser power and scanning speeds. The curves show that for almost all process conditions, samples with fewer layers exhibit higher stress values, while prints with more layers show significantly lower stresses. In most cases, an increase in sxx is observed up to 5 layers, then the stress intensity decreases with the number of layers (Fig. 22a). Whatever the process conditions, the peak sxx values are obtained for a small number of layers (either for 3 or 5). The maximum value (644 MPa) is obtained when laser power is 200 W and scanning speed is 0.6 m.s− 1. As shown in Fig. 22b, syy follows a similar trend with the number of layers, but magnitude-wise it is significantly lower than sxx, with a maximum value of about 250 MPa. A similar stress anisotropy was reported in44. In the scanning direction, the stresses are purely tensile in nature (no compressive stresses), while in the perpendicular direction, compressive stresses can be observed, especially for some samples with at least 50 layers. Evolution of residual stress as a function of number of layers (n), for all investigated process conditions. Initially, the developed model was validated with the experimentally measured residual stresses for samples produced with at least 5 layers and all combinations of laser power and scanning speed, as shown in Fig. 23. For these first layers, the experimental and computed values are generally consistent. Moreover, for a combination of laser power and scanning speeds with higher volumetric energy density Ev = 65.4 J.mm− 3 (P = 300 W and s = 0.9 m.s− 1, and P = 200 W and s = 0.6 m.s− 1), the simulation results are in reasonable agreement with experimental values for almost all the number of layers (Fig. 23a-b). However, when Ev = 32.7 J.mm− 3 (P = 300 W and s = 1.8 m.s− 1, and P = 100 W and s = 0.6 m.s− 1), the difference between the measured and the computed values increases, as can be seen in Fig. 23c-d. And this difference becomes even larger for the combinations of laser power and scanning speed, giving an Ev of 21.8 J.mm− 3 (P = 300 W and s = 2.7 m.s− 1, P = 200 W and s = 1.8 m.s− 1), shown in Fig. 23e-f. In many cases, the numerical values exceed the measured ones by more than double. The simulated results also captured the variation in stress values with the number of layers, showing a decrease followed by an increase in stress magnitude with an increase in n for all cases. Comparison between experimentally measured and computed values of longitudinal residual stresses (sxx) for different combinations of laser power (P) and scanning speed (s): (a) P = 300 W and s = 0.9 m.s− 1, (b) P = 200 W and s = 0.6 m.s− 1, (c) P = 300 W and s = 1.8 m.s− 1, (d) P = 100 W and s = 0.6 m.s− 1, (e) P = 200 W and s = 1.8 m.s− 1, (f) P = 300 W and s = 2.7 m.s− 1. To understand the discrepancies between the measured and simulated stress values, SEM micrographs of the samples were analyzed. Figure 24 shows SEM images of the surface of the samples produced with a laser power of 100 W and a scanning speed of 0.6 m.s⁻¹ and different numbers of layers (from 5 to 200, to comply with Fig. 23). It shows that the 200-layer sample, which has a very low measured stress value (about 47 MPa), exhibits the highest porosity (Fig. 24f), while the 5-layer sample - which has a much higher measured stress (about 388 MPa) - presents the lowest porosity (Fig. 24a). This correlation indicates that increased porosity might have contributed to a substantial drop in residual stresses in the prints. The porosity rate of the samples is not being taken into account in the simulation, which can explain the differences previously highlighted between the measured and computed stress values. It should be noted that for other combinations of laser power and scanning speed, samples with a low number of layers exhibit lower porosity than samples with a higher number of layers. SEM observations of the samples produced with P = 100 W, s = 0.6 m.s− 1 and a different number of layers (n): (a) n = 5, (b) n = 10, (c) n = 20, (d) n = 50, (e) n = 100, and (f) n = 200. Additional SEM investigations were carried out to check for different anomalies. In Fig. 25, tilted views of the 200-layer samples produced under different combinations of laser power and scanning speed, allowing both the top and side faces to be visualized, are compared. As shown in Fig. 25a, a laser power of 300 W and a scanning speed of 1.8 m.s− 1 (Ev = 32.7 J.mm⁻³) resulted in a consolidated material without any observable porosity. This sample is also the one that, as the sample produced at 300 W and 0.9 m.s− 1 (Ev = 65.4 J.mm⁻³), shows the best agreement between the measured and numerical values (approximately 400 MPa), as shown in Fig. 23c. On the contrary, the sample produced with a laser power of 100 W and a scanning speed of 0.6 m.s− 1 (Ev = 32.7 J.mm−³) resulted in unconsolidated material on the top surface normal to the build direction (Fig. 25b), likely due to a lower laser power that is not sufficient for full melting. Similarly, the sample produced with 300 W laser power at 2.7 m.s⁻¹ scanning speed (Ev = 21.8 J.mm⁻³) showed a bad surface quality with delamination defects on the side surface and regions of unconsolidated material as well (Fig. 25c). The occurrence of such delaminations and areas of lack of fusion/unconsolidated regions has been reported in other studies, at lower Ev71,72. These detrimental effects that result at low Ev values lead to lowering of the residual stresses. These two samples are those with the lowest measured stress value (around 50 MPa) and the largest deviation between the measured and calculated values (Fig. 23d and f), thus confirming the previous hypothesis. SEM observation of samples built with 200 layers: (a) P = 300 W, s = 1.8 m.s− 1 (Ev = 32.7 J.mm− 3), showing a well-consolidated material on the top surface; (b) P = 100 W, s = 0.6 m.s− 1 (Ev = 32.7 J.mm− 3), whose top surface is made of unconsolidated material; (c) P = 300 W, s = 2.7 m.s− 1 (Ev = 21.8 J.mm− 3), showing delamination on the side surface; (d) schematic representation of the sample indicating the directions and surfaces shown in the images. [In all cases, it can be noted that the side surfaces are very degraded by numerous unmelted particles]. Pores are a detrimental defect that can lead to crack initiation, as evidenced by the