Microstructure evolution in laser powder bed fusion-built Fe-Mn-Si shape memory alloy

Sample preparation

The Fe and Si powders used in the LPBF fabrication of Fe-Mn-Si alloy were gas atomised and provided by TLS Technik, Germany, while the Mn powder was from Merelex Corp, USA. Both the Fe and Si powders had a purity of > 99 wt.%, and the purity of Mn was > 98 wt.%, as estimated using the Malvern Panalytical Epsilon ED X-ray fluorescence spectroscopy (XRF), Supplementary Table 1. Figure 1A-C shows the scanning electron microscope (SEM) micrographs of the Fe, Mn, and Si powder. Their particle size and cumulative size distributions were measured using the Malvern Mastersizer 3000 and are shown in Figure 1D-F, respectively. A nominal powder composition of Fe-30Mn-6Si (wt.%) was homogenised for 4 h using the Turbula® T2F 3D mixer and then used as powder precursor. Meanwhile, the nominal concentration of Fe-30Mn-6Si reference as-cast alloy was prepared using the arc-melting technique from Fe, Mn, and Si high purity (> 99.9%) chips from Sigma-Aldrich. The arc-melted product was subsequently hot-rolled at ~800 °C and then homogenised at 1100 °C for 14 h in an argon-purged furnace. Homogenisation was performed by loading the sample at room temperature, heating it at 5 °C/min to 1100 °C, and followed by furnace cooling. The resulting sample is referred to hereafter as reference as-cast alloy and its properties were treated as a reference in the following investigations.

Figure 1. SEM micrographs of (A) Fe, (B) Mn, and (C) Si powders. Their corresponding particle size frequency and cumulative distributions are shown in (D-F), respectively.

The LPBF fabrication was carried out using the Mlab Cusing 200R from Concept Laser GmbH equipped with 200 W Yb:YAG fibre laser and the print chamber atmosphere was maintained up to 0.2 vol.% O₂ using a high purity Ar gas. Only freshly homogenised powder was used, and all printed parts were built on a sandblasted stainless steel substrate. In identifying the optimum LBPF parameters, a 125 mm³ cube model was prepared using the Materialise Magics v24 software. The influence of both laser power and laser scan speed on the density of the built part was investigated by varying the laser power (P) from 100 W to 175 W and the laser scan speed (ν) from 400 mm/s to 600 mm/s while keeping a constant laser hatch spacing, layer thickness, and scan strategy at 45 μm, 50 μm, and island scan strategy, respectively. The scan strategy is unique to Concept Laser^[32], where each island was maintained at 5 × 5 mm² and was scanned by the laser in one direction. The laser scan direction was rotated by 90° between the neighbour islands, and finally, this whole pattern was rotated by 45° in the subsequent layer. A laser re-scan strategy was also included. This was done by scanning the solidified layer again at a varying percentage of laser power (0%, 50%, and 100%) that was applied in the first scan, laser scan speed from 400 mm/s to 600 mm/s, and a similar scan strategy to increase the laser linear energy density (LED). LED is a simplified energy parameter defined as the P/ν ratio and was considered when the layer thickness and laser space hatching were unchanged^[33-35]. Table 1 summarises the parameters that were investigated.

LPBF product quality assessment

The density of the LPBF built parts was measured by applying the Archimedes method and using the Mettler Toledo XS105 balance with a density kit. All surfaces of the samples were ground down to 1200 SiC paper and then dried. Measurements were done on three replicates. The measured density was then divided by the theoretical density (7.408 g/cm³) and reported as relative density. In addition, the surface of the LPBF-built parts along the build direction was viewed under the Hitachi TM4000Plus bench-top SEM coupled with a Bruker X-Flash 630Hc EDS detector to further evaluate the product quality.

Microstructure characterisation

The crystallography and the phase volume fraction of both the reference alloy and the LPBF fabricated alloy were evaluated using the PANalytical Empyrean with a Co target (λ = 1.79 Å) and a scan range from 40 to 130° 2θ at a step size of 0.02°. The volume fraction of the phases in the samples was computed by applying a Rietveld refinement^[36,37] using the HighScore Plus v5.1 Suite^[38,39]. Refinement parameters such as the expected profile R-value, profile R-value, weighted profile R-value, goodness of fit, Bragg R-value, and the difference plot between the experimental and calculated XRD pattern were closely monitored during the refinement process. Also, refinement was done at least three times to verify the results, and the γ-austenite and ε-martensite phase volume fractions were subsequently validated using the electron backscattered diffraction (EBSD) technique [Supplementary Figure 1].

The EBSD was carried out to characterise the microstructure, i.e., grain morphology, crystallographic texture, and grain characteristics, of the reference alloy and built products. The sample surfaces for EBSD analysis were final polished up to OPU finish and then ion milled using the Hitachi IM4000 at 30° and 6 kV for 1 h to remove any polishing artefacts. The Zeiss Auriga Crossbeam Field-emission SEM equipped with a NordlysF detector for EBSD and an Oxford Instruments X-Max 20 mm² silicon drift detector for EDS was used for the simultaneous SEM-EDS/EBSD analysis. The EDS and EBSD data were gathered using the AZtec 3.3 and then analysed using the AZtecFlex software suite. For EBSD analysis, a 500 × 500 μm² area was scanned at a step size of 1 μm and only the results with at least 90% hit rate were analysed. Data cleaning was done by removing the wild spikes, and then using the “zero solutions removal” tool from level 1 up to level 4 while preserving any defects, i.e., cracks and pores, on the cleaned dataset. The grain size and shape analysis were subsequently done on the computed prior austenite grains^[40] using the Shoji-Nishiyama orientation relationship^[41] to reveal the likely austenite grains.

