Smart-Alloying for tailored microstructures in laser-based metal additive manufacturing

As an additive manufacturing process, laser powder bed fusion (PBF-LB/M) enables the production of structural components with a geometrically high spatial resolution compared to conventional processes. However, it often leads to location-dependent material properties. The latter is determined by the solidification and cooling of the material volume limited to a small melt pool generated by the laser and the crystallization and growth kinetics that take place in the process. Depending on the material, process parameters, and thermal conditions under which the process takes place, heterogeneous microstructures can result. These microstructures are characterized by pronounced textures, uneven grain size distribution, or locally different precipitation states, all of which are responsible for both the global and local mechanical properties achieved.^1–5 

In addition to the approaches that specifically influence the mechanisms responsible for the formation of the microstructure via the process conditions and, thus, use of them for the on-demand grading of their properties,^6–9 there are also other methods. One approach achieves functional grading in additive manufacturing (FGAM) by changing the local material composition. This is done by applying powder with a different composition and processing it with the laser as usual for PBF-LB/M.^10,11 In such functional grading, also known as multimaterial additive manufacturing, various methods have been developed in the recent past to reduce cross-contamination of the powder remaining at the end of the process by the localized application of powders of different compositions. The powders are either selectively deposited on the already applied powder layer using a separate dispensing system^12,13 or applied simultaneously in one layer using a special multimaterial coating system.^10,14 However, during the joint laser additive processing of these two adjacent powder components, defects can occur in the manufactured structure, particularly at the material transitions, such as cracks, oxides, and pores.¹⁵ The compatibility of the material components in terms of thermophysical properties must be considered in order to avoid defects.¹⁶ 

A method with which the local chemical composition can be gradually adjusted in contrast to the previously mentioned methods is made possible by the so-called Smart-Alloying approach.¹⁷ In this process, nanoparticulate solids of the required alloying elements dispersed in a water-based solvent are applied in situ to a heated substrate, which can be done extremely precisely due to the special dosing options for liquids. After evaporation, the solids remain on the substrate and are melted by the laser together with the applied powder as usually done via PBF-LB/M and incorporated into the solid solution during solidification. In this publication, it is shown that through the use of Smart-Alloying, the carbon content of steel can be significantly increased locally, allowing the microstructure to be tailored by the modification of the chemical composition. Furthermore, variation of the process parameters of PBF-LB/M enables additional adjustments to the microstructure.

A large number of steels have already been processed using PBF-LB/M. The complex thermal conditions, such as high temperature gradients, high cooling rates, and repeating temperature cycles, can lead to challenges. In the as-built state, the elongation at break and fatigue strength do not reach the levels of samples from the comparable states in conventional manufacturing processes, in some cases. This is due to process defects such as pores or cracks, as well as local residual stresses. In addition, the combination of process parameters significantly influences the melt pool geometry and cooling conditions, which have a direct impact on the resulting microstructure. For this reason, steels of identical chemical composition can have different mechanical properties.^18–20 

In this work, the carbon content of ferritic stainless steel X2Cr12 is varied locally and the influence of the process parameters on the microstructure is investigated. Chromium, as an alloying element, constricts the γ-region under equilibrium conditions due to its strong tendency to form carbides and its ferrite-stabilizing effect. Moreover, chromium lowers the martensite start temperature. The interaction of the two elements, chromium and carbon, on the microstructure is particularly noticeable with regard to the expansion of the constricted γ-region with increasing carbon content. The limit of the range of the homogeneous γ-phase is about 12 wt. % Cr in the equilibrium state of the low-carbon or carbon-free Fe-Cr system. Contents above this lead to a completely ferritic solidification of the microstructure. Below 500 °C, chromium-containing ferrite (α′) can form, which, like the sigma phase (σ), can lead to embrittlement of the microstructure.²¹ Independent of these phases, by increasing the carbon content to around 0.6 wt. %, the homogeneous γ-phase can be maintained up to a chromium content of 19 wt. %, which once again demonstrates the major influence of carbon as austenite-stabilizing element.²² 

