Abstract
Residual stress and plastic strain in additive manufactured materials can exhibit significant microscopic variation at the powder scale, profoundly influencing the overall properties of printed components. This variation depends on processing parameters and stems from multiple factors, including differences in powder bed morphology, nonuniform thermostructural profiles, and interlayer fusion. In this research, we propose a powderresolved multilayer multiphysics simulation scheme tailored for porous materials through the process of selective laser sintering. This approach seamlessly integrates finite element method (FEM) based nonisothermal phasefield simulation with thermoelastoplastic simulation, incorporating temperature and phasedependent material properties. The outcome of this investigation includes a detailed depiction of the mesoscopic evolution of stress and plastic strain within a transient thermostructure, evaluated across a spectrum of beam power and scan speed parameters. Simulation results further reveal the underlying mechanisms. For instance, stress concentration primarily occurs at the necking region of partially melted particles and the junctions between different layers, resulting in the accumulation of plastic strain and residual stress, ultimately leading to structural distortion in the materials. Based on the simulation data, phenomenological relation regarding porosity/densification control by the beam energy input was examined along with the comparison to experimental results. Regression models were also proposed to describe the dependency of the residual stress and the plastic strain on the beam energy input.
Similar content being viewed by others
Introduction
In the recent decade, additive manufacturing (AM) emerged from a niche technology to a widely applicable means of production. Among a variety of AM techniques, powder bed fusion (PBF) stands out as a strong technique for producing intricate geometries while maintaining good structural integrity and superior material properties^{1,2,3,4,5}. In PBF, layers of powder get fused by a beam one after another to build up threedimensional (3D) geometries. Compared to conventional ones, this manufacturing process offers several advantages, such as great design flexibility, reduction in waste, enabling rapid tooling and optimizing production cycles.
Selective laser sintering (SLS) is one prominent example of PBF, which utilizes the tuned incident beam (mostly continuous laser scan or laser pulse) to bind the free powders layer by layer. Due to its wellcontrolled powder bed temperature compared to other PBF techniques like selective laser melting (SLM), where the significant melting phenomenon can be captured, SLS enables only partial melting of particles and produces samples with relatively high porosity^{6,7,8,9,10}. In this regard, SLS has great potential in applications that require complex geometries with high porosity. For instance, SLS is applicable in manufacturing porous components for medication applications, especially medical scaffolds, and artificial bones^{11,12,13}. It is feasible in manufacturing functional materials with a low processing temperature, such as ferromagnetic materials^{14,15,16,17,18}. By precisely controlling the geometry and structure of the printed material, it may also be possible to create the desired strain gradients and electric polarization necessary to generate the flexoelectric effect^{19,20,21}.
Residual stress has been a critical issue since it can affect mechanical properties, dimensional accuracy, corrosion resistance, crack growth resistance, and performance of AM parts^{22}. There are already investigations focusing on the spatial distribution of the residual stress within the AMbuilt parts, which were carried out by experimental measurements and theoretical estimations^{23,24,25,26}. Mercelis et al. explained the origins of residual stress in parts produced by AM. They experimentally measured the residual stress distribution in an SLS produced part and compared it with analytical and numerical solutions^{23}. Pant et al. employed a layerbylayer simulation by finite element method (FEM) to predict the residual stress and validated the model using measurements from neutron diffraction^{25}. Ibraheem et al. predicted the thermal profile and residual stress of SLS processed H13 tool steel using a FEM model. High porosity was observed in the range of 25–40%, yet a homogenized powder bed was utilized to simulate the thermal evolution and subsequently the residual stress^{27}.
The calculation of residual stress in literature is often performed in two sequential steps: first the transient temperature is simulated in the entire domain and then the thermal history is used to calculate the stress and strain evolution in thermomechanical simulations. The accuracy of these calculations critically relies on the quality of the thermal history. The temperature gradient mechanism (TGM) model and the cooldown stage (CDS) model are two commonly used models to explain the development of plastic strain and residual stress formation mechanisms^{23,28,29}. In the TGM model, the deformation of the material in the molten pool/fusion zone is restrained by surrounding materials during both the heating and cooling stages due to a large temperature gradient inside and outside the overheated region (where the onsite temperature is beyond the melting point). In the heating stage, the plastic strain is developed in the surrounding material due to an expansion of the heated material. As the heated material cools down, it starts to shrink, which is, however, counteracted by the plastically deformed surrounding material. Thus, a tensile residual stress develops. The CDS model, on the other hand, was proposed to elucidate the residual stress due to the layerwise features of the AM process and the AMmanufactured parts. This occurs because, in both the heating and cooling stages, the deformation of the upperlayers is restrained by the fused lowerlayers and/or substrate. Meanwhile, fused material in the lowerlayers will undergo a remelting and resolidification cycle. Both factors result in tensile residual stress in the top layer due to the shrinkage of the material during the cooling stage. These two models provide phenomenological aspects by employing idealized homogeneous layers for the understanding of the formation mechanism of residual stresses. They have been widely adopted in numerical analysis of residual stress at singlelayer scale models^{30,31,32,33} and partscale multilayer models^{34,35,36}.
In these aforementioned works, homogeneous layers are used for analyses of the thermal history and the mechanical response in an AMmanufactured part. However, due to the complex morphology of the powder bed on the powder scale and the resultant inhomogeneity of the local thermal properties, the thermal profile is subjected to a high level of nonuniformity as well. This implies heterogeneity of the thermal stress, the plastic deformation and the residual stress on the powder scale. In fact, mechanical properties of AMbuilt parts, such as fracture strength and hardening behavior, are significantly influenced by the local defects and local stress^{37,38}. The diversity in the powder bed structure and the packing density introduces thermal heterogeneity in the form of mesoscopic highgradient temperature profile and asynchronous onsite thermal history. Such variability leads to varying degrees of thermal expansion and, therefore, thermal stresses. Additionally, the presence of stochastic interparticle voids and lackoffusion pores contribute to evolving disparities in thermomechanical properties^{7,39,40}. The thermal gradient across the powder particles is very sharp as per the modeling results by Panwisawas et al.^{41}, which is in agreement with experimental findings^{42} and is also recaptured in our former numerical works^{7,39,40}. Furthermore, the layerwisely buildup process results in local interaction between the newly deposited and previously deposited layers with high surface roughness. Thus, the residual stress formation on the powder scale should be revisited for the SLS process.
For this purpose, the powder bed morphology, the heat transfer and the porous microstructure evolution during the SLS process should be primarily considered. One promising approach is the phasefield simulation. For instance, phasefield simulations of the AM process can reveal the inprocess microstructure evolution, providing insights into key features such as porosity, surface morphology, temperature profile, and geometry evolution^{7,43}. Based on nonisothermal phasefield multilayer simulation results of SLS processed porous structures, thermoelastic simulations and homogenization of the elastic properties have been investigated in our previous work^{39}. In combination with the singlelayer phasefield thermostructural simulations, the first powderresolved elastoplastic simulations of the residual stress in a SLSprocessed magnetic alloy have also been performed^{40}. Results show that during the cooling stage, the partially melted and interconnected particles result in plastic deformation due to the shrinkage of the fusion zone, but mostly at the necking region because of stress concentration in porous microstructures. By combining the thermostructural phasefield method and the thermomechanical calculations, this work demonstrates the promising capability of the multiphysics simulation scheme for the local residual stress analysis of AMbuilt materials on the powder scale. In the current article, we extend this multiphysics simulation scheme for elastoplastic multilayer SLS process, providing systematic simulations of the local plastic deformation and the residual stress. These allow us to reveal the influential aspects related to the printing process and the powder bed.
Results
In this work, the simulation workflow is arranged in three stages, as shown in Fig. 1a. The powders were first deposited into a prime simulation domain based on the discrete element method (DEM) under a given gravitational force. Then, nonisothermal phasefield simulations were performed for the coupled thermostructural evolution during the SLS process. Upon completion of a layer, the resulting microstructure is voxelized and reimported into the DEM program for the deposition of the next powder layer until reaching the desired layer number or height. Meanwhile, the subsequent thermoelastoplastic calculations were performed to investigate the development of plastic strain and stress under the quasistatic thermostructures by mapping the nodal values of the order parameters (OP, indicating the chronologicalspatial distribution of the phases) and temperature to a subdomain. It is worth noting that the thermoelastoplastic calculation of the next layer was restarted from the former one with continuous history of the residual stress and accumulated plastic strain, though the OPs and temperature were reiterated in the prime domain. As it is usually done in the literature, we assume thereby that heat transfer is only strongly coupled with microstructure evolution (driven by diffusion and underlying grain growth), but mechanics impose negligible effects on both heat transfer and microstructure evolution.
SLS processes with a constant beam spot diameter D_{L}, and various combinations of beam power P as well as scan speed v were simulated. The processing window P ∈ [15, 30] W and v ∈ [75, 150] mm s^{−1} was set in order to compare it with previous studies^{7,39}. D_{L} = D_{FWE2} = 200 μm (FWE2 represents the fullwidth at 1/e^{2}) was adopted as the nominal diameter of the beam spot, within which around 86.5% of the power is concentrated. The fullwidth at half maximum intensity (FWHM) then equals to 0.588D_{FWE2} = 117.6 μm, characterizing 50% power concentration within the spot. The powder bed was preheated with a preheating temperature T_{0} = 0.4T_{M} = 680 K, which was embodied by temperature initial condition (IC) and boundary condition (BC). Both powder and substrate are SS316L. The Prime simulation domain has a size of 250 × 500 × 400 μm^{3}, which contains the free space of 150 μm height for positioning powders and a substrate of 250 μm thickness. Intotal four layers of SLS processes were performed for each pair of processing parameters (i.e., P and v, hereinafter as P − v pair). Each newly deposited layer maintained the same powder size distribution and a volume fraction of 12% with respect to the free space of 250 × 500 × 150 μm^{3}. This guarantees each deposited layer sufficiently holds one to two particles in thickness, as illustrated in Fig. 1b. The powder size follows Gaussian distribution with a mean diameter of μ_{d} = 20 μm, a standard deviation ς_{d} = 5 μm, and a cutoff bandwidth d_{cut} = 10 μm. Fig. 1c presents the powder size distribution of the first layer (common for all P − v combinations). After each layer’s beam scan (i.e., the beam leaves the prime domain), the simulation continues for twice the duration of the scanning period to emulate a natural cooling stage. Details regarding the process as well as thermoelastoplastic modeling, simulation setup, and parameterization are explicitly given in the Methods section.