Simulation of melt pool thermal properties

Simulation of the 3D thermal profile during LPBF was conducted using the COMSOL™ Multiphysics software suite, in which a simplified finite element analysis (FEA) thermal model was developed. A tetrahedral mesh was used for the model geometry with a minimum and maximum mesh size of 2 μm and 16 μm, respectively. A 1 × 6 × 1 mm³ smooth flat plate model with no powder was used to provide a heat sink effect sufficient for simulating the single line scan. Laser irradiation heat input is modelled as^[42,43],

where Q is volumetric heat flux (W/m²), P is laser power, A is laser absorption coefficient, R is laser beam radius, δ is laser penetration depth^[44,45], |z| is z-coordinate absolute value, and x and y are laser x- and y-coordinates, respectively. The theoretical density was used as the material density for the simulation. Specific heat capacity and thermal conductivity at room temperature were taken from literature for the similar material Fe-28Mn-6Si-5Cr and are 544.2 J/(Kg∙K) and 8.37 W/(mK), respectively^[46,47]. A single-track scan was first performed on a polished material surface of similar composition. Then, in the numerical simulation, the coordination of x and y scan speeds, as well as P and R, were the same as those used for the single-track scan. Determination of penetration depth δ and absorption coefficient A was conducted via an iterative process by matching the simulated melt pool with the observed melt pool [Supplementary Figure 2]. Initial build platform and surrounding gas temperature were taken as 30 °C, surface emissivity was 0.3, and convective heat transfer coefficient was 10 W/m²∙K.

Hardness test

The resistance to localised plastic deformation of the LPBF built and reference alloy was measured using the Struers DuraScan hardness machine following a standard procedure^[48]. The samples were resin mounted and then the surface was OPU polished. At least 10 indentations were made on each sample.

LPBF product quality

The interaction of the powder bed and the laser system in the LPBF fabrication of Fe-30Mn-6Si was assessed by looking into the influence of relative density as a function of LED, as shown in Figure 2A. The P/ν ratio is the LED of the laser system where a high P value and a low ν value translate to a large amount of LED, and this is similar to what is being used in conventional fusion welding techniques^[49]. The lowest LED of 0.17 J/mm returned the lowest density at 95%, and the density was found to increase linearly with LED until ~0.30 J/mm. The relative density levelled at mostly above 99%, where the 0.44 J/mm LED recorded the highest value of 99.9%.

Figure 2. (A) The relationship between relative density and LED, and the representative SEM micrographs at (B) low (0.17 J/mm), (C) mid (0.25 /mm), and (D) high (0.88 J/mm) LEDs. Arrows in black highlight the defects in the built parts, the build direction is from bottom to top, and the error bars in (A) represent the standard error of the mean.

SEM micrographs of all samples built at different LBPF are analysed and their representative at low, mid, and high LEDs are shown in Figure 2B-D, respectively. These micrographs represent the LEDs labelled in Figure 2A. Pores of over 250 μm are seen at the low LED of 0.17 J/mm, and their morphologies are irregularly shaped, as seen in Figure 2B. At the mid LED of 0.25 J/mm, the large pores are not apparent, but chemical inhomogeneity is noted by the difference in contrast in the backscattered electron (BSE) micrograph and they are highlighted by the black arrows in Figure 2C. The high LED of 0.88 J/mm has no observed chemical segregation; defects like cracks and spherical pores are however noted, and they are marked by the black arrows in Figure 2D. Densification of LPBF fabricated parts is directly associated with the reduction of defects at high LED, and this is shown by the SEM micrographs in Figure 2B-D. Nevertheless, the high relative density and relatively few defects in the high LED above 0.44 J/mm, i.e., the lack of large irregularly shaped pores [Figure 2D and Supplementary Figure 3], make these parameters ideal for the LPBF fabrication of Fe-30Mn-6Si from homogeneously blended powder.

The recommended LED for Fe-30Mn-6Si alloy (≥ 0.44 J/mm) is higher than the suggested for pure Fe (≥ 0.33J/mm)^[50,51], 316 L (0.21 to 0.30 J/mm)^[52,53] and 304 L (0.14 J/mm)^[54] stainless steels that were fabricated from pre-alloyed powders. Much higher energy is needed for the homogenised powder than for pre-alloyed powder, as noted in Al-Si^[55] and FeCoCrNi^[56], because more energy and a slow melt pool solidification rate are needed for the melting and alloying of the homogenised powder. The recommended energy for the LPBF of Fe-30Mn-6Si alloy is, however, less than of the Fe-17Mn-5Si-10Cr-4Ni alloy (0.53 J/mm), and the difference may be due to the high melting requirement of Cr in the latter.

Chemical composition

LPBF uses a high energy laser for melting and alloying of the metallic powders. Such energy may evaporate volatile components and alter the final composition of the alloy. An EDS analysis of the major elements Fe, Mn, and Si as a function of LED in Fe-30Mn-6Si LPBF alloy is shown in Figure 3A-C, respectively. As the LED is increased, the Fe content in the alloy is shown to increase, Mn is slightly decreased, and Si remains almost the same. The Mn element has the least melting point [Supplementary Table 2], high vapour pressure^[57,58], and therefore more volatile than the other precursors^[22,28]. Thus, its evaporation is expected during LPBF, resulting in a decrease in Mn concentration. Meanwhile, the slight increase in Fe likely influenced the relative density value. Fe had the highest density among the raw materials and the increase of its concentration likewise increased the alloy density. Therefore, careful selection of processing parameters and the adjustment of Mn concentration are necessary for the LPBF processing of homogenised powders.