For the tailored modification of the microstructure in PBF-LB/M, local in situ re-alloying was used, which was published in Ref. 17 as Smart-Alloying. The process was developed using the ferritic stainless steel X2Cr12. These works are adopted by the authors and expanded upon by selectively influencing the re-alloyed structure with the system-side laser. The process steps of PBF-LB/M supplemented with Smart-Alloying are illustrated in Fig. 1. In this process, the suspension, consisting of solvent deionized water, surfactant sodium dodecyl benzenesulfonate at a concentration of 8.71 mM, and nanoparticulate carbon at a concentration of 2.5 wt. %, is applied to the already solidified layer. The solvent dries instantly due to the existing surface heat and the acting protective argon gas flow. As the platform is lowered, coating with the powdered base material X2Cr12 begins. Finally, melting is performed with the high-energy laser source, whereby the base material is transferred into the liquid phase with the carbon particles beneath it, and solidified with a modified chemical alloy composition.

The chemical powder composition of base material X2Cr12 is visible in Table I. The powder was purchased from Ametek, where it was produced by water atomization and classified to a particle size distribution of a d₉₀ of 71.5 μm.

The fluid system used for Smart-Alloying is an additional module in our AconityMIDI+ system (Aconity3D GmbH, Herzogenrath, Germany). This system enables a three-dimensional movement with two additional axes to the recoating direction and suspension dosing by a print head from Nordson EFD. The 3D-printer features a 700 W fiber laser with a wavelength of approximately 1070 nm. The diameter at the focal plane is 63.0 μm at 50% of the maximum power. The inert gas atmosphere (argon 4.6) contained a maximum oxygen content of 100 ppm during the process, and no preheating of the build platform was used.

The experiments conducted in this study were carried out with a platform reduction to a diameter of 100 mm. The process parameters were kept constant across all experiments, including a layer thickness of 50 μm, a hatch distance of 0.09 mm, a scanning speed of 650 mm/s, and a primary exposure laser power of 350 W. Additionally, a layer rotation of 67° was applied. From this parameter combination, a volume energy of 119.7 J/mm³ and a linear energy of 0.54 J/mm resulted, leading to a relative density evaluated by an image analysis of over 99.99%.

The reference state of a tailored microstructure was generated by re-alloying a test specimen in the last five upper layers [Fig. 2(a)]. Basically, cylindrical geometries were used which had a diameter of 10 mm and a height of 3 mm. The top five layers were each alloyed with a single suspension application with a pulse time of 60 ms, resulting in a volume of 3.9 μl each. After wire EDM separation of the samples from the build plate, the samples were additionally separated along the build direction. Subsequent preparation was carried out using standard metallographic preparation methods before electron microscopic examinations were conducted. Using the CBS detector of a Dual Beam Helios G4 PFIB CXe (Thermo Fisher Scientific, Waltham, MA, USA), backscattered electron (BSE) images were taken at a high voltage of 5 kV, with a current of 0.4 nA at magnifications ranging from 1500x to 35 000x. The patterns were recorded using the Ametek EDAX detector, and phase analyses regarding ferrite, austenite, and chromium-rich carbides M(Fe, Cr)₃C, M(Fe, Cr)₇C₃, and M(Fe, Cr)₂₃C₆ were performed. The results of the phase analysis are overlaid in all figures with the confidence index, displayed as a gray value representation at 70% transparency.

For the exact determination of the present chemical composition and existing phases, the regions with complex phases resulting from local in situ re-alloying were analyzed by using transmission electron microscopy (TEM). Therefore, a lamella of an area showing ferritic as well as a complex microstructure composed of M₃C, M₇C₃, and austenite [see Figs. 3(b) and 3(c)] was extracted by using a dual-beam focused ion beam-scanning electron microscope (FIB-SEM) FEI Helios Nanolab 600 (Thermo Fisher Scientific, Waltham, MA, USA) equipped with multiple detectors for BSE imaging, energy dispersive x-ray (EDX) diffraction, and electron backscatter diffraction (EBSD). The TEM lift-out was performed perpendicular to the image plane through ion milling with Ga+ ions at acceleration voltages between 5 and 30 kV (Fig. 4).

The TEM investigations were performed on a FEI Talos F200X (Thermo Fisher Scientific, Waltham, MA, USA) operated at an acceleration voltage of 200 kV. The instrument is equipped with multiple detectors for STEM imaging, a XFEG high-brightness gun, a large-area Super-X EDX detector, and a Gatan Continuum ER spectrometer. The microstructure and chemical composition were analyzed by high-angle annular dark-field STEM (HAADF-STEM) in combination with STEM-EDX.