Thermostructural evolution during multilayer SLS
Figure 2a_{1}a_{4} present the transient thermostructural profiles for a single scan of P = 20W and v = 100 mm s^{−1} where the beam spot is consistently positioned along scan direction (SD) across various building layers. In the overheated region (T ≥ T_{M}), particles can undergo complete or partial melting. This prompts molten materials to flow from convex to concave points, making the fusion of particles possible and, therefore, forging the fusion zone. In regions where the temperature remains below the melting point (T < T_{M}), however, the temperature is still high enough to cause diffusion as the diffusivity grows exponentially w.r.t. temperature. This is evidenced by the necking phenomenon among neighboring particles. Temperature profiles demonstrate a strong dependence on the local morphology. Notably, the concentrated temperature isolines can be observed around the neck region among particles, indicating an increased temperature gradient (up to around 50–100K μm^{−1}, comparing to 1K μm^{−1} in the densified region). It is worth noting that such thermal inhomogeneity induced by stochastic transient morphology can hardly be resolved by the simulation works employing homogenized powder bed^{26,29,44}.
The heat dissipation in the powder bed also changes considerably as upperlayers are continuously built and processed. This can be noticed by a varying shape and significance of the overheated region across layers. As a comparison, the overheated region at 4^{th} layer (Fig. 2a_{1}) is greater in comparison to the initial layer (Fig. 2a_{4}). This can simply be reasoned by increasing porosity in fused lowerlayers that block the heat dissipation. At the initial layer, this dissipation is affected the least as there is no lowerlayer but substrate. We further probed the temperature history at surface points located at the center of the scan path, as shown in Fig. 2c_{1}, where the recurring peaks in every singlelayer SLS stage (shaded section) present the thermal cycle. The first peaks of distinctive cycles (probed on different surface points) nearly match. In an identical cycle (e.g., C1), the peaks gradually decrease as the SLS processes advance. Once the spot leaves the probing point, the temperature drop is rapid at first (mostly due to the heat conduction driven by the highgradient temperature profile) and then becomes moderate (due to heat convection and radiation) as the cooling stage begins, when the heat convection and radiation are effective. This steep temperature drop gradually disappears at points from fused lowerlayers as the upperlayers are continuously built, comparing the thermal cycles at point C1. Meanwhile, comparing the thermal cycles at points C1C4, the temperature drop at distinct points gradually coincides as the cooling stage proceeds.
The multilayer thermostructural evolution is also significantly affected by processing parameters, as the profiles of varying beam powers and scan speeds presented in Fig. 2b_{1}–b_{4}. Increasing the beam power and/or decreasing the scanning speed is observed to intensify heat accumulation at the beam spot, resulting in an enlargement of the overheated region. When v is held constant at 100 mm s^{−1}, decreasing P from 30W (Fig. 2b_{1}) to 15W (Fig. 2b_{2}) results in a less pronounced overheated region. On the other hand, maintaining a constant P = 20 W and increasing v from 75 mm s^{−1} (Fig. 2b_{3}) to 150 mm s^{−1} (Fig. 2b_{4}) leads to a reduction in the overheated region as well. It is also evident that increasing P and/or decreasing v produces less porosity in fused lowerlayers, which may further change the thermal conditions and improve the heat dissipation. Figure 2c_{2} presents the probed thermal cycle of point C1 under various P with v = 100 mm s^{−1} maintained. One can tell that the temperature of the case P = 30W quickly drops from a higher peak to the value coincided with one of the case P = 20 W at the end of scan duration, implying an improved heat dissipation. It should also be noted that surface point C1, located at the initial layer, experienced three peaks of temperature that are beyond T_{M} due to the extended depth of the fusion zone at higher beam power. In the Discussion section, we will continue the discussion regarding the relation between porosity and processing parameters.
Stress and plastic strain evolution during multilayer SLS process
To analyze the overall stress and plastic strain evolution during SLS, the average quantities (incl. von Mises stress σ_{e}, effective plastic strain p_{e}, and temperature T) in the powder bed are defined as
where \({\Omega }^{{\prime} }\) is the volume of the simulation domain without the substrate meshes, and ρ is the OP indicating the substance with ρ = 1. They are hereinafter termed as PBaveraged quantities. Figure 3a presents that \({\bar{\sigma }}_{e}^{p}\) develops as the SLS process advances on distinctive layers. When the beam spot enters the domain, accompanied by the appearance of the overheated region, \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\) begins to decrease while \({\bar{T}}^{{{{\rm{P}}}}}\) continues to rise to its peak. This is attributed to losing structural integrity caused by full/partial melting within the overheated region, resulting in zero stress as shown in Fig. 3b_{1}. The areas surrounding the overheated region also exhibit relatively low stress due to a significant reduction in stiffness at elevated temperatures. Stress accumulates faster around concave features, such as surface depressions and particle sintering necks, compared with the stress around convex features as well as unfused powders away from the fusion zone. This is due to localized temperature gradients. As cooling progresses, stress decreases in convex features and unfused powders, yet remains concentrated around concave features (see Fig. 3b_{2}−b_{3}). The stress at the end of each cooling stage (Fig. 3b_{3}−b_{6}) serves as the residual stress of corresponding processed layers. After the deposition of new layers, \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\) has an instant drop due to the addition of stressfree substances (powders), then a new cycle begins. It is worth noting that the peak values of the stress cycle (which are also residual stresses of each processed layer) are almost identical, implying that each singlelayer SLS generates nearly the same amount of residual stress by average in the powder bed. For upperlayers, σ_{e} experiences an additional reduction in the cooling stage, which may be attributed to the more delicate variations in the onsite stress components and will be discussed in Fig. 4.
The accumulated plastic strain p_{e}, on the other hand, presents an overall increasing tendency vs time during the singlelayer SLS and cooling stage, as shown in Fig. 3c. This distinguishes the plastic strain history from the stress cycle, which suffers a reduction during SLS due to loss of structural integrity. The continuous accumulation of p_{e} leads to a locally concentrated profile in the vicinity of the overheated region (see Fig. 3d_{2}−d_{3}). The high temperature gradient at the front and bottom of the overheated region during singlelayer SLS also results in the localized p_{e} surrounding the fusion zone (see Fig. 3d_{3}−d_{6}), where the existing high thermal stress locally activates the plastic deformation. A similar explanation applies to the concentrated p_{e} around pores and concave features such as sintering necks near the fusion zone. In contrast, unfused powders and the substrate away from the melt zone show minimal p_{e}, since the thermal stress at these locations is insufficient to induce material plastification due to lower local temperatures.
We probed the history of the stress components at six points, where L_{1}0L_{4}0 locates at the center of 1^{st} − 4^{th} layers, L_{1}1 locates at the boundary of the fusion zone, and L_{1}2 locates outside the fusion zone in the 1^{st} layer, as shown in the inset of Fig. 4. The results mainly show the difference of the stress field inside and outside the fusion zone. Since the temperature cools down from above the melting temperature to preheating temperature, the rise of the elasticity leads to a significant increase of σ_{e} on point L_{1}0 for temperature below 0.65T_{M}. As a comparison, the increament of the σ_{e} becomes smaller for the point L_{1}1 and the point L_{1}2, as shown in Fig. 4a_{1}−a_{3}. It can also be seen that σ_{e} for the mentioned three points (L_{1}0L_{1}2) during SLS of 2^{nd} and 3^{rd} layers are very similar when they are all outside the fusion zone. Similar results can be observed for the development of normal stress components σ_{ii} (i = x, y, z). The main differences in normal stress for the three points are mostly for the SLS process of the 1^{st} layer. σ_{ii} are tensile in the fusion zone while compressive outside the fusion zone. This is due to the thermal stress in the fusion zone being tensile while it is compressive outside the fusion zone. Curves for the 2^{nd} and 3^{rd} layers are similar because the three points are all outside the fusion zone. Shear stress components σ_{ij} (i, j = x, y, z; i ≠ j) remain small inside and outside the fusion zone. The variation of the shear stress is comparably more visible at the boundary of the fusion zone (L_{1}1), as shown in Fig. 4a_{2}. This is due to a large gradient of the thermal strain and material mechanical properties across the boundary of the fusion zone. With a larger variation of the shear stress components, variation of σ_{e} at the highest temperature during SLS process of each layer is also more visible. The saturation of the stress field can be observed after the 2^{nd} layer. It means that the upper layer has limited influence on the plastic deformation of previous layers, which experience lower temperatures and lower temperature gradients outside the fusion zone.
For points in the layerwise direction (L_{1}0 to L_{4}0 at the boundary of each layer), the results in Fig. 4b show a direct comparison between the development of the stress field of a point in the microstructure during SLS process of different layers. It is worth noting that the development of the σ_{e} at each point is very similar at the beginning. The saturation of the σ_{e} for each point can be observed in the layerwise direction when the points are outside the fusion zone. However, variation of the σ_{e} is much more pronounced when the overheated region is moving close to the point. This indicates that the point is located around the boundary of the fusion zone, which encounters a large variation of mechanical properties of the material w.r.t. the highgradient temperature and thus undergoes a large variation of the shear stress.