Figure 3. Chemical composition of (A) Fe, (B) Mn, and (C) Si in LPBF alloy as a function of LED.

Microstructure

The influence of LPBF processing parameters on the microstructure was investigated using the EBSD micrographs of the LPBF-built alloy surface parallel to the building direction. The LEDs at 0.25 J/mm and 0.44 J/mm represent the low (100 W) and high (175 W) laser power, 0.29 J/mm and 0.44 J/mm for the low (400 mm/s) and high (600 mm/s) scan speed, and 0.44 J/mm and 0.88 J/mm for the 0% (175W, 400 mm/s) and 100% (175W, 400 mm/s and then 175 W, 400 mm/s) re-scan strategy, respectively. Figure 4A, D, G, and J show the prior austenite grains^[40,41] EBSD IPF map of the surface parallel to the build direction that seems to respond to the changes in the LPBF processing parameters. In Figure 4C, F, I, and L, the grain size distribution and the area-weighted average are also shown, and their corresponding aspect ratios are presented in Supplementary Figure 4. At a scan speed of 400 mm/s and a low laser power of 100 W (0.25 J/mm), the grain size was fine and nearly equiaxed [Figure 4A] with a size of 64 μm and an aspect ratio of 2.03. This changed to a coarse and columnar microstructure for the laser power of 175 W (0.44 J/mm),Figure 4G. Such an increase in laser power generated a grain size of almost three times and elongated the grains by ~90% compared to that of the low laser power. The microstructure was further modified for a constant laser power of 175 W when the scan speed was increased from 400 mm/s to 600 mm/s. The subsequent grain size was rather coarse (105 μm), and the grains were nearly equiaxed (aspect ratio = 1.97),Figure 4D and F. The columnar grains generated at high LED (0.44 J/mm) are seemingly retained when the already solidified layer was re-scanned at 100% (175 W, 400 mm/s) to generate an LED of 0.88 J/mm [Figure 4J]. However, the coarse columnar grains in a non-re-scanned alloy [Figure 4G] are replaced with a fine columnar grain in a laser re-scanned alloy. In addition, the clustering of fine grains is observed in the laser re-scanned alloy, as seen in the marked areas in Figure 4J, and this effectively reduced the grain size from 242 to 191 μm and the aspect ratio from 3.83 to 2.88.

Figure 4. EBSD parent grain reconstructed IPF map, HCP ε-martensite IPF map, and the grain size distribution of Fe-30Mn-6Si LPBF-built alloy from LED of (A-C) 0.25 J/mm (100 W, 400 mm/s), (D-F) 0.29 J/mm (175 W, 600 mm/s), (G-I) 0.44 J/mm (175 W, 400 mm/s), and (J-L) 0.88 J/mm (175 W, 400 mm/s, and 100% re-scan), respectively, and the IPF colour key for (M) FCC and (N) HCP. The build direction is from right to left and the grain boundaries are outlined in black.

The LPBF parameters also strongly influenced the crystallographic texture of the alloy. The predominance of a single colour in the EBSD IPF map indicated a substantially preferred orientation or strong texture. Figure 4G and J show that most of the cubic grains are aligned with their <001> direction parallel to the build direction in the 0.44 J/mm and 0.88 J/mm, whereas the texture was weak (broad range of colours) in 0.25 J/mm and 0.29 J/mm, Figure 4A and D, respectively. Therefore, the high LED and laser re-scan strategy generated prior austenite columnar grains that grow in their <001> direction parallel to the building direction. Meanwhile, Figure 4B, E, H, and K display the randomly orientated HCP martensite phase within the austenite grains.

The X-ray spectra of the Fe-30Mn-6Si reference alloy and the LPBF alloy made from different processing parameters were gathered and quantified using the Rietveld refinement method. The results are then shown in Figure 5 and Table 2. The major phases identified in the LPBF alloy are γ-austenite and ε-martensite because of their intense XRD peaks and composition that is ≥ 19 wt.%, as seen in Table 2. A dual-phased microstructure is expected in the Fe-30Mn-6Si alloy that underwent post-process treatment^[59], while the homogenised alloy may be single-phase austenite^[60], and such is observed in Figure 5. The existence of the γ and ε phases in the LPBF alloy is due to the far-from-equilibrium process conditions of the technique. Table 2 also reveals three other phases in the LPBF alloy; α-FeMn, α-FeSi, and FeO. The presence and composition of these phases are observed to vary in the LPBF alloy prepared for different parameters. For example, FeO was identified at 0.25 J/mm and 0.44 J/mm but not at 0.29 J/mm and 0.88 J/mm. Upon close inspection at 54.3° 2θ in the 0.29 J/mm, its 10$$\overline{2}$$ peak is visible. Several trials were made to include the low-intensity peaks from those three phases for a detailed analysis, but the quality of the resulting Rietveld refinement was unsatisfactory. A more detailed XRD scan is therefore necessary for a comprehensive analysis of those three phases.

Figure 5. The XRD patterns of the Fe-30Mn-6Si reference alloy and the LPBF alloy prepared at different process settings.