Additionally, three more samples were examined for carbon content measurements using spark optical emission spectroscopy (SOES) with an ARL 3460 (Thermo Fisher Scientific). For this purpose, only the surface of the samples was lightly ground to create a uniform structure. A grinding wheel made of zirconium dioxide (ZrO₂) was used for this purpose.

Starting from the reference samples with approximately 3 mm total height, the distribution of carbon depending on subsequently printed base material without carbon addition was considered [Fig. 2(b)]. For better handling during SOES measurements, the sample height was increased to 5 mm where the last five layers were alloyed with the carbon suspension of the same parameters as for the reference state. After Smart-Alloying, the conventional PBF-LB/M process was continued, printing three samples each with an additional sample height of 0.25, 0.5, 1.0, 1.5, and 2.0 mm of the base material on top of the carbon-modified layers.

The carbon profile along the build direction was measured using SOES. The area around the re-alloyed zone was covered with grinding steps from 0.05 to 0.1 mm, with distances increased in the base material area from 0.2 to 0.25 mm. With the known influence of the subsequent layers on carbon distribution, the sample with an additional height of 1.5 mm base material was used as a reference for investigating the influence of double exposure [Fig. 2(c)]. Again, the re-alloying process was carried out after a sample height of 5 mm in five layers. However, both the re-alloyed layers and the subsequent base material of a height of 1.5 mm were always provided with a second additional exposure. Following the primary exposure, four specimens were produced with a secondary exposure laser power of 200, 350, and 500 W each. Carbon profiles were measured using SOES on three of the samples, and one additional sample underwent image analysis. For the optical microscopy of the relevant microstructural areas, the prepared longitudinal sections were etched with V2A etchant for 60 s at 60 °C.

With an increase in carbon content per layer, the influence on the deliberately modified microstructure was further increased. The approach for this involved a higher amount of suspension per layer [Fig. 2(d)]. In total, five dosing processes per layer were realized, with a pause time of 20 s between each to prevent overflow of the sample edges. The sample geometry consisted of a diameter of 10 mm and a sample height of 5 mm. Smart-Alloying was carried out in the uppermost five layers of the geometry. This approach was additionally expanded by subjecting a specimen to a secondary exposure [Fig. 2(e)]. Thus, in the top five layers, the two-component powder layer was first melted with the base parameter set. Then, the chemically altered layer was subjected to the base parameter set again.

The difference in increased carbon dosing compared to the initially determined reference state was examined using the image analysis methods of FIB-SEM. Similarly, the influence of secondary exposure on the locally modified microstructure of increased carbon content was analyzed. In comparison to the reference state, additional carbon analysis with SOES was conducted in the re-alloyed area.

With the implementation of in situ re-alloying in the top five layers, local carbon content of approximately 0.1 wt. % was measured during SOES investigations. This decreases below the re-alloyed layers to the carbon content of the base material. The localized increase in carbon content is evident in the image quality of the upper half of the sample magnified by a factor of 10. The theoretically re-alloyed region of 250 μm is delineated by two dashed orange lines [Fig. 3(a)]. Grain boundaries within these five layers are highlighted more distinctly by contrasting appearances than those in the region below the dashed lines. Along the re-alloying zone, this grain boundary occupancy decreases from the upper surface layer to the lower boundary. This gradient also accompanies a change in grain morphology. The region near the surface layer is characterized by globular grains, some of which extend partially beneath the re-alloyed layers. With decreasing sample height, there is grain growth and an increasingly columnar primary structure, comparable to that of the initial microstructure. Within a distance of 250 μm below the lower dashed line, contrasting deposits are still present at the grain boundaries, highlighted by orange arrows [Fig. 3(a)]. The density of these phases decreases compared to the re-alloyed zone and is no longer visible in the lower image area. Upon closer examination of a defined area of interest within the re-alloying zone using the CBS detector, substructures along the grain boundaries are highlighted [Fig. 3(b)]. The angular deposits partially surround complete grains and, emanating from the grain boundaries, extend into the surfaces in a needle-like manner (orange arrows). With the enlargement of the intricate areas in Fig. 4(c), these are determined by a fine, needle-like structure. Phase analysis of the region reveals a similarly cubic-centered structure surrounded by carbides of type M₃C and M₇C₃ [Fig. 3(d)]. These are martensitic grains, and thus tetragonally distorted, which cannot be distinguished from ferrite in the phase analysis of EBSD. This is additionally confirmed by the presence of small amounts of retained austenite. Austenite formed during the cooling process is not stable down to room temperature with the existing carbon content and transforms into martensite when the temperature falls below the martensite start temperature. Due to the primary precipitation of chromium-rich carbides, the structure locally depletes in the ferrite-stabilizing element. The locally expected high amounts of carbon cannot be further bound in carbides and a fast redistribution occurs. The austenite remains down to room temperature due to a further, locally reduced martensite start temperature, thereby complementing the as-built microstructure with retained austenite. Below the theoretical re-alloying zone, these substructures are only sporadically visible. This results from the melting of the underlying layers that were not re-alloyed with carbon. With further printing progress, the proportion of un-alloyed layers decreases during the remelting process, leading to an increase in the gradient up to a maximum carbon content of 0.1 wt. %. The confidence index indicates the existing uncertainties in the entire image section. Therefore, quantitative classification of the two carbide forms should be approached with caution and interpreted primarily, qualitatively. For example, the carbide structures displayed within the grains, which are rather questionable due to the formation mechanisms and can be attributed to measurement uncertainties [Fig. 3(d)].