Powderresolved mesoscopic residual stress formation mechanism
For SLSprocessed porous microstructures, the residual stress is generally related to stress concentration under thermal loading in the microstructure. In Fig. 5a_{3}, a_{5}, the particles undergo a low degree of fusion (low level of overheating) due to relatively low beam power or high scan speed. The stress concentration mostly occurs at the necking region, leading to accumulated plastic strain and residual stress. In Fig. 5a_{1}, a_{4}, with a higher degree of fusion due to a relatively high beam power or low scan speed, accumulated plastic strain and residual stress tend to concentrate on both necking region and interlayer region (as indicated by the black lines in Fig. 5c, d). Note that the interlayer boundary has a high surface roughness which also causes stress concentrations. In Fig. 5a_{2}, with the combination of highest P and lowest v within the processing window, residual stress can be observed in the whole fusion zone, which makes it difficult to identify the residual stress concentration. Besides, accumulated plastic strain and residual stress can be observed on the boundary of the substrate and the 1^{st} layer, especially at high beam power. This is because a high beam power generates a much larger fusion zone, which even penetrates into the substrate. When the substrate with continuum material is fused, it generates a large plastic deformation and thus residual stress, which can be significantly larger than those in the porous microstructure, as shown in Fig. 5a_{2}. These results show that the residual stress is directly related to the degree of fusion and thus related to SLS processing parameters.
To gain a better understanding of the distribution of residual stress, we analyzed the stress state of a few representative positions in the microstructure, as shown in Fig. 5c, d. The former surfaces of fused layers are marked by black lines, and the bottom of the fused strut, a.k.a., the fusion zone boundary (FZB) of the 1^{st} layer, is indicated by white lines. Fig. 5c_{1}, d_{1} show that the residual stress and the accumulated plastic strain are solely concentrated in the necking region and interlayer region for the case with a low degree of fusion. In Fig. 5c_{2}, with the high degree of fusion (the smallest porosity achieved in the selected processing window), the residual stress and the accumulated plastic strain still tend to be solely concentrated in the interlayer region, yet less distinguishable compared to the former case. By comparing Lamé’s stress ellipsoids at six points on the boundaries of different layers, it is worth noting that the stress states close to the boundary between layers are very similar, e.g. the stress states of P_{2} and P_{3} are similar to that of P_{5} and P_{6}, no matter how high the degree of fusion is. It means that the residual stress is still formulated due to stress concentration at the interlayer region. On the other hand, the stress states of P_{1} and P_{4} at the top layer are quite different. This is because the residual stress of the top layer is directly determined by the thermal loading which depends on the surface morphology and morphologyinduced thermal inhomogeneity.
Based on the above observations, we propose a powderresolved mesoscopic residual stress formation mechanism, which is summarized in the schematic illustration in Fig. 6. The residual stress formulated in porous microstructures manufactured by a multilayer SLS process contains two primary contributions:

(i)
Residual stress is directly caused by the stress concentration at the necking region of partially melted particles under thermal loading. The partially melted particles are interconnected via necking regions, which is the weak link of the microstructure. Therefore, both the thermal expansion in the overheated region and the thermal contraction during the cooling stage cause severe plastic deformation at the necking region, which is one primary source of the accumulated plastic strain and residual stress.

(ii)
Residual stress due to interaction between the upper and lower/fused layers in the layerwisely buildup process. In the cooling stage, the shrinkage of the upperlayer after overheating results in tensile stress on itself and compressive stress on the lower/fused layer, which causes plastic deformation of the porous structure, especially at the interlayer region with stress concentration due to a high surface roughness, which is the other primary source of the accumulated plastic strain and residual stress, as indicated by the white lines in Fig. 6a.
The proposed mechanism is based on the detailed simulation results using the powderresolved thermomechanical model, which evidently demonstrates the formation and distribution of residual stress in the porous structure. The accumulated plastic strain results in structural distortion of the fused strut, which is schematically illustrated in Fig. 6a and demonstrated by the contour plot of deformation component u_{z} in Fig. 6b. Compared to the aforementioned TGM and CDS models proposed by Mercelis and Kruth^{23} correspondingly, which ignored the difference between parts manufactured by SLM and SLS, the partial melting as the main fusion mechanism plays an important role in the residual stress in the microstructure. There are some similarities, the concept of the TGM model also applies to the proposed mechanism because the temperature gradient indeed directly leads to residual stress. However, in the SLS process stage, since the powder bed is fully relaxed before fusion, particles provide very weak constrain on the fusion zone and thus have limited plastic deformation. The plastic deformation is majorly accumulated during the cooling stage due to thermal contraction and stress concentration at the necking region. For the CDS model, the proposed model also considered the shrinkage of the top layer resulting in a tensile residual stress in the top layer and a compressive residual stress in the previous layers. However, the top layer in the simulated SLS process has very complex boundaries with previous layers. These boundaries are the source of the stress concentration which leads to the dominant residual stress at the interlayer regions.
Based on the TGM model and the CDS model, Mercelis and Kruth^{23} also introduced a theoretical model to predict the relationship between residual stress and the part height. Similarly, we can use the proposed powderresolved model and mesoscopic residual stress formation mechanism to predict the dependence of the residual stress on the microstructure porosity, which is also a characteristic geometric parameter of a porous microstructure. Since the porous structure induces complex inhomogeneity of both material properties and thermal and mechanical loading, and we also need to consider the porosity is directly related to SLS processing parameters, it is more straightforward to propose phenomenological models based on our simulation results to predict the relationship between the residual stress and SLS processing parameters, as will be discussed in the next section.
Discussion
Controlling the porosity of processed sample through the tuning of the processing parameters plays a central role in tailoring the endup performance of the porous materials in AM, as many properties are determined by (or related to) porosity, including but not limited to mechanical strength, permeability, acoustic/optical absorptivity and various effective conductivities. We start the discussion with phenomenologically relating the porosity to the processing parameters (in this work P and v). In this work, a nominal porosity is defined as
with the substance OP ρ. Ω^{″} represents the volume of a postprocessed simulation domain with surface and unfused powders sufficiently removed (termed as “virtual polishing”). Figure 7a presents the local porosity that evaluated segmentwise along z (BD) and ydirection (perpendicular to SD) to help identify the representative domain with a width W_{R} and a height H_{R} for porosity calculation, while the complete length (L_{R} = 250 μm) along xdirection (SD) is included due to exhibited relatively minor variations in local porosity^{39}. We then conduct the virtual polishing on simulated multilayer SLSprocessed microstructures with H_{R} = 60 μm (containing the coordinate range z ∈ [ − 135, − 75] μm) and W_{R} = D_{FWHM}, and proceed calculating their porosity using Eq. (2) with the postprocessed domain size Ω^{″} = L_{R}H_{R}W_{R}. The resulting P − v map of porosity is in shown Fig. 7c. Selected microstructures illustrated in Fig. 7b_{2}−b_{4} for the cases varying P with constant v; Fig. 7b_{5}−b_{7} for the cases varying v with constant P. Figure 7b_{1} is the microstructure under a reference processing parameter (P = 20 W, v = 100 mm s^{−1}), while Fig. 7b_{8}, b_{9} are respectively the one with minimum and maximum porosity. It demonstrates that improved densification can be achieved by either increasing P or decreasing v, where smoothed surface morphology implies an enhanced partial melting. Porosity drops from 31% to 18% for the increase of P from 15 to 30 W (v = 100 mm s^{−1}) and from 29% to 21% for the decrease of v from 150 to 75 mm s^{−1} (P = 20 W). The presented tendencies imply a possible allometric relation φ ∝ P^{−m}v^{n} with positive indices m and n. On the other hand, combining beam power and scan speed as one characteristic quantity, P/v is widely adapted to define a specific energy density, notably the volumetric energy input as
where H is the powder layer thickness (here H = 60 μm) and W is the scan track width (here W takes D_{FWHM}). Evidently, φ exhibits an overall decrease tendency with rising U, meaning lower porosity can be achieved as higher specific energy input.
We then examined the simulation results with a proposed phenomenological relation following refs. ^{7,45}, read as
with a densification factor defined as
ϱ indicates the ratio between a reduced porosity w.r.t. the maximum porosity reduction in the chosen processing window, as φ_{0} and \({\varphi }_{\min }\) represent the initial and minimum achieved porosity, respectively. \({\varphi }_{\min }\) usually varies between 3% and 30% for metallic materials^{45}. In pursuit of uniformity, we have chosen \({\varphi }_{\min }=3 \%\) in this study, aligning with our previous singlelayer SLS simulation^{7}. φ_{0}, however, is difficult to be determined for multilayer SLS simulation since there is solely one or two particles in thickness for every new layer while the old layers have been fused. In this sense, we evaluate φ_{0} by three routes:

(i)
Assuming that the porosity of the powder bed region, which is away from the beam spot (i.e., without significant densification), is the initial one of the powder bed, φ_{0} can be read as the converged value from segmentwise porosity evaluation along ydirection (Fig. 7a_{2}), which is φ_{0} = 27.8–33.6%.

(ii)
As the particle size distribution in this work is assumed to be Gaussiantype, φ_{0} is then estimated by statistic randomclosepacking model of spherical particle as \({\varphi }_{0}=0.3660.0257\left({\varsigma }_{d}/{\mu }_{d}\right)\) with μ_{d} and ς_{d} the mean and standard deviation of the particle diameter^{44,46}. Taking μ_{d} = 20 μm and ς_{d} = 5 μm, it can be calculated as φ_{0} = 36.0%.