Key microstructural features associated with LPBF processing, such as the types and volume fraction of phases present, solidified grain size, morphology, and texture of the processed samples, were strongly influenced by the laser power, scan speed, and re-scan strategy. This shows that the desired microstructure is tailored by controlling laser power and scan speed to change the LED. The information on the thermal history of the resultant product is, however, necessary to completely understand the development of the microstructure.

Melt pool of single laser track scan

A polished surface of the reference alloy was subjected to single track laser scans at various LEDs. This resulted in the melting and subsequent solidification along the laser tracks, which generated a certain melt pool morphology for a given LED, when viewing a cross section perpendicular to the laser track. The effect of LED on the cross section of melt pool morphology is shown in Figure 6. Figure 6C and D show that a high LED creates both a deep and wide melt pool that penetrates at least 120 μm below the polished surface. In contrast, a low LED generates a relatively shallow melt pool of 50 μm deep [Figure 6A]. In Figure 6B, the melt pool became wide and deep when the LED was slightly increased from 0.25 J/mm to 0.29 J/mm by increasing the laser power from 100 W to 175 W and scan speed from 400 mm/s to 600 mm/s. Overall, there is sufficient lateral overlap of the melt pool tracks because the width of the melt pool is wider than the 0.45 μm distance of the parallel laser tracks.

Figure 6. The single laser scan melt pool profile on the polished Fe-30Mn-6Si reference alloy at varying LEDs of (A) 0.25 J/mm, (B) 0.29 J/mm, (C) 0.44 J/mm, and (D) 0.88 J/mm.

The melting mode at low LED (0.25 J/mm), as defined by Tenbrock et al., is conduction mode, and the rest, 0.29 J/mm to 0.88 J/mm, are in keyhole mode^[61]. In the authors’ single laser track investigation on 316 L stainless steel, the group used the melt pool depth-to-width ratio threshold of less than 0.8 as the conduction mode; above 0.8, the keyhole mode of melting transpired. Conduction mode of melting was observed at low LED, where the underlying regions are heated through the energy conducted from the surface^[62]. In the keyhole mode of melting, the high LED evaporated the metal and left a vapor cavity in the melt pool that enhanced laser absorption and enabled a deeper melt pool than in conduction mode^[63].

Simulated melt pool thermal profile

Using the melt pool profiles and a finite element analysis technique^[64] on the LPBF of Fe-30Mn-6Si alloy, the thermal profiles through the penetration distance of the melt pools were calculated as a function of laser scan strategy. The derived thermal conditions at different LPBF process settings as a function of melt pool depth are presented in Figure 7. It is known that the solidified microstructure prepared using the LPBF technique follows the solidification theory^[65-67], where the morphology is affected by the extent and direction of the temperature gradient and the solidification rate of the melt pool^[68]. Likewise, the cooling rate, a product of temperature gradient and the solidification rate^[69], dictates the size of the solidified structure^[70]. Therefore, the temperature gradient and the solidification and cooling rates are computed, and the temperature profiles are also derived.

Figure 7. Computed thermal profile of the LPBF fabricated Fe-30Mn-6Si alloy. (A) temperature gradient, (B) solidification rate, (C) cooling rate, and (D) maximum temperature as a function of melt pool depth at different LEDs. (E) Temperature profile for 0% (0.44 J/mm) and 100% re-scan (0.88 J/mm) at varying melt pool depths and processing time.

Figure 7A shows the variation in the temperature gradient within the melt pool for different LEDs. A low temperature gradient is initially observed from the surface of the melt pool, and it increases as solidification proceeds, leaving the bottom layer with the highest temperature gradient values of over 10⁴ K/m. The LPBF process has a typical temperature gradient range of 10⁴ to 10⁷ K/m^[71]. Temperature gradients between 10⁴ to 10⁵ K/m were associated with large melt pools^[72], and such were observed in the melt pool profile [Figure 6C and D]. The 0.25 J/mm has the highest temperature gradient at the surface at 2.32 × 10³ K/m as compared to the 1.03 × 10³ K/m, 9.83 × 10² K/m, and 7.72 × 10² K/m for 0.29 J/mm, 0.44 J/mm, and 0.88 J/mm, respectively. Moreover, the 0.25 J/mm has the steepest slope in the temperature gradient, followed by 0.29 J/mm. The temperature gradient of 0.44 J/mm and 0.88 J/mm are almost constant up to 50 μm melt pool depth and it increased gradually afterwards. The low temperature gradient for 0.44 J/mm and 0.88 J/mm at 0-50 μm was caused by their comparatively wide melt pool size in this area. Therefore, the temperature of the surrounding material is high, and the heat sink effect is low. As the distance from the top of the melt pool is increased, the melt pool achieves a lower width and lower surrounding temperature.

The solidification and cooling rates were numerically evaluated and remarked to be significantly influenced by scan speed than by laser power^[73,74]; hence, their influence at varying LED was evaluated. In the LPBF process, the solidification of the molten melt pool proceeds as the laser track leaves the melt pool. The rate of solidification at varying LED was presented in Figure 7B, where the rate at the surface of the melt pool was similar to the applied scan speed. The 0.29 J/mm LED had a faster solidification rate at the surface compared to the rest of the LEDs because a scan speed of 600 mm/s was applied. All the solidification rates decreased sharply from the surface of the melt pool until the 50 μm depth, but the slope was noticeably steeper at 0.25 J/mm and 0.29 J/mm LEDs than at 0.44 J/mm and 0.88 J/mm LEDs. After the 50 μm depth, the solidification rate for 0.29 J/mm LED decreased slowly. For 0.44 J/mm and 0.88 J/mm LEDs, the slope of the solidification rate only changed after ~90 μm and then became stable at 50 mm/s.