To determine the present phases and the chemical composition of the specific microstructure resulting from the local in situ re-alloying (Fig. 3), a site-specific TEM lamella was extracted from the region of interest (see Fig. 4). Figure 5 shows an HAADF-STEM overview image of the TEM lamella (a) and the area selected for a detailed analysis of the present chemical composition and phases (b) as well as a summary of the determined phases within this representative area.

An analysis via selected area diffraction (SAD) has revealed that the microstructure of the lamella mainly consists of α-ferritic grains [Fig. 6(a)] with small γ-austenitic grains [Fig. 6(b)]. It is difficult to finally clarify if an α-ferritic or a martensitic microstructure is present as a differentiation between both phases is not possible via SAD due to the very similar diffraction pattern, e.g., lattice parameters, of α-ferrite and martensite. But, due to the high chromium content (12 wt. %) and the characteristics of the LPBF process, e.g., high cooling rates, it is assumed that a martensitic microstructure is present.

In Fig. 7(a), the HAADF-STEM image is depicted together with the corresponding EDX maps. In general, the containing chemical elements are homogeneously distributed over the surface area. It is conspicuous that enrichments of chromium and carbon as well as iron depletion can be found along a main grain boundary. It is suspected that these are chromium-rich carbides. However, further detailed analysis is required for the determination of the present phase(s) or types of these carbides along the main grain boundary in the HAADF-STEM image in Fig. 7, as the particles are too small for a successful SAD analysis.

Following the process of in situ re-alloying, additional layers of the un-alloyed base material result in the formation of a carbon gradient in the upper transition region. This gradient does not reduce the maximum carbon content in the re-alloyed zone significantly. Regardless of the subsequent sample height, a local increase in carbon content was observed (Fig. 8).

Neither the position of the maximum carbon content reached along the build direction nor the elemental content itself were significantly influenced by the subsequently printed sample height. This can be attributed to the aforementioned primary precipitation of chromium-rich carbides. Consequently, comparatively fast-diffusing carbon is bound within the microstructure and remains localized despite additional temperature influences. The additional remelting processes of the subsequent layers affect carbon only when the prior solidified layer is completely melted, leading to a decrease in the density of the substructures accordingly. Pure thermal influence seemingly cannot induce additional distribution; instead, it favors carbide growth through carbon enrichment from the martensitic areas.