(iii)
We also piled multiple densificationfree powder stacks using DEM method with the same domain volume fraction of the particles (48.0%) as the overall one of the multilayer SLS simulation, in which each layer deposits powders of 12% domain volume fraction. The measured nominal porosity is thereby regarded as the initial one, as φ_{0} = 40.0–40.5%.
φ_{0} from route (i) directly reflects the porosity of onsite unfused powders, but the influences from the potential necked particles due to thermal processes cannot be well eliminated. φ_{0} from route (ii) is based on the statistics of randomclosedpacked particles, which is rather ideal compared to the practical powder bed created by powder spreading. Since route (iii) creates particle stacks via simulating the deposition in powder spreading, the calculated φ_{0} may be still inflated as the morphological variability of the deposition surface is missing. Considering all of these factors, we have selected φ_{0} = 36.0%, which stands at the midpoint among all the assessed values.
In Fig. 7d, linear regression between \(\ln (1\varrho )\), calculated from simulated φ, and U is presented. For comparison, regressions on data by singlelayer SLS simulation and experiment are also illustrated^{7,45}. The regressed line together with the 95% confidence interval (CI_{9}_{5%}) of the multilayer SLS simulations is right in between those of the singlelayer SLS simulation and experiments. As the coefficient K_{ϱ} is related to the material and the size distribution of the powders, the multilayer result K_{ϱ} = 0.016 ± 0.001 mm^{3} J^{−1} demonstrates an improved coherence with the experimental one K_{ϱ} = 0.013 mm^{3} J^{−1} compared the singlelayer one K_{ϱ} = 0.019 mm^{3} J^{−1}, which can suffer from inflated porosity mostly due to insufficient volume microstructure for porosity calculation. It is also worth noting that this relation between \(\ln (1\varrho )\) and U is examined on the microstructures both with and without fusion zone. To characterize the formation of the fusion zone, a spatialtemporal indicator ξ was added in the system to emulate the phenomenological fusion of the strut during the multilayer SLS process following our former work^{40,44}, which is initialized as zero and irreversibly transitions as one once T ≥ T_{M}. In Fig. 7b_{1}–b_{9} the fusion zones are denoted for selected P and v. The map can be thereby divided into characteristic regions where a continuous/discontinuous fusion zone or no fusion zone is formed. Increasing U from 17.01 to 22.68 J mm^{−3}, the transition from microstructures without fusion zone to with continuous fusion zone also implies the switch of densification mechanism from pure solidstate sintering to partialmelting–induced fusion (similar to the liquidstate sintering). This may explain the change of the tendency of \(\ln (1\varrho )\) when U < 22.68 J mm^{−3} in Fig. 7d.
Taking the PBaveraged residual von Mises stress \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\) and effective plastic strain \({\bar{p}}_e^{{{{\rm{P}}}}}\) (defined in Eq. (1)) at the end of the simulations, Fig. 8 presents the distributions of \({\bar{\sigma }}_{e}^{\rm{P}}\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\) w.r.t. the P and v in the chosen processing window. Generally, the rise in \({\bar{\sigma }}_{e}^{\rm{P}}\) corresponds to an increase in the specific energy input U (i.e., increase P with maintained v or increase v with maintained P), resulting in further concentrated residual stress around concave features (surface depressions and particle sintering necks) and across layers, as already presented in Fig. 5a, b. Meanwhile, as the locally concentrated residual stresses and plastic strain are evident in the processed powder bed, understanding how these local quantities are distributed and how much effect the extreme values (here the local maximums) impose on the evaluated averages \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\) and \({\bar{p}}_e^{{{{\rm{P}}}}}\) becomes essential. In that sense, we sampled all nodal points in the processed powder bed and performed statistics on σ_{e} and p_{e} of these points. The probability distribution of local σ_{e} and p_{e} are presented in Supplementary Fig. 1a. The third quartile (Q3) is employed to characterize the concentration of the local quantities. Then, P − v regions with Q3(σ_{e}) and Q3(p_{e}) correspondingly beyond the referencing values, which are PBaverages \({\sigma }_{{{{\rm{e0}}}}}^{{{{\rm{P}}}}}=247\,{{{\rm{MPa}}}}\) and \({p}_{{{{\rm{e0}}}}}^{{{{\rm{P}}}}}=0.012\) from the case P = 20 W and v = 100 mm s^{−1}, are denoted in Fig. 8a, b (the complete P − v maps for Q3(σ_{e}) and Q3(p_{e}) are shown in Supplementary Fig. 2). These can be statistically interpreted as that the processed powder bed with P and v selected within the regions have at least 25% local substances obtaining σ_{e} and p_{e} beyond the referencing \({\sigma }_{{{{\rm{e0}}}}}^{{{{\rm{P}}}}}\) and \({p}_{{{{\rm{e0}}}}}^{{{{\rm{P}}}}}\), respectively, while the rest of local substances have adequate σ_{e} and p_{e} up to these referencing values. In other words, the SLS processes with P and v selected from these regions tend to have processed powder bed with relatively higher localized residual stresses and accumulated plastic strains.
To understand the relationship between \({\bar{\sigma }}_{\rm{e}}^{\rm{P}}\) and U, we interpreted \({\bar{\sigma }}_{\rm{e}}^{\rm{P}}\) as the stored mechanical energy density after the multilayer SLS processes, which can be regarded as the residue of the energy density imported via the beaminduced thermal effect in the powder bed after all types of inprocess dissipation. In this regard, the nonlinear regression analysis of an energy conversion law \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{p}}}}}={\sigma }_{\infty }^{{{{\rm{P}}}}}\{1\exp [\frac{1}{{U}_{\sigma }^{{{{\rm{P}}}}}}(U{U}_{{{{\rm{th}}}}}^{{{{\rm{P}}}}})]\}\) was conducted on the simulation data. The parameters \({U}_{{{{\rm{th}}}}}^{{{{\rm{P}}}}}\) and \({\sigma }_{\infty }^{{{{\rm{P}}}}}\) adopt the physical meanings as the volumetric energy input at stressfree state and the saturated stress at infinity energy input, respectively, and \({\sigma }_{\infty }^{{{{\rm{P}}}}}/{U}_{\sigma }^{{{{\rm{P}}}}}\) is the increasing rate (slope) of \({\bar{\sigma }}_{\rm{e}}^{\rm{P}}\) vs. U at stressfree state, as shown in the inset of Fig. 8c. The result in Fig. 8c presents a high correlation (R^{2} = 98.41%) between simulated \({\bar{\sigma }}_{\rm{e}}^{\rm{P}}\) vs. U with narrow confidence interval, demonstrating the applicability of proposed energy conversion law in predicting the residual stress in the powder bed.
Unlike the residual stress, the effective plastic strain p_{e} is a measure of cumulative plastic deformation at any given moment during the process and lacks a clear physical picture of its relationship with volumetric energy input U. Therefore, the nonlinear regression analysis of an allometric scaling law \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}={C}_{p}^{{{{\rm{P}}}}}{(U)}^{{I}_{p}^{{{{\rm{P}}}}}}\) with parameters \({C}_{p}^{{{{\rm{P}}}}}\) and \({I}_{p}^{{{{\rm{P}}}}}\) was conducted to phenomenologically relate \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\) to U, as shown in Fig. 8d. The analysis gives a relatively low correlation coefficient R^{2} = 87.4% of the \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}(U)\) relation compared with one of \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{p}}}}}(U)\), with an expanded confidence interval at high U range. Regressed \({I}_{p}^{{{{\rm{P}}}}}=1.17\pm 0.13\) indicates an almost linear scalability of accumulated plastic strain in the processed powder bed on energy input via beam scan. It is worth noting that the proposed scaling law can be challenged as U appears to not being able to uniquely identify \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\), in other words, U of certain value can correspond to multiple p_{e} values, as also depicted in Fig. 8d. Nonetheless, this scaling law can find it feasible in estimating strength of inprocess plastification of the microstructure at the given specific energy input.