The fast-moving laser in LPBF imparts a high cooling rate of 10⁴ to 10⁶ K/s^[75-77]. Presently, a cooling rate of 10⁵ K/s was noted in the Fe-30Mn-6Si LPBF-built alloy. The relationship between LED and the cooling rate was observed in Figure 7C. From the surface of the melt pool, at 0 μm melt pool depth, the cooling rate was constant and then dropped at a particular depth depending on the LED. Its magnitude was also dependent on LED. The low LED (0.25 J/mm) had a stable cooling rate of 9.28 × 10⁵ K/s up to ~30 μm melt pool depth, and for 0.29 J/mm LED, it was stable at 6.21 × 10⁵ K/s until ~60 μm deep. For the LEDs of 0.44 J/mm and 0.88 J/mm, the cooling rates were rather stable until ~120 μm deep at 3.93 × 10⁵ K/s and 3.09 × 10⁵ K/s, respectively. Moreover, the percentage change in the cooling rates after varying the laser power and scan speed was the same at ~58%. When the laser power was raised from 100 W to 175 W, the cooling rate dropped from 9.28 × 10⁵ K/s to 3.93 × 10⁵ K/s, while the increase in scan speed from 400 mm/s to 600 mm/s increased the cooling rate from 3.93 × 10⁵ K/s to 6.21 × 10⁵ K/s.

The maximum calculated temperature in the melt pool as a function of melt pool depth for different LEDs derived from the FEA analysis is shown in Figure 7D. The melt pool temperature responds positively to the increase in LED, and it is observed to decrease within the melt pool. For example, at the melt pool surface, a temperature of roughly 2081 °C, 2173 °C, 3019 °C, and 3256 °C was computed for 0.25 J/mm, 0.29 J/mm, 0.44 J/mm, and 0.88 J/mm LEDs, and it decreased to 1360 °C, 1755 °C, 2484 °C, and 2716 °C at 50 μm melt pool depth, respectively. As a guide for the melting of the powder, the melting points of the constituent elements (Fe = 1535 °C, Si = 1410 °C, and Mn = 1245 °C) in the blended powder are likewise inscribed in Figure 7D. The observed temperatures at the melt pool surface are beyond the melting temperature of the powder, which may have likely evaporated some elements. In particular, the loss of manganese is expected when an LED over 0.25 J/mm is applied because of its relatively low boiling temperature (Fe = 2750 °C, Si = 2357 °C, and Mn = 2062 °C), Figure 3B.

The solidified surface of the LPBF-built alloy fabricated at 0.44 J/mm LED was re-scanned at 175 W and 400 mm/s (100% re-scan) to further promote the alloying of the blended powders. The added step doubled the LED from 0.44 J/mm to 0.88 J/mm at each layer and consequently raised the temperature in the melt pool, as shown in Figure 7E. At the surface, the maximum temperature for 0.44 J/mm LED reached 3019 °C, and this was similar for 0.88 J/mm LED, as depicted by the first peak in the temperature profile on the surface of the laser re-scanned LPBF-built alloy. The alloy was expected to have solidified after the first scan because the temperature dropped to almost 270 °C. However, the additional re-scan step reached a much higher temperature of 3256 °C, as seen in the second peak, than in the first scan because the re-scan started at a relatively high temperature of 270 °C, and there is a difference in the thermal conductivity of the powder and the alloy^[78,79]. A similar pattern showing the two temperature peaks was observed at different melt pool depths when the re-scan strategy was applied.

Microstructure evolution

Both the highly directional heat flow conditions and large temperature gradients generated during laser melting of an outermost layer of metal powder, which usually also resulted in the partial remelting of the already solidified grains of the underlying built substrate, favour epitaxial growth of these existing grains in certain crystallographic directions towards the heat source (i.e., they grow antiparallel to the direction of heat flow into the underlying substrate). For certain laser input conditions, a highly directional columnar morphology and strong texture were frequently observed in LPBF-built alloys^[80,81]. However, the EBSD maps shown in Figure 4 revealed a gradual change in the microstructure from a nearly equiaxed to columnar grain structure as the LED was increased.

The equiaxed-to-columnar transition in the grain structure was commonly observed within the melt pool, and this transition depends on the alloy chemistry and the heat transfer conditions according to the LPBF processing conditions^[64,82-86]. A near-homogeneous grain structure with weak texture was achieved when Attard et al. applied the island scan strategy or checkerboard style^[87], a standard parameter unique to the setup^[32], and when Ewald et al. heated the build platform to 500 °C^[21]. Attard’s group associated this with the even distribution of heat in the island scan strategy. Meanwhile, the heated build platform in the 0.5 mm sized product reduced the temperature gradient in Ewald et al.’s LPBF product, which also reduced the temperature gradient and promoted a nearly homogeneous and equiaxed microstructure^[21]. The lack of grain morphology transition in the melt pool in the present Fe-30Mn-6Si LPBF alloy may have been caused by the island scan strategy with 45° scan rotation in the subsequent layers, leading to a homogeneous grain morphology in each parameter setting.