On average, the maximum carbon content was 0.098 ± 0.012 wt. % and was measured at a mean sample height of 4.19 ± 0.03 mm (Fig. 8). Since the actual deposition occurred at a height of 5 mm, the deviation from the theoretical build height is already apparent. This deviation is also reflected in the absolute sample height, which deviates on average by approximately 0.75–1.0 mm from the target value. This is attributed to the comparatively high volume energy density required for processing the powder material.²³ The powder is water atomized with irregular particle morphology, resulting in low bulk density. Consequently, remelting leads to a slight collapse of the layer and thus an offset from the target value. This is a well-known issue in additive manufacturing and is conventionally compensated for, although it was not done in this study. The region of elevated carbon content extended on an average, over a sample height of 0.82 mm, indicating an enlargement of the theoretical re-alloying zone (0.25 mm) (Fig. 8). Comparing the carbon gradient below and above the maximum, it is noted that the carbon content below the re-alloyed zone approaches the carbon content of the base material over a shorter distance than above the re-alloyed zone. In general, the remelting processes following the deposition result in a shallower gradient because carbon is carried toward the heat source by the high temperature in the melt pool. However, this effect occurs only in the directly subsequent layers that lead to complete melting; hence, no influence was observed in the millimeter range of the examined sample heights.

Also, the integration of secondary exposure in the form of varying laser powers of 200, 350, up to 500 W did not lead to a significant influence on both the carbon content and carbon distribution. Instead, the additional energy input initiated an in situ heat treatment of the microstructure, resulting in optical differences in the re-alloying zone (Fig. 9). For the initial state solely induced by primary exposure and for a secondary laser power of 200 W, no changes in the grain boundary-revealing substructures are apparent. Increasing the laser power resulted in the densification of fine structures in Figs. 9(c) and 9(d), with an almost complete overlay of the ferritic base structure. The substructures are no longer exclusively present at the grain boundaries. With the increase in laser power, the predominantly globular structures were increasingly complemented by needle-like structures.

Starting from an achieved increase in the local carbon content up to 0.1 wt. %, a stronger alteration of the microstructure was aimed for by increasing the amount of suspension per layer. Consequently, with five dosages per layer, a significant change in the locally re-alloyed area occurred. Figure 10 provides an overview of the re-alloyed zone, supplemented by EBSD investigations of two differently appearing microstructures. The overview of the re-alloyed microstructure in Fig. 10(a) allows classification into two different appearances, occurring alongside the microstructure of the base material visible in the lower region. Within the theoretically re-alloyed layer height of 250 μm, delineated by dashed lines, veil-like contrast differences prevail. Occasionally recognizable line formations indicate a fine structure without revealing specific grains. These are illustrated in the phase analysis in Fig. 10(b), showing a fine structure consisting predominantly of a cubic-centered structure, supplemented by areas of cubic face-centered retained austenite, and carbides, M₃C and M₇C₃. While carbides mainly form at the grain boundaries, austenite is found in an almost homogeneous distribution in the intermediate areas of predominantly elongated grains. SOES measurements yielded a carbon content of 0.204 ± 0.044 wt. %, resulting predominantly in martensitic solidification of the microstructure.

Due to the primary precipitation of chromium-rich carbides, the surrounding microstructure becomes depleted in chromium. With the low diffusion rate of the ferrite-stabilizing element, further carbide precipitation is prevented, and the remaining carbon remains interstitial, thereby increasing the stability of the austenite down to room temperature.

As the distance from the upper sample edge increases and exceeds the lower marking in Fig. 10(a), the brightness of the intricate area decreases until a cloud-like boundary to a needle-like structure appears. These needles can be attributed to the formation of Widmanstätten austenite. Since this is not stable at room temperature, the needles consist of a martensitic substrate. As described earlier, the initial re-alloyed layer results in a mixing with the layers of the base material below. Consequently, ferritic solidification still dominates below the theoretical re-alloyed zone. At the lower boundary of the solidification front, according to Karayagiz et al.,²⁴ the cooling rate is the lowest and is further heated to high temperatures by the subsequent layers. The longer time intervals allow for the growth of needle-like Widmanstätten austenite into the underlying ferritic grains. However, the austenite does not remain stable again until room temperature and transforms into the tetragonal-centered structure upon reaching the martensite start temperature. With the subsequent layers, the density of the needle-like structures increases until the gradual increase in carbon content exceeds a threshold, suppressing ferritic solidification. The fully austenitic microstructure then completely transforms into a fine structure of martensitic needles surrounded by carbides. The maximum carbon content achievable with five droplet repetitions is reached within the theoretical re-alloyed zone of 250 μm, resulting in a retained austenite content of 5%. This proportion of the cubic face-centered microstructure could be increased to up to 16% by adding secondary exposure of the same laser power (Fig. 11).