Since both \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\) and \({\bar{\sigma }}_{\rm{e}}^{\rm{P}}\) consider a complete powder bed with unfused particles, we further defined two average quantities that only take the residual stress and effective plastic strain inside the fused strut into account, denoted respectively \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\), which are calculated as
with the fusion zone indicator ξ. Figure 9 presents the distribution of \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) w.r.t. P and v in the chosen processing window. Contrasting with those shown in Fig. 8, a notable distinction in the maps presented in Fig. 9c, d is the emergence of regions where the selected P and v fail to form continuous fused strut. Meanwhile, the profiles of \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) receive significant influences from the size of the strut. Comparing Fig. 9a_{1} with Fig. 9a_{2}, a_{3} and Fig. 9a_{4}, a_{5}, the increase of \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) follows the direction of increasing U as well, which simultaneously leads to the enlargement of the fused strut. As depicted in Figs. 5c and 9a, concentrated residual stress is primarily found within the upperlayers of the strut. When the strut size is enlarged, it encompasses a larger volume with elevated maximum σ_{e} and highσ_{e} region, resulting in an increased \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) with the rise in U. \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) exhibits similar tendency w.r.t. P and v as \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\), with an increase in U resulting in an increase in \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\). Nonetheless, regions with highly accumulated p_{e} locates at the bottom of each layer’s fusion zone, as illustrated in Figs. 5d and 9b. At high U, such accumulation intensifies, especially at the bottom of the strut (which is also the bottom of the initial layer’s fusion zone). Simultaneously, enlarged depth of a fusion zone indicates an extended remelting in the fused lowerlayers, which removes some accumulated p_{e} by the process or former layer within the strut, as shown in Fig. 6a. It also concentrates the highp_{e} region further to the bottom of the strut, notably the profile in Fig. 9b_{4}. Eventually, the rise of \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) w.r.t. U is relatively less “rapid” than the one of \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\), evident in slightly sparser contours in Fig. 9d compared to Fig. 9c. Statistics of the local σ_{e} and p_{e} in the fused strut are also performed (Supplementary Fig. 1b), and the P − v regions characterizing Q3(σ_{e}) and Q3(p_{e}) beyond the referencing values, which are strut averages \({\sigma }_{{{{\rm{e0}}}}}^{{{{\rm{S}}}}}=215\,{{{\rm{MPa}}}}\) and \({p}_{{{{\rm{e0}}}}}^{{{{\rm{S}}}}}=0.016\) from the same case P = 20 W and v = 100 mm s^{−1}, are denoted in the Fig. 9c,d (the complete P − v maps for Q3(σ_{e}) and Q3(p_{e}) are shown in Supplementary Fig. 3). As many local points outside the fusion zone are excluded from the statistics, extended regions of \({{{\rm{Q3}}}}({\sigma }_{{{{\rm{e}}}}})\, > \,{\sigma }_{{{{\rm{e0}}}}}^{{{{\rm{S}}}}}\) and \({{{\rm{Q3}}}}({p}_{{{{\rm{e}}}}})\, > \,{p}_{{{{\rm{e0}}}}}^{{{{\rm{S}}}}}\) are evident, demonstrating that more P − v combinations can lead to the fused strut containing at least 25% local σ_{e} and p_{e} are beyond the averages \({\sigma }_{{{{\rm{e0}}}}}^{{{{\rm{S}}}}}\) and \({p}_{{{{\rm{e0}}}}}^{{{{\rm{S}}}}}\) of the same chosen reference case. One can also conclude that most formed continuous struts more or less have localized residual stresses and accumulated plastic strains, based on the observation of narrow unmarked P − v regions between one with the discontinuous strut and one with Q3 of local quantities beyond the chosen references. In this sense, secondary thermal treatment should be positively considered to anneal the residual stresses inside the SLSprocessed struts.
Nonlinear regression analyses were also conducted on \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) employing the proposed energy conversion law for residual stress and scaling law for effective plastic strain. Results are correspondingly presented in Fig. 9e, f. Notably, comparing to those of \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{p}}}}}(U)\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}(U)\), the correlation of \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\) present decline, respectively, with enlarged confidence interval at both low and high U ranges. This attributes to the removal of the influences from the unfused particles. Moreover, for \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\), regressed threshold energy input presents a negative value as \({U}_{{{{\rm{th}}}}}^{{{{\rm{S}}}}}=11.11\pm 17.43\,{{{\rm{J}}}}\,{{{{\rm{mm}}}}}^{3}\), indicating a required energy output to achieve stressfree state. This is explainable as the σ_{e} already exists in a justformed strut. In other words, primary SLS process (i.e., the SLS without subsequent thermal postprocessing to release residual stress) cannot achieve samples with stressfree strut. Regressed saturated stress \({\sigma }_{\infty }^{{{{\rm{S}}}}}=260.05\pm 24.5\,{{{\rm{MPa}}}}\) also presents decline comparing to the regressed one on PBaveraged data. For \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\), a reduced regressed index of the scaling law \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\), i.e., \({I}_{p}^{{{{\rm{S}}}}}=0.30\pm 0.06\) in Fig. 9f, demonstrates a sublinear scalability of \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\) compared with one of \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}(U)\), which is almost linear. It also reflects a reduced growth rate of \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) at high U, and the underlying intensified remelting, which removes some accumulated p_{e} within the strut while further concentrates p_{e} around the strut bottom, and the comparably faster increase in strut size shall be the reason. Nonetheless, information conveyed by \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\) is more relevant for practical application, as unfused particles shall be removed during postprocessing of a practical SLS. Further examination and validation of the proposed laws for residual stress and plastic strain w.r.t. specific energy input are expected in the future numerical and experimental studies.
In summary, we proposed a powderresolved multilayer simulation scheme for producing porous materials using SLS in this work, combing FEMbased nonisothermal phasefield simulation and thermoelastoplastic simulation with temperaturedependent material properties. This work has presented the mesoscopic evolution of stress and plastic strain on a transient thermostructure under various beam power (P) and scan speed (v). Processproperty relationships between porosity, residual stress and effective plastic strain and the volumetric energy input (U ∝ P/v) have also been demonstrated and discussed. The following conclusions can be drawn from the present work:

(i)
We proposed in this work a powderresolved mesoscopic residual stress formation mechanism for porous materials manufactured by the SLS process, which leads to the structural distortion appeared in fused strut. It was demonstrated with simulation results that the stress concentration at the necking region of the partially melted particles and interlayer region between different layers provide dominant accumulated plastic strain and residual stress in the porous material.

(ii)
Based on the proposed residual stress formation mechanism, we examined the proposed phenomenological relation between the porosity (densification) and the volumetric energy input U. Regression analysis on the resulting porosity from multilayer SLS simulations suggested an improved coherence with the experimental data, as the regressed densification coefficient K_{ϱ} = 0.016 ± 0.001 mm^{3} J^{−1} in this work is compared to the experimental K_{ϱ} = 0.013 mm^{3} J^{−1} and the one from our former singlelayer simulation K_{ϱ} = 0.019 mm^{3} J^{−1}.

(iii)
Two types of average quantities, namely PBaveraged (\({\bar{\sigma }}_{\rm{e}}^{\rm{P}}\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\)) and strutaveraged ones (\({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\)), were defined to characterize the residual stress and plastic strain within the powder bed and fused strut, correspondingly. The relationships between these quantities and volumetric energy input (U) are unveiled by conducting nonlinear regression analyses. The average residual stress (\({\bar{\sigma }}_{\rm{e}}^{\rm{P}}\) and \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\)) relates to U by the energy conversion law, while the average effective plastic strain (\({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\)) relates to U by the allometric scaling law.

(iv)
Attributing to the removal of influences from unfused particles, the correlation of the relations \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\) present drops compared with one of \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}(U)\) and \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}(U)\), respectively. Saturation behavior is observed on both \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}(U)\) and \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\), while the linear scalability in \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{P}}}}}(U)\) degenerates into sublinear one in \({\bar{\sigma }}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}(U)\), demonstrating a reduced growth rate of \({\bar{p}}_{{{{\rm{e}}}}}^{{{{\rm{S}}}}}\) at high U.
Despite the feasibility of the multilayer simulation scheme in recapitulating mesoscopic formation of porosity, residual stress and plastic strain under given processing parameters; and the proposed mechanism in explaining the structural distortion of SLSproduced samples, several points should be further examined and discussed in future works:

(i)
The present work omits the consideration of chronologicalspatial distribution of the thermalelastoplastic properties among polycrystals, as the properties such as elasticity and crystal plasticity vary from grain to grain with distinct orientations.

(ii)
The present findings are examined at relatively low specific energy input U and, correspondingly, low generated residual stress and accumulated plastic strain. Pore formation is also limited to lackoffusion mechanism. It is anticipated to conduct further simulations with relatively high energy input, where the mechanisms like keyholing coexist with high residual stress and plastic strain. Extension of the proposed mechanism into the highU range together with the following examination and validation are also expected.
Methods
Nonisothermal phasefield model for multilayer SLS processes
The nonisothermal phasefield model employed in this work to simulate the multilayer is based on our former works^{7,39,40,47}. For the sake of completeness, in this subsection we summarize the essentials of the employed model. For clarity, vectors are denoted by italic symbols with accented arrow (e.g., \(\overrightarrow{v}\)). Bold symbols (e.g., K, M, σ, ε) are for 2^{nd}order tensors, and blackboard bold symbols (e.g., \({\mathbb{C}}\)) are for 4^{th}order ones.
The model adopts both conserved and nonconserved OPs for representing the microstructure evolution of polycrystalline material. The conserved OP ρ indicates the substance, including the unfused and partially melted regions, while the nonconserved OPs {η_{i}}, i = 1, 2, … distinguish particles with different crystallographic orientations. To guarantee that the orientation fields ({η_{i}}) get valued only in the substance (ρ = 1), a numerical constraint ∑_{i}η_{i} + (1 − ρ) = 1 is imposed to the simulation domain. This constraint also applies to the substrate where ρ = 1 always holds.
The thermostructural evolution is governed by
where the local free energy density is formulated as
Here f(T, ρ, {η_{i}}) contains multiple local minima, representing the thermodynamic stability of the states, i.e., substance, atmosphere/pore and grains with distinct orientations, with the coefficients \({\underline{W}}_{{{{\rm{sf}}}}}(T)\) and \({\underline{W}}_{{{{\rm{gb}}}}}(T)\) related to the barrier heights among minima, varying with temperature^{7,40}. The heat contribution f_{ht} thereby manifests the stability of states by shifting the minima according to the local temperature field. c_{r} is a relative specific heat landscape formulated by the volumetric specific heats in the substance c_{ss} and pore c_{at}, i.e., c_{r} = h_{ss}c_{ss} + h_{at}c_{at} with corresponding interpolation functions h_{ss}, h_{at} indicating the substance (including solid and liquid) and the atmosphere/pore, respectively. When the material reached melting point T_{M}, the extra contribution due to the latent heat \({{{{\mathcal{L}}}}}_{{{{\rm{M}}}}}\) is mapped by the interpolation function h_{M}, which approaches unity when T → T_{M}^{7,40}. It is worth noting that \({\underline{W}}_{{{{\rm{sf}}}}}(T)\), \({\underline{\kappa }}_{{{{\rm{sf}}}}}(T)\), \({\underline{W}}_{{{{\rm{gb}}}}}(T)\), and \({\underline{\kappa }}_{{{{\rm{gb}}}}}(T)\) inherit their temperature dependencies from the surface energy γ_{sf}(T) and grain boundary energy γ_{gb}(T), respectively (see Boundary conditions and model parameters subsection). At the equilibrium, they can be related to γ_{sf}(T), γ_{gb}(T), and a diffuseinterface width of the grain boundary l_{gb}, which have been explained in our former work^{7,40}.