The similar microstructures of LPBF processed parts and conventionally welded components make it convenient to describe the solidified LPBF microstructure in terms of the well-established physical metallurgy principles associated with fusion welding^[69]. Grain shape and scale were defined by the solidification theory, and may be controlled by the temperature gradient G, solidification velocity R, the temperature solidification range of an alloy ∆T, and the liquid diffusion coefficient DL^[65,67]. The relationships between these key solidification parameters are given below^[66]:

where the G/R ratio and the G·R product, which is the cooling rate, can predict the morphology and dimensions of the solidified microstructure, respectively. For example, a low G/R value correlates to equiaxed grains, with the morphology transitioning to columnar dendritic, cellular, and then to planar for increasing values of G/R, and the high cooling rate resulted in a fine solidified grain structure^[70]. Investigation of the thermal history of LPBF-processed alloy was necessary for understanding its expected final microstructure, and in Figures 4 and 7, the select parameters showed that the different thermal profiles affected both the morphology and dimensions of the solidified grains.

The solidification of grains in LPBF-processed alloys follows the well-established nucleation and growth processes in solidified metals and alloys. Li and Tan^[88] provided the general grain characteristics of LPBF alloys and summarised two possible nucleation mechanisms: (i) bulk nucleation; and (ii) epitaxial or surface nucleation. Bulk nucleation occurs on the top side of the melt pool and at the head of a solidification front^[88]. Nuclei also form from the partially melted powder in the melt pool^[89], and they can survive given a sufficient volume of surrounding undercooled liquid metal^[90]. These formed grains then assume an equiaxed morphology due to the low G/R ratio on the top side of the melt pool^[91]. Epitaxial nucleation occurs at the interface of the melt pool and the substrate, or at the previously solidified layer^[92]. A high LED and a low solidification rate in the melt pool encouraged grains to grow in a preferred crystallographic orientation^[93], which was <100> for cubic and <1010> for hexagonal metals, respectively^[69]. Grains possessing these favoured orientations outgrew grains with less favourable orientations^[65], eventually generating a highly textured, columnar microstructure^[94].

The prevalence of a highly textured and columnar grain morphology at the high LED settings [Figure 4G] suggested an epitaxial mechanism. Without an added and known potent nucleating particle in the elemental mixture and because of the steep temperature gradient on melting and solidification, the previously solidified layer would act as a suitable substrate for continued growth into the melt pool, whereby the partly melted grains propagate by epitaxial “nucleation” towards the heat source. Equiaxed grains may form on the top surface of the melt pool when the melt pool trail ended because of the low G/R ratio in this region, and such was seen on the last fabricated layer in NiTi^[86]. In the Fe-30Mn-6Si, at 0.44 J/mm LED, a temperature gradient = 9.83 × 10² K/m, cooling rate = 4 × 10⁵ K/s, the slow solidification rate of R = 3.98 × 10² mm/s, and the melt pool temperature of ~1300 °C at 140 μm melt pool depth were sufficient to melt the solidified equiaxed grains in the previous laser scan and then subsequently re-solidify into columnar grains. A similar grain morphology holds for the laser re-scanned LPBF alloy, albeit grains were relatively fine and less columnar when laser re-scanning was applied.

The re-scan strategy had been reported to improve surface quality^[66], increase density^[95], and reduce residual stress^[96] in AM components. This additional step was included in this study to enhance the alloying of the blended powders, and this resulted in a notably different microstructure from that of a non-re-scanned alloy. The melt pool width, depth and overall area associated with the re-scan strategy [Figure 6D] were considerably larger than after single scanning [Figure 6C], and this is caused by the higher thermal conductivity of the solidified layer than the powder material^[97]. Hence, the enhanced heat transfer in the solidified layer resulted in a more pronounced melt pool, which was reflected in the calculated thermal profile. A coarse and columnar grain structure was still expected in the laser re-scanned LPBF alloy because the parent grains in the non-re-scanned alloy have solidified into columnar grains. The relatively gentle slope of solidification for 0.88 J/mm LED [Figure 7B] and its low cooling rate (3 × 10⁵ K/s) promoted the epitaxial growth of columnar grains, but its temperature profile shown in Figure 7D suggested that remelting of the previously solidified layer had occurred.

Completely remelting an alloy reshapes its microstructure, and such was evident in this work by the decrease in the average grain size and aspect ratio in the remelted LPBF alloy. A region of coarse and refined grains was apparent on close inspection in the re-solidified structure (marked areas in Figure 4J). Xiong et al. reported a similar observation in pure tungsten^[79]. During re-scanning, the laser irradiated heat initially remelted the surface and the convection current in the melt pool^[98] engulfed and remelted the remaining solid within the melt pool. This caused some of the initially formed columnar grains to be separated and these freed grains became the nuclei for growth^[66]. The fast-moving laser that drives the rapid cooling rate (10⁵ K/s) in the LPBF process curbs the growth of the newly nucleated grains and freezes them into a fine microstructure^[99], thereby forming regions of non-uniform microstructure.

For the low LED (0.25 J/mm), the melt pool temperature at 50 μm pool depth was 1360 °C which was enough to melt the blended powders and potentially melt the surface of the previously solidified layer. However, the high cooling rate of 9 × 10⁵ K/s and the high solidification rate at this setting resulted in the retention of the equiaxed grain morphology. Moreover, the chemical segregation [Figure 2C] preserved in this LED suggests the presence of partially alloyed powder both in the melt pool and the solidified layer when the next layer was melted. The bulk nucleation mechanism was favoured in the presence of partially alloyed powder since they can act as heterogeneous nucleation sites and impede the epitaxial growth of the previously solidified equiaxed grains at the bottom of the melt pool^[100].