Such parameter dependence is also observed for duplex steels, where an increased austenite content is primarily associated with an increase in laser power or energy input. With the change in this parameter, lower cooling rates are associated, leading to both increased segregation of austenite-stabilizing elements and more time for the solid-phase transformation from δ-ferrite to austenite. In relation to secondary exposure, this does not necessarily lead to changed cooling conditions. According to Pfaff et al.,²⁵ for example, increased cooling rates only apply to melt tracks of low depth, as the influence of a thermally insulating powder layer is significant here. Against the background of the generally high volume energy input due to powder characteristics, however, a comparatively high melting depth can be assumed. Assuming comparable cooling rates, as a result of double exposure, improved homogenization of the microstructure is expected, as also demonstrated in the area of fine-scale changes in alloy composition (Fig. 9). Compared to the exclusively singly exposed state in Fig. 10, a significantly reduced gradual increase in carbon in the form of martensitic Widmanstätten needles is apparent.

The additional melt bath times lead to a more even distribution of carbon into the underlying ferritic zones. The homogeneous enrichment of the austenite stabilizer promotes complete transformation into the cubic face-centered lattice structure. This formation is further supported by the generally expected higher temperature level resulting from repeated energy input. Along the build direction, the heat-affected zones tend to reach higher temperatures, providing more time for austenitic transformation. Furthermore, generally higher melt pool temperatures can be achieved, potentially leading to the evaporation of the ferrite-stabilizing element chromium, and thus, an additional destabilization of ferrite. Carbides present sporadically in austenite serve as nucleation sites for the subsequently formed martensite. Upon reaching the martensite start temperature, transformation begins at those carbides present there. This grows needle-like through the austenite region (Fig. 11).

• By using Smart-Alloying, the carbon content can be significantly increased locally, and thus, the microstructure can be tailored by changing the chemical composition.

• Chromium-rich carbide compounds can be found along a main grain boundary. However, further analysis is required to determine the type or phase of these carbide compounds, which are also visible as small particles along the grain boundary in the HAADF.

• The precipitation of chromium-rich carbides binds the re-alloyed carbon locally, causing it to remain in the target areas regardless of the subsequent geometry.

• The use of double or secondary exposure with laser power at least equal to the base parameter leads to an increase in the melting bath times and temperature level, which improves the distribution of the re-alloyed carbon in the local area.

• The combination of Smart-Alloying and locally modified parameters leads to further modification of the local microstructural areas in order to produce customized components.

We are grateful to D. Knoop for his first fundamental feasibility studies in the development of the Smart-Alloying technology and to U. Fritsching and M. Maas for their support in the development of the used suspension. M. Rickers, F. Walter, D. Hallmann, and P. Meier are acknowledged for their continuous engagement in sample preparation, metallographic preparation, and analysis. Part of this work was performed at the DFG-funded Micro- and Nanoanalytics Facility (MNaF) of the University of Siegen (INST 221/131-1) utilizing its major TEM instrumentation (DFG INST 221/93-1, DFG INST 221/126-1) and sample preparation equipment. The authors would like to express their gratitude to Yilmaz Sakalli for supporting the TEM preparation and analysis. This work was supported by the Leibniz Association in the form of the Leibniz Junior Research Group “AURORA - Additive manufacturing of graded structures from iron-based shape memory alloys” (Project No. J119/2021) and by the German Research Foundation (DFG) (Project No. TO 1395/1-1).

The authors have no conflicts to disclose.

Anastasiya Toenjes: Conceptualization (lead); Funding acquisition (lead); Investigation (supporting); Methodology (equal); Project administration (lead); Resources (lead); Supervision (lead); Visualization (supporting); Writing – original draft (equal); Writing – review & editing (lead). Carolin Zinn: Data curation (supporting); Formal analysis (supporting); Investigation (supporting); Visualization (equal); Writing – original draft (equal); Writing – review & editing (supporting). Axel von Hehl: Formal analysis (supporting); Investigation (supporting); Resources (supporting); Writing – original draft (equal); Writing – review & editing (supporting). Hongcai Wang: Data curation (equal); Investigation (supporting); Visualization (equal). Kerstin Hantzsche: Data curation (equal); Investigation (supporting); Visualization (equal). Marcel Hesselmann: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (equal); Visualization (lead); Writing – original draft (lead); Writing – review & editing (supporting).

The data that support the findings of this study are available from the corresponding author upon reasonable request.