The CahnHilliard mobility employed in Eq. (6) equation adopts the anisotropic form. Contributions considered in this work contain not only the mass transfer through substance, atmosphere, surface and grain boundary, but also the diffusion enhancement due to possible partial melting^{40,47,48}, i.e.
where D_{path} is the effective diffusivity of the mass transport via path = ss, at, sf, gb, and h_{ss}, h_{at}, h_{sf} and h_{gb} are again interpolation functions to indicate substance, atmosphere/pore, surface and grain boundary, respectively, which obtain unity only in the corresponding regions^{40,47,48}. Mass transport along surface and grain boundary is also regulated by corresponding projection tensor T_{gb} and T_{sf}. It is worth noting that the partial melting contribution M_{M} is treated as an enhanced surface diffusion when T → T_{M} due to the assumption of a limited melting phenomenon around the surface of particles. In this sense, formulations in Eqs. (6) and (10) disregard the contributions from melt flow dynamics as well as from intercoupling effects between mass and heat transfer, i.e., Soret and Dufour effects^{47}. The isotropic AllenCahn mobility is explicitly formulated by the the temperaturedependent GB mobility G_{gb}, GB energy γ_{gb} and gradient coefficient \({\underline{\kappa }}_{\eta }\) as^{49,50}
The phasedependent thermal conductivity tensor takes the continuity of the thermal flux along both the normal and tangential directions of the surface into account, and is formulated as^{51,52}
where K_{ss} and K_{at} are the thermal conductivity of the substance and pore/atmosphere. N_{sf} is the normal tensor of the surface^{40,48,51}. It is worth noting that radiation contribution via pore/atmosphere is considered in K_{at} as K_{at} = K_{0} + 4T^{3}σ_{B}ℓ_{rad}/3 with K_{0} the thermal conductivity of the Argon gas and σ_{B} the StefanBoltzmann constant. ℓ_{rad} is the effective radiation path between particles, which usually takes the average diameter of the powders^{53}.
The beamincited thermal effect is equivalently treated as a volumetric heat source with its distribution along the depth direction formulated in a radiation penetration fashion, as the powder bed is regarded as an effective homogenized optical medium, i.e.
in which the inplane Gaussian distribution p_{xy} with a moving center \({\overrightarrow{r}}_{\!O}(\overrightarrow{v},t)\). P is the beam power and \(\overrightarrow{v}\) is the scan velocity with its magnitude \(v= \overrightarrow{v}\) as the scan speed, which are two major processing parameters in this work. The absorptivity profile function along depth da/dz which is calculated based on refs. ^{7,54}. It is obvious that when one ignores the effects of those latent heat induced by microstructure evolution (i.e. the evolution of the pore/substance as well as unique grains), Eq. (8) can be degenerated to the conventional Fourier’s equation for heat conduction with an internal heat source.
Elastoplastic model for thermomechanical analysis
The transient temperature field and substance field ρ, resulting from the nonisothermal phasefield simulations, are then imported into the quasistatic elastoplastic simulation as the thermal load and the phase indicator for the interpolation of mechanical properties. The linear momentum equation reads
where σ is the 2^{nd}order stress tensor. Taking the VoigtTaylor interpolation scheme (VTS), where the total strain is assumed to be identical among pores/atmosphere and substance, i.e. ε = ε_{ss} = ε_{at}^{55,56,57}. In this regard, the stress can be eventually formulated by the linear constitutive equation
where the 4^{th}order elastic tensor is interpolated from the substance one \({{\mathbb{C}}}_{{{{\rm{ss}}}}}\) and the pores/atmosphere one \({{\mathbb{C}}}_{{{{\rm{at}}}}}\), i.e.,
In this work, \({{\mathbb{C}}}_{{{{\rm{ss}}}}}\) is calculated from the temperaturedependent Youngs’ modulus E(T) and poison ratio ν(T), while \({{\mathbb{C}}}_{{{{\rm{at}}}}}\) is assigned with a sufficiently small value.
The thermal eigenstrain ε^{th} is calculated using interpolated coefficient of thermal expansion, i.e.,
Here I is the 2^{nd}order identity tensor. Meanwhile, for plastic strain ε_{pl} and isotropic hardening model with the von Mises yield criterion is employed. The yield condition is determined as
with
where σ_{e} is the von Mises stress. s is the deviatoric stress, and σ_{y} is the isotropic yield stress when no plastic strain is present. The isotropic plastic modulus \({\mathsf{H}}\) can be calculated from the isotropic tangent modulus E_{t} and the Young’s modulus E as \({\mathsf{H}}=E{E}_{{{{\rm{t}}}}}/(E{E}_{{{{\rm{t}}}}})\). p_{e} is the effective (accumulated) plastic strain, which is integrated implicitly by the plastic strain increment ε^{pl} employing radial return method and will be elaborated in Numerical implementation and parallel computing subsection.
Boundary conditions and model parameters
Together with the governing equations shown in Eqs. (6)–(8), the following BCs are also employed for the process simulations
with the convectivity \(\underline{h}\), StefanBoltzmann constant σ_{B}, the hemispherical emissivity ε, and the preheating (environmental) temperature T_{0}. Γ_{T} and Γ_{B} are correspondingly the top and bottom boundaries of the simulation domain, and Γ_{S} is the set of all surrounded boundaries. Γ_{B} ∪ Γ_{S} ∪ Γ_{T} = Γ, as the schematic shown in the inset of Fig. 1b. \(\overrightarrow{n}\) is the normal vector of the boundary. Eq. (18) physically shows the close condition for the mass transfer, restricting no net mass exchange with the environment. Eq. (19) shows the heat convection and radiation, which are only allowed via the pore/atmosphere at the boundary (masked by h_{at}). Eq. (20) emulates a semiinfinite heat reservoir under the bottom of the substrate with constant temperature T_{0}, consistently draining heat from the simulation domain and reducing its temperature back to T_{0}.
As summarized in Nonisothermal phasefield model for multilayer SLS processes subsection, this simulation requests following parameters/properties: the thermodynamic parameters \(\,{\underline{W}}_{{{{\rm{sf}}}}}\), \(\,{\underline{W}}_{{{{\rm{gb}}}}}\), \(\,{\underline{\kappa }}_{\rho }\), and \(\,{\underline{\kappa }}_{\eta }\); the kinetic properties (diffusivities and GB mobility) D_{path} (path = ss, at, sf, gb) and G_{gb}; and the thermal properties K_{ss} and K_{at}. Among them, \(\,{\underline{W}}_{{{{\rm{sf}}}}}\), \(\,{\underline{W}}_{{{{\rm{gb}}}}}\), \(\,{\underline{\kappa }}_{\rho }\), and \(\,{\underline{\kappa }}_{\eta }\) are parameterized by the temperaturedependent surface as well as GB energies, and a nominal diffuse interface width l_{gb} as
with normalized tendencies τ_{sf}(T) and τ_{gb}(T) that reach unity at T_{M}. \(\,{\underline{W}}_{{{{\rm{sf}}}}}\,\approx \,{W}_{{{{\rm{sf}}}}}{\tau }_{{{{\rm{sf}}}}}(T)\), \(\,{\underline{W}}_{{{{\rm{gb}}}}}={W}_{{{{\rm{gb}}}}}{\tau }_{{{{\rm{gb}}}}}(T)\), \(\,{\underline{\kappa }}_{\rho }\,\approx \,{\kappa }_{\rho }{\tau }_{{{{\rm{sf}}}}}(T)\), and \(\,{\underline{\kappa }}_{\eta }={\kappa }_{\eta }{\tau }_{{{{\rm{gb}}}}}(T)\). Constants W_{sf}, W_{gb}, κ_{ρ}, κ_{η} also satisfy a constraint (W_{sf} + W_{gb})/κ_{ρ} = 6W_{gb}/κ_{η} derived from the coherent diffuseinterface profile at equilibrium (Supplementary Note 1 of ref. ^{7}). These parameter/properties are collectively shown in Table 1.
For the thermoelastoplastic calculations, a 250 × 250 × 250 μm^{3} subdomain was selected from the center of the prime domain to eliminate the boundary effects (Fig. 1a). The above momentum balance is subjected to the following rigid support BC as
restricting the displacement \(\overrightarrow{u}\) along the normal direction of all boundaries except the top (Γ_{T}), which is traction free. As summarized in Elastoplastic model for thermomechanical analysis subsection, temperaturedependent E, ν, α, E_{t}, and σ_{y} are as listed in Table 2, where piecewise linear interpolation was employed to implement their temperature dependence.
Numerical implementation and parallel computing
The theoretical models are numerically implemented via FEM within the program NIsoS, developed by authors based on MOOSE framework (Idaho National Laboratory, ID, USA)^{58}. 8node hexahedron Lagrangian elements are chosen to mesh the geometry. A transient solver with preconditioned JacobianFree NewtonKrylov method (PJFNK) and backward Euler algorithm was employed for both problems.