For the high laser power and fast scan speed (175 W, 600 mm/s, 0.29 J/mm), the melt pool depth of 110 μm could get through an equivalent of two powder layers and had enough heat to sufficiently remelt the previously solidified layer and re-solidify them into a full-columnar structure. However, the solidified grains shown in Figure 4D were equiaxed and rather coarse (105 μm) compared to the finer grains (64 μm) associated with the low LED (0.25 J/mm) in Figure 4A. The partially melted powder observed at this setting could have induced the bulk nucleation of the grains and interrupted the epitaxial growth of grains.

The significant influence of the studied processing parameters on the resultant LPBF microstructure presents an opportunity to control the microstructure and texture, and therefore the properties of any given component. For instance, a columnar and textured grain structure is ideal for the pseudoelastic behaviour seen in Fe-Mn-Al SMAs^[101,102] and the unrestricted martensitic phase transformation for shape memory in Cu-based SMAs^[103].

Possible factors influencing hardness

Effect of grain size

Figure 8 shows the hardness of the LPBF-built alloy prepared as a function of LED. The two low LEDs (0.25 and 0.29 J/mm) have a close hardness value (278 ± 7.6 and 273 ± 3.9 HV2, respectively). The hardness in the two high LEDs (0.44 and 0.88 J/mm) is also close (287 ± 5.5 and 292 ± 3.6 HV2, respectively.) Meanwhile, the reference as-cast alloy had the lowest hardness (226 ± 6.7 HV2). The hardness of the material varies with grain size according to the classic Hall-Petch relation^[104,105]. Also, in Figure 8, the grain size increases with an increase in LED up to 0.44 J/mm, and then drops when the laser re-scanning step was added to achieve 0.88 J/mm LED. This change was associated with the thermal history of the LPBF alloy [Figure 7]. The reference alloy has a lower hardness than each of the LPBF-fabricated alloys. This is due to the coarse, equiaxed grains generated in the reference alloy by hot working and the 14-h homogenisation^[17,106,107]. In the LPBF alloy, the hardness is seen to increase together with the grain size, thereby negating the established influence of grain size on hardness. This suggests that some other factor affects the hardness of the LPBF-fabricated alloys.

Figure 8. Relationship between grain size and hardness for both the reference alloy and the LPBF alloy fabricated at different LEDs.

Effect of phase types

Figure 9 shows the relationship between the volume fraction of phases and hardness as the LED is increased. This parameter was also found to influence the relative volume fractions of austenite and ε-martensite in the LPBF alloy, whereby austenite decreases while ε-martensite increases with increasing LED. Martensite is formed from austenite by either a stress or thermally induced transformation^[41,108-110], which results in the observed inverse relationship between the two phases. The effect of LED on the volume fractions of the phases was associated with the grain size, and that is, fine grains are detrimental to the formation of the ε-martensite phase^[111,112]. The increase in ε-martensite volume fraction may also be caused by the decrease in Mn concentration at high LED [Figures 3B and 9]^[113]. Hardness as a function of phase volume fraction is also given in Figure 9, where it appears that hardness directly correlates with the amount of ε-martensite in the microstructure. This confirms that both the type and volume of phases present in the LPBF-fabricated alloy have a very strong effect on hardness.

Figure 9. The relationship between γ-austenite and ε-martensite phase fractions and hardness for both the reference alloy and the LPBF alloy fabricated at different LEDs.

Boundaries exist between the phases in a multi-phased material, and each phase has a distinct characteristic^[114]. The reference alloy was fully austenitic, whereas the LPBF alloy contains both austenite and ε-martensite, and other minor phases [Figure 5 and Table 2]. Since austenite is much softer than ε-martensite^[115], this resulted in the low hardness of the reference alloy. In comparison, the amount of austenite and the pre-existing ε-martensite in the LPBF alloy, for example, in 0.44 J/mm LED, were 62% and 31%, respectively. The relationship between hardness and the volume fraction of ε-martensite has also been reported in a powder metallurgy fabricated Fe-30Mn-6Si alloy^[116]. A high hardness was found in the as-sintered condition, but it decreased after heat treatment because of the corresponding decrease in ε-martensite. The addition of 5 wt.% Cr, an austenite stabiliser^[117,118], in an as-cast Fe-30Mn-6Si alloy also resulted in a soft alloy due to the absence of ε-martensite^[119].

Pre-existing ε-martensite has been reported to block plastic flow, which leads to high work hardening^[120]. The impeding action of pre-existing ε plates was observed by Sato et al. using TEM, and they also reported a hardened Fe-30Mn-1Si alloy^[1]. The group likened the ε plate phase boundary to a grain boundary. In the Fe-30Mn-6Si reference and LPBF alloys, ε plates may have nucleated and grown in the austenite grains during the hardness test to accommodate strain. But the thick pre-existing ε-martensite plates [Figure 4B, E, H and K] may have restricted the nucleation and growth of the stress-induced ε plates. This is a contributing factor to the higher hardness found in the LPBF alloy than in the reference alloy and suggests that an increase in hardness is strongly related to the increase in the amount of ε-martensite in the microstructure.