delamination highlighted in the specimen produced at low volumetric energy density (Ev). Increasing Ev contributes to reducing the viscosity and increasing the size of the melt bath. Furthermore, this improves Marangoni convection, leading to better spreading and melting of the melt bath, thus reducing porosity73. This was also well shown in Figs. 13a-b for the samples produced at 300 W and 0.9 m.s− 1 (Ev = 65.4 J.mm− 3), whose surface is well compacted and free of porosity, unlike samples produced at the same laser power but higher speed (thus lower Ev). Conclusions LPBF-printed Ti-6Al-4V samples were fabricated with different numbers of layers (from 1 to 200) using a zigzag strategy and various combinations of laser power and scanning speed, resulting in volumetric energy densities ranging from 21.7 to 65.4 J.mm− 3. The top surface integrity of the as-built samples was characterized by roughness and residual stress measurements. The main findings of this study are the following: A cumulative effect of surface defects was demonstrated, with a sharp increase in roughness amplitude and surface complexity up to a few dozen layers, followed by stabilization beyond. A cumulative effect of surface defects was demonstrated, with a sharp increase in roughness amplitude and surface complexity up to a few dozen layers, followed by stabilization beyond. Due to the unidirectional scanning strategy, a strong anisotropy of the surface texture was observed for the first layers, which quickly faded or even disappeared as additional layers were added due to the presence of numerous symmetrical defects (unmelted powders and porosities) and undulations on the surface. Due to the unidirectional scanning strategy, a strong anisotropy of the surface texture was observed for the first layers, which quickly faded or even disappeared as additional layers were added due to the presence of numerous symmetrical defects (unmelted powders and porosities) and undulations on the surface. The best surface quality was achieved with a volumetric energy density of 65.4 J.mm− 3, resulting from better scanning track overlap and improved melt pool stability. In contrast, low energy density promoted irregular surfaces, leading to the formation of porosity and undulations. The best surface quality was achieved with a volumetric energy density of 65.4 J.mm− 3, resulting from better scanning track overlap and improved melt pool stability. In contrast, low energy density promoted irregular surfaces, leading to the formation of porosity and undulations. An excess height of several hundred micrometres - or even more than one millimetre for 200 layers - was observed on samples of 50 layers and more, which could be linked to a lack of compactness due to incomplete fusion of the powders and to the irregularity of the surface, and likely to lead to geometric deviations from the specified dimensions. An excess height of several hundred micrometres - or even more than one millimetre for 200 layers - was observed on samples of 50 layers and more, which could be linked to a lack of compactness due to incomplete fusion of the powders and to the irregularity of the surface, and likely to lead to geometric deviations from the specified dimensions. Anisotropy of the residual stress fields was highlighted, with the main stress oriented along the scanning direction. Anisotropy of the residual stress fields was highlighted, with the main stress oriented along the scanning direction. Low-porosity samples showed good agreement between measured residual stresses and simulated stress values by a thermomechanical numerical model. However, the unconsolidated material led to residual stress releasing, explaining the gap between the experimental and numerical values. Low-porosity samples showed good agreement between measured residual stresses and simulated stress values by a thermomechanical numerical model. However, the unconsolidated material led to residual stress releasing, explaining the gap between the experimental and numerical values. Acknowledgements The authors would like to thank Pr. Jean-Michel Sprauel and Mr. Fadi Hermes from MRX for their substantial support in evaluating residual stresses. Abbreviations Laser powder bed fusion Finite element Focus variation microscopy Non-measured points Selective laser melting Scanning electron microscope Volumetric energy density Hatch spacing Number of layers Total flatness deviation Measured sample height Theoretical sample height Number of layers Arithmetic mean roughness Root mean square slope Root mean square (RMS) roughness Maximum profile peak height Skewness parameter Root mean square deviation of the profile Total amplitude of roughness Maximum profile valley depth Maximum amplitude of roughness Arithmetic mean height Rate of the average roughness evolution between two steps Root mean square gradient Interfacial area ratio Mean hills volume Peak density Texture aspect ratio Width of weld beads Apparent width of laser scanning track Dimensions of the elliptical crater Specific heat capacity Thermal conductivity Volumetric heat generation Heat transfer coefficient at the surface Residual stress along the scanning direction Residual stress perpendicular to the scanning direction Laser intensity Laser power Scanning speed Layer thickness Laser absorption coefficient Radial distance from the centre of the laser beam Laser spot size Elastic stiffness matrix Nodal displacement vector Nodal force vector due to the inherent strain Stiffness matrix Greek symbols Laser scan orientation θ Density Stefen-Boltzmann constant Emissivity Strain Size of the Gaussian filter Rotation angle Tilt angle Author contributions S.L. and M.D. carried out experiments, collected data, analysed the data wrote the initial draft version of manuscript. S.M., B.V. and N.L conducted simulations, plotted the results, wrote the draft version of the numerical part of the manuscript. S.L, M.D, K.M., A.H, and V.V. secured the funding and supervised overall work. All the authors contributed equally in editing and finalizing the final version of the draft. Data availability All the data necessary to support the results are provided in the manuscript. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement All the data necessary to support the results are provided in the manuscript.