For nonisothermal phasefield simulations, the CahnHilliard equation in Eq. (6) was solved in a split way. The constraint of the order parameters was fulfilled using the penalty method. To reduce computation costs, hadaptive meshing and timestepping schemes are used. The additive Schwarz method (ASM) preconditioner with incomplete LUdecomposition subpreconditioner was also employed for parallel computation of the vast linear system, seeking the balance between memory consumption per core and computation speed^{59}. Due to the usage of adaptive meshes, the computational costs vary from case to case. The peak DOF number is on the order of 10,000,000 for both the nonlinear system and the auxiliary system. The peak computational consumption is on the order of 10,000 corehour. More details about the FEM implementation are shown in the Supplementary Note 4 of ref. ^{7}.
For thermoelastoplastic simulations, a static mesh was utilized to avoid the hanging nodes generated from hadaptive meshing scheme. In that sense, the transient fields T and ρ of each calculation step were unidirectionally mapped from the nonisothermal phasefield results (with hadaptive meshes) into the static meshes, assuming a weak coupling between thermostructural and mechanical problems in this work. This is achieved by the MOOSEembedded SolutionUserObject class and associated functions. The parallel algebraic multigrid preconditioner BoomerAMG was utilized, where the EisenstatWalker (EW) method was employed for determining linear system convergence. It is worth noting that a vibrating residual of nonlinear iterations would show without the employment of EW method for this work. The DOF number of each simulation is on the order of 1,000,000 for the nonlinear system and 10,000,000 for the auxiliary system. The computational consumption is approximately 1,000 CPU corehour.
A modified radial return method was employed to calculate the integral of the plasticity as well as to determine the yield condition during the process with the temperaturedependent elastoplastic properties at any time t with the time increment Δt. This method employs the trial stress σ^{⋆} calculated assuming an elastic new strain increment Δε^{⋆}
where the elasticity tensor is obtained under the current temperature field T_{t}. Once the trial stress state is outside the yield condition in Eq. (17), i.e., the plastic flow exists, the stress is then projected onto the closet point of the expanded yield surface with the normal direction determined as \({{{{\bf{n}}}}}_{{{{\rm{y}}}}}=3{{{{\bf{s}}}}}^{\star }/2{{\sigma }_{{{{\rm{e}}}}}}^{\star }\) with the von Mises \({{\sigma }_{{{{\rm{e}}}}}}^{\star }\) and deviatoric trial stress s^{⋆}^{60,61}. Meanwhile, assuming isotropic linear hardening under T_{t} of every timestep, the amount of the effective plastic strain increment Δp for returning the stress state back to the yield surface is calculated in an iterative fashion adopting Newton’s method
where p_{t} and Δp_{t} are updated as p_{(t+Δt)} and Δp_{(t+Δt)} at the end of the timestep (t + Δt). G(T_{t}) and \({\mathsf{H}}({T}_{t})\) are shear and isotropic plastic modulus calculated under the local temperature at the current timestep (T_{t}). Here G(T_{t}) = E(T_{t})/[2 + 2ν(T_{t})] with E(T_{t}) and ν(T_{t}) correspondingly the temperaturedependent Young’s modulus and Poisson ratio. Knowing that the plastic strain increment Δε^{pl} = Δpn_{y} following the normality hypothesis of plasticity^{60}, the updated stress and plastic strain at the end of timestep are thereby obtained by
in which the vanishing of the stress as well as the plastic strain beyond a stressfree temperature T_{A} (in this work T_{A} = T_{M}) is also considered.
Eqs. (23)(27) are sequentially executed on every timestep and update the quantities under T_{t}. It is worth noting that the linearinterpolated \({\mathsf{H}}\) is implemented for both p and Tdependence
with
where \({\mathsf{f}}({\hat{T}}_{i},{p}_{t})\) is a piecewise function with the grids \({\hat{T}}_{i}\) (i = 1, 2, . . . ) and T_{i} is inside the section bound by \({\hat{T}}_{i}\) and \({\hat{T}}_{i+1}\), i.e., \({T}_{t}\in ({\hat{T}}_{i},{\hat{T}}_{i+1}]\). Considering reduction of the nonlinearity, the pindependent \({\mathsf{H}}(T)\) was practically employed in the simulations, as formulated in Eq. (24).
Data availability
The authors declare that the data supporting the findings of this study are available within the paper. The temperaturedependent parameters, thermodynamic database, simulation results, supplementary data and utilities are cured in the online dataset (https://doi.org/10.5281/zenodo.10940625).
Code availability
Source codes of MOOSEbased application NIsoS and related utilities are available and can be accessed via the online repository bitbucket.org/mfm_tuda/nisos.git.
References
Gu, D. et al. Materialstructureperformance integrated lasermetal additive manufacturing. Science 372, eabg1487 (2021).
Ladani, L. & Sadeghilaridjani, M. Review of powder bed fusion additive manufacturing for metals. Metals 11, 1391 (2021).
Zhang, Y. et al. Additive manufacturing of metallic materials: a review. J. Mater. Eng. Perform. 27, 1–13 (2017).
Panwisawas, C., Tang, Y. T. & Reed, R. C. Metal 3d printing as a disruptive technology for superalloys. Nat. Commun. 11, 1–4 (2020).
Körner, C., Markl, M. & Koepf, J. A. Modeling and simulation of microstructure evolution for additive manufacturing of metals: A critical review. Metall. Mater. Trans. A 51, 4970–4983 (2020).
Kim, W. R. et al. Fabrication of porous pure titanium via selective laser melting under lowenergydensity process conditions. Mater. Design 195, 109035 (2020).
Yang, Y., Ragnvaldsen, O., Bai, Y., Yi, M. & Xu, B.X. 3d nonisothermal phasefield simulation of microstructure evolution during selective laser sintering. npj Comput. Mater. 5, 1–12 (2019).
Gu, D., Meiners, W., Wissenbach, K. & Poprawe, R. Laser additive manufacturing of metallic components: Materials, processes and mechanisms. Int. Mater. Rev. 57, 133–164 (2012).
Li, J.Q., Fan, T.H., Taniguchi, T. & Zhang, B. Phasefield modeling on laser melting of a metallic powder. Int. J. Heat Mass Transfer 117, 412–424 (2018).
Li, R., Liu, J., Shi, Y., Du, M. & Xie, Z. 316l stainless steel with gradient porosity fabricated by selective laser melting. J. Mater. Eng. Perform. 19, 666–671 (2009).
Shuai, C. et al. Construction of an electric microenvironment in piezoelectric scaffolds fabricated by selective laser sintering. Ceram. Int. 45, 20234–20242 (2019).
Duan, B. et al. Threedimensional nanocomposite scaffolds fabricated via selective laser sintering for bone tissue engineering. Acta Biomater. 6, 4495–4505 (2010).
Eshraghi, S. & Das, S. Mechanical and microstructural properties of polycaprolactone scaffolds with onedimensional, twodimensional, and threedimensional orthogonally oriented porous architectures produced by selective laser sintering. Acta Biomater. 6, 2467–2476 (2010).
Mapley, M., Pauls, J. P., Tansley, G., Busch, A. & Gregory, S. D. Selective laser sintering of bonded magnets from flake and spherical powders. Scr. Mater. 172, 154–158 (2019).
Wendhausen, P., Ahrens, C., Baldissera, A., Pavez, P. & Mascheroni, J. Additive manufacturing of bonded ndfeb, process parameters evaluation on magnetic properties. In 2017 IEEE International Magnetics Conference (INTERMAG), 1–1. IEEE (IEEE, 2017).
Sing, S. L. et al. Direct selective laser sintering and melting of ceramics: A review. Rapid Prototyping J. 23, 611–623 (2017).
Jhong, K. J. & Lee, W. H. Fabricating soft maganetic composite by using selective laser sintering. In 2016 IEEE International Conference on Industrial Technology (ICIT), 1115–1118. IEEE (IEEE, 2016).
Huang, J., Yung, K. C. & Ang, D. T. C. Magnetic properties of smco5 alloy fabricated by laser sintering. J. Mater. Sci. Mater. Electron. 30, 11282–11290 (2019).
Zhang, B. et al. Flexoelectricity on the photovoltaic and pyroelectric effect and ferroelectric memory of 3dprinted BaTiO3/PVDF nanocomposite. Nano Energy 104, 107897 (2022).
Zhang, M., Yan, D., Wang, J. & Shao, L.H. Ultrahigh flexoelectric effect of 3d interconnected porous polymers: Modelling and verification. J. Mech. Phys. Solids 151, 104396 (2021).
Yan, D., Wang, J., Xiang, J., Xing, Y. & Shao, L.H. A flexoelectricityenabled ultrahigh piezoelectric effect of a polymeric composite foam as a straingradient electric generator. Sci. Adv. 9 (2023).
Chen, S.g et al. Review on residual stresses in metal additive manufacturing: Formation mechanisms, parameter dependencies, prediction and control approaches. J. Mater. Res. Technol. 17, 2950–2974 (2022).
Mercelis, P. & Kruth, J. Residual stresses in selective laser sintering and selective laser melting. Rapid Prototyping J. 12, 254–265 (2006).
Roberts, I., Wang, C., Kibble, K., Stanford, M. & Mynors, D. Numerical and Experimental Studies on the Laser Melting of Steel Plate Surfaces, 535–538 (Springer London, 2010).
Pant, P. et al. A simplified layerbylayer model for prediction of residual stress distribution in additively manufactured parts. Metals 11, 861 (2021).
Mukherjee, T., Zhang, W. & DebRoy, T. An improved prediction of residual stresses and distortion in additive manufacturing. Comput. Mater. Sci. 126, 360–372 (2017).
Ibraheem, A. K., Derby, B. & Withers, P. J. Thermal and residual stress modelling of the selective laser sintering process. MRS Proc. 758, LL1–8 (2002).