Effect of residual strain

The far-from-equilibrium processing conditions in LPBF introduce residual strains that may also influence hardness. Since residual strain is associated with crystal misorientation^[121-124], the relationship between crystal misorientation and hardness is presented in Figure 10. Comparing the reference alloy and the LPBF alloy prepared at 0.44 J/mm LED, the hardness of the former was significantly lower (227 HV2) than the latter (287 HV2). The corresponding average Kernel average misorientation (KAM) in the reference alloy is also lower (0.44°) than in the LPBF alloy (0.64°). A high residual strain has been associated with a high density of low-angle grain boundaries^[125] and, as such, the density of these boundaries (2° to 10° misorientation) in both the reference and LPBF alloys were measured by EBSD to be 3% and 7%, respectively. Hu et al. reported that in pure Ti sheet, the hardening effect due to low-angle boundaries was dependent on the level of strain^[126]. At strains up to 30%, the high-angle boundaries (HAGB) contributed to the hardness, but for strains above 30%, the density of the low-angle boundaries (LAGB) increased. The latter was suggested to be the biggest contributor to hardness. This was also noted in both 304 L stainless steel and Ni-Co alloys, whereby the hardness increased with increasing residual strain^[127], and in a Fe-Ni alloy, the hardness decreased when the residual strain was relieved^[128]. A dislocation has to overcome the grain boundary energy, both high- and low-angle, for it to move through the boundary, and the magnitude of the LAGB interfacial energy is a function of the degree of crystallographic misalignment^[114]. Thus, the high hardness in the LPBF alloys as compared to the reference alloy was also caused by the inherent residual strain that resisted the localised deformation.

Figure 10. The relationship between residual strain (average KAM value) and hardness for both the reference alloy and the LPBF alloy fabricated at different LEDs.

Figure 10 shows a positive correlation between the average crystal misorientation from EBSD analysis and the computed temperature gradient using the FEA of the melt pool as a function of LED at differing depths from the melt pool surface. On the top surface of the melt pool, the highest temperature gradient (2.32 × 10³ K/m) was computed for 0.25 J/mm LED. It then decreased as the LED increased, with 0.88 J/mm LED having the lowest temperature gradient (7.72 × 10² K/m). At 50 μm depth from the melt pool surface, the temperature gradient in 0.25 J/mm LED increased substantially to 1.90 × 10⁴ K/m (~7× increase) while it remained almost constant in 0.88 J/mm LED at 1.15 × 10³ K/m (~0.5× increase). This then corresponds to an average misorientation of 0.65° and 0.49°, respectively, and suggests that a high average KAM indeed correlates with a high temperature gradient.

The residual strain in the LPBF alloy is caused by the local heat application of the laser, which introduces tensile stress in the molten layer and compressive stress in the solidified lateral and underlying layers^[129]. These stresses, if not released, result in residual plastic strains. Several authors looked into minimising thermal stress in the LPBF-fabricated alloy. Vrancken et al., Lu et al., and Liu et al. agreed that a short laser scan length introduced less thermal stress, while Mishurova et al. emphasised the importance of large melt pool volume to lessen thermal stress^[130-133]. The scan strategy was maintained during the LPBF of Fe-30Mn-6Si alloy, but the melt pool for 0.88 J/mm LED was comparably large than for the other LEDs [Figure 6D]. However, Liu et al. added that a low LED is necessary for a small thermal stress, and these workers pointed out that a low thermal stress in short laser scan length was caused by the release of stress through cracking^[132]. A low average misorientation (0.49°) and a high hardness [Figure 10] in the highest LED (0.88 J/mm) suggest otherwise. More so, the ε-martensite formation in 0.88 J/mm LED may be stress induced and its volume fraction was high (45.8%). This entirely suggests that the residual strain may have been released through the formation of cracks since the LPBF alloy fabricated at 0.88 J/mm LED had comparably more cracks than the 0.25 J/mm LED (Figure 2C and D, respectively). A more thorough investigation is, however, warranted to understand the residual strain in the LPBF alloy fabricated from a homogeneously mixed Fe-30Mn-6Si powder.

The residual strain in the LPBF alloy, according to the average crystal misorientation data, was also shown in Figure 10 to decrease with increasing LED. This was reported to increase the hardness^[127], and when the residual strain was relieved, the hardness decreased^[128]. For this reason, the relationship between the average KAM value and hardness in both the reference alloy and LPBF alloy at different processing conditions was investigated. However, the decrease in residual strain corresponded to an increase in hardness, as seen in Figure 10, which differed from the previous reports^[126-128]. Therefore, the residual strain may have indeed been relieved from the LPBF alloy through the formation of cracks, particularly in 0.88 J/mm LED.

The influence of grain size, presence and volume of phases, and residual strain was analysed to identify the possible factor affecting the hardness of the LPBF-fabricated Fe-30Mn-6Si alloy. Hardness is known to increase as the grain size decreases, ε-martensite volume fraction increases, and residual strain increases. It was observed that the increase in hardness was mainly influenced by ε-martensite at high LEDs of 0.44 J/mm and 0.88 J/mm [Figure 9], while grain size and residual stress were not seen to influence their hardness according to the accepted theories^[104,105] and observations^[126-128]. The sub-grain phase boundaries between the different variants of ε-martensite increased the hardness in those large-grained microstructures. Also, high hardness value was found in the LPBF alloy with low residual strain [Figure 11]. This suggests that the residual strain at the high LED was relieved after cracks were formed. At 0.25 J/mm LED, hardness has increased relative to the reference alloy because of the low grain size and high vol.% martensite (Figures 8 and 9, respectively). However, the hardness for 0.29 J/mm LED slightly decreased (2%) despite the increase in martensite vol.% (26%) as compared to that of 0.25 J/mm LED. Concurrently, the grain size of the former increased by 65%. From the given correlations, the increase in grain size was likely the main mechanism for the slight decrease in hardness from 0.25 to 0.29 J/mm LED.

Figure 11. The relationship between crystal misorientation (average KAM) and temperature gradient derived from the simulated thermal profile of the melt pool as a function of LED at differing depths from the melt pool.