Simson, T., Emmel, A., Dwars, A. & Böhm, J. Residual stress measurements on aisi 316l samples manufactured by selective laser melting. Addit. Manuf. 17, 183–189 (2017).
Takezawa, A., Guo, H., Kobayashi, R., Chen, Q. & To, A. C. Simultaneous optimization of hatching orientations and lattice density distribution for residual warpage reduction in laser powder bed fusion considering layerwise residual stress stacking. Addit. Manuf. 60, 103194 (2022).
Yi, M., Xu, B.X. & Gutfleisch, O. Computational study on microstructure evolution and magnetic property of laser additively manufactured magnetic materials. Comput. Mech. 64, 917–935 (2019).
Hussein, A., Hao, L., Yan, C. & Everson, R. Finite element simulation of the temperature and stress fields in single layers built withoutsupport in selective laser melting. Mater. Des. (19802015) 52, 638–647 (2013).
Parry, L., Ashcroft, I. & Wildman, R. Understanding the effect of laser scan strategy on residual stress in selective laser melting through thermomechanical simulation. Addit. Manuf. 12, 1–15 (2016).
Vastola, G., Zhang, G., Pei, Q. & Zhang, Y.W. Controlling of residual stress in additive manufacturing of ti6al4v by finite element modeling. Addit. Manuf. 12, 231–239 (2016).
Chiumenti, M. et al. Numerical modelling and experimental validation in selective laser melting. Addit. Manuf. 18, 171–185 (2017).
Ganeriwala, R. et al. Evaluation of a thermomechanical model for prediction of residual stress during laser powder bed fusion of ti6al4v. Addit. Manuf. 27, 489–502 (2019).
Carraturo, M. et al. Modeling and experimental validation of an immersed thermomechanical partscale analysis for laser powder bed fusion processes. Addit. Manuf. 36, 101498 (2020).
Chen, W. et al. Microscale residual stresses in additively manufactured stainless steel. Nat. Commun. 10, 4338 (2019).
Bartlett, J. L. & Li, X. An overview of residual stresses in metal powder bed fusion. Addit. Manuf. 27, 131–149 (2019).
Zhou, X. et al. 3dmultilayer simulation of microstructure and mechanical properties of porous materials by selective sintering. GAMMMitt. 44, e202100017 (2021).
Yang, Y., Oyedeji, T. D., Zhou, X., Albe, K. & Xu, B.X. Tailoring magnetic hysteresis of additive manufactured feni permalloy via multiphysicsmultiscale simulations of processproperty relationships. npj Comput. Mater. 9, 103 (2023).
Panwisawas, C. et al. On the role of thermal fluid dynamics into the evolution of porosity during selective laser melting. Scr. Mater. 105, 14–17 (2015).
Arısoy, Y. M., Criales, L. E. & Özel, T. Modeling and simulation of thermal field and solidification in laser powder bed fusion of nickel alloy in625. Opt. Laser Technol. 109, 278–292 (2019).
Zhang, Z., Wang, Y., Ge, P. & Wu, T. A review on modelling and simulation of laser additive manufacturing: Heat transfer, microstructure evolutions and mechanical properties. Coatings 12, 1277 (2022).
Yang, Y. et al. Validated dimensionless scaling law for melt pool width in laser powder bed fusion. J. Mater. Process. Tech. 299, 117316 (2022).
Simchi, A. Direct laser sintering of metal powders: Mechanism, kinetics and microstructural features. Mater. Sci. Eng. A 428, 148–158 (2006).
Desmond, K. W. & Weeks, E. R. Influence of particle size distribution on random close packing of spheres. Phys. Rev. E 90, 022204 (2014).
Yang, Y., Oyedeji, T. D., Kühn, P. & Xu, B.X. Investigation on temperaturegradientdriven effects in unconventional sintering via nonisothermal phasefield simulation. Scr. Mater. 186, 152–157 (2020).
Oyedeji, T. D., Yang, Y., Egger, H. & Xu, B.X. Variational quantitative phasefield modeling of nonisothermal sintering process. Phys. Rev. E 108, 025301 (2023).
Moelans, N., Blanpain, B. & Wollants, P. Quantitative analysis of grain boundary properties in a generalized phase field model for grain growth in anisotropic systems. Phys. Rev. B 78, 024113 (2008).
Turnbull, D. Theory of grain boundary migration rates. JOM 3, 661–665 (1951).
Yang, Y. et al. A diffuseinterface model of anisotropic interface thermal conductivity and its application in thermal homogenization of composites. Scr. Mater. 212, 114537 (2022).
Nicoli, M., Plapp, M. & Henry, H. Tensorial mobilities for accurate solution of transport problems in models with diffuse interfaces. Phys. Rev. E 84, 046707 (2011).
Denlinger, E. R., Jagdale, V., Srinivasan, G., ElWardany, T. & Michaleris, P. Thermal modeling of inconel 718 processed with powder bed fusion and experimental validation using in situ measurements. Addit. Manuf. 11, 7–15 (2016).
Gusarov, A. V., Yadroitsev, I., Bertrand, P. & Smurov, I. Model of radiation and heat transfer in laserpowder interaction zone at selective laser melting. J. Heat Transfer 131, 072101 (2009).
Voigt, W. Ueber die beziehung zwischen den beiden elasticitätsconstanten isotroper körper. Ann. Phys. (Berlin) 274, 573–587 (1889).
Schneider, D. et al. Phasefield elasticity model based on mechanical jump conditions. Comput. Mech. 55, 887–901 (2015).
Durga, A., Wollants, P. & Moelans, N. A quantitative phasefield model for twophase elastically inhomogeneous systems. Comput. Mater. Sci. 99, 81–95 (2015).
Permann, C. J. et al. MOOSE: Enabling massively parallel multiphysics simulation. SoftwareX 11, 100430 (2020).
Balay, S.et al. Petsc Users Manual (rev. 3.11) (Argonne National Laboratory, 2019).
Dunne, F. & Petrinic, N.Introduction to Computational Plasticity (Oxford University Press, 2005).
Simo, J. C. & Hughes, T. J.Computational Inelasticity, vol. 7 of Interdisciplinary Applied Mathematics Ser. (Springer Science & Business Media, New York, NY, 2006).
Price, A., Holl, H. & Greenough, A. The surface energy and self diffusion coefficient of solid iron above 1350 c. Acta Metall. 12, 49–58 (1964).
Blakely, J. & Mykura, H. Studies of vacuum annealed iron surfaces. Acta Metall. 11, 399–404 (1963).
Mead, H. W. & Birchenall, C. E. Selfdiffusion of iron in austenite. JOM 8, 1336–1339 (1956).
Liu, F., Zhang, Q., Zhou, W., Zhao, J. & Chen, J. Micro scale 3d fem simulation on thermal evolution within the porous structure in selective laser sintering. J. Mater. Process. Tech. 212, 2058–2065 (2012).
Hoshino, T., Mito, K., Nagashima, A. & Miyata, M. Determination of the thermal conductivity of argon and nitrogen over a wide temperature range through data evaluation and shocktube experiments. Int. J. Thermophys. 7, 647–662 (1986).
Chase Jr, M. W. Nistjanaf thermochemical tables. J. Phys. Chem. Ref. Data, Monograph 9 (1998).
Di Schino, A., Kenny, J. M. & Abbruzzese, G. Analysis of the recrystallization and grain growth processes in aisi 316 stainless steel. J. Mater. Sci. 37, 5291–5298 (2002).
Mills, K. C. Recommended Values of Thermophysical Properties for Selected Commercial Alloys (Woodhead, Cambridge, 2002), 1. publ. edn.
Li, Y. et al. Modeling temperature and residual stress fields in selective laser melting. Int. J. Mech. Sci. 136, 24–35 (2018).
Acknowledgements
We acknowledge the financial support of German Research Foundation (DFG) in the framework of the Collaborative Research Centre Transregio 270 (CRCTRR 270, project number 405553726, subprojects A06 and A07), the Research Training Groups 2561 (GRK 2561, project number 413956820, subproject A4), the Priority Program 2122 (SPP 2122, project number 493889809). X.Z. acknowledges the support from the National Natural Science Foundation of China (project number 12302231), Sichuan Science and Technology Program (project number 2023NSFSC0910), and China Postdoctoral Science Foundation (project number 2023M732433). We also greatly appreciate the access to the Lichtenberg II HighPerformance Computer (HPC) and the technique supports from the HHLR, Technische Universität Darmstadt. The computating time on the HPC is granted by the NHR4CES Resource Allocation Board under the project “special00007”. Y.Y. and S.B. also greatly thanks Dr. B. Lin for helping with the setup of layerwise powder deposition, and P. Kühn and J. Ma for their extensive help with language and grammar checks.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Contributions
Conceptualization: B.X.X., X.Z. and Y.Y.; methodology: X.Z., Y.Y., J.S. and B.X.X.; software: Y.Y. and X.Z.; investigation: Y.Y., S.B., and N.F.; formal analysis: Y.Y., S.B., and N.F.; resources, X.Z., J.S. and B.X.X.; data curation: Y.Y.; writingoriginal draft preparation: Y.Y., S.B., N.F., X.Z. and B.X.X.; writingreview and editing: Y.Y., S.B., N.F., X.Z., J.S. and B.X.X.; visualization: Y.Y. and S.B.; supervision: B.X.X.; consultation and discussion: J.S.; funding acquisition: B.X.X. All authors have read and agreed to the published version of the manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Yang, Y., Bharech, S., Finger, N. et al. Elastoplastic residual stress analysis of selective laser sintered porous materials based on 3Dmultilayer thermostructural phasefield simulations. npj Comput Mater 10, 117 (2024). https://doi.org/10.1038/s41524024012965
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41524024012965