Controlling splat boundary network evolution towards the development of strong and ductile cold sprayed refractory metals: The role of powder characteristics

Feedstock powder characteristics such as composition (specific alloying elements and concentrations and impurity levels), microstructure, thickness/composition of surface oxide layers, and particle size distribution play a crucial role in determining the overall mechanical properties of cold sprayed deposits. Herein, we report on two deposits consolidated via cold spray processing from differently-sourced batches of nominally identical elemental refractory powders under identical spraying conditions, which exhibit bending strength and ductility values that differ by more than a factor of two – and with the weakest sample displaying negligible effective ductility. Through chemical, microstructural and micromechanical characterization of both the feedstock powders and cold sprayed deposits, we consider the possible influences of feedstock characteristics on the mechanical performance of cold spray consolidated deposits. It is shown that while differences in interstitial oxygen and hydrogen content may result in differences in the intrinsic yield characteristics of the feedstock material, both feedstocks maintain the ductile behavior required to induce good metallurgical bonding upon impact of optimally sized powder particles. We conclude that the deposits formed from the two feedstock powders are indistinguishable and exhibit high ductility when characterized locally within the relatively undeformed bulk of a single particle or splat. However, the two sprayed deposits show low ductility or brittle behavior when loaded in tension across intersplat boundary domains comprising material that has undergone extensive deformation. In addition, one feedstock incorporates a broader particle size distribution, with a long tail of larger-than-optimal particles. These larger particles are accelerated below the critical impact velocity, resulting in observed imperfect metallurgical bonding at splat interfaces and higher porosity at such intersplat boundaries. This result, combined with the overall differences in the morphology of the intersplat boundary network between the two deposits, explains the difference in macroscopic mechanical behavior. We conclude that accurate control of the powder feedstock particle size distribution is essential for optimizing the mechanical integrity of cold sprayed refractory elements and alloys.


Introduction
Cold spray (CS) is a solid-state additive manufacturing (AM) process based on the acceleration of micron-sized powders to supersonic velocities (300-1200 m/s) through a gas jet nozzle towards a substrate [1][2][3][4][5].The CS process utilizes kinetic energy to attain bonding and does not require melting of the feedstock powder, as opposed to other thermal spray and AM processes such as plasma spray (PS), high velocity oxy-fuel spray (HVOF), laser powder bed fusion (LPBF) and directed energy deposition (DED) [4][5][6].Upon impact, powder particles undergo severe plastic deformation, building a dense deposit on the substrate [1][2][3]5].High strain rate impact during cold spray causes particle flattening and out-flow of material around the outer edges of the particle, a phenomenon known as jetting [1,[6][7][8][9].Bonding of particles to the substrate and to each other occurs through mechanical interlocking and/or metallurgical bonding, the latter requiring clean metal contact at the atomic level [5,6,[10][11][12][13].The enormous shear strains generated by jetting allow for fracture and spallation of the oxide layer of the feedstock powder particles, thus enabling metal/metal contact [6][7][8].Jetting and bonding during cold spray can only be achieved if the particle velocity upon impact exceeds the critical impact velocity (v cr ) [1,2], which is a material-specific property.The particle impact velocity (v pi ) depends on both the processing conditions (such as gas temperature and pressure [14,15], standoff distance [16,17] and deposition gas type [5,17]) and the feedstock material characteristics (including materials density, particle size distribution (PSD) [15,[17][18][19][20], and powder morphology [19,21]).The critical velocity can be influenced by factors such as the concentration of interstitial inclusions, powder microstructure [12] and surface oxide layer properties [6,22].
Comparatively low process temperatures mitigate oxidation and formation of undesirable phases during deposition, and limit the development of thermal stresses [2,11,13,23,24], making the CS process particularly suitable for materials with high melting temperatures and materials intrinsically susceptible to cracking with deformation and thermal stresses, such as refractory metals.Refractory metals, including niobium, tungsten, tantalum, and molybdenum alloys, possess unique properties, including high strength, high temperature stability, and great corrosion and wear resistance [25][26][27], rendering them uniquely suitable for a range of applications involving extreme environments.Nonetheless, their high strength, severe susceptibility to oxidation, and high melting temperatures pose significant manufacturing challenges, resulting in high processing costs and low production rates [25][26][27][28].CS deposition has been shown to enable production of high-quality dense deposits of refractory metal powders [26][27][28][29][30].However, the mechanical properties of cold sprayed deposits are found to be highly variable and significantly affected by the microstructure, chemical composition, and particle size distribution of the powder feedstock, as well as the thickness, chemistry and microstructure of existing surface oxide layers on the powders [6,12,14,15,17,19,21,22].Understanding the underlying mechanisms of how these factors influence the properties of CS deposits is crucial for obtaining well-bonded interfaces and optimizing the resulting mechanical properties of the sprayed deposits and articles.In particular, the applicability of cold spray for deposition of wear and corrosion resistant coatings [24,31], repair of critical assets [32], and additive manufacturing (AM) of structural components [4,17,33,34] demands a fully mechanistic understanding of the relation between powder feedstock and cold sprayed deposit properties.While a few studies on cold spray of refractory metals have been carried out [21,29,[35][36][37][38][39][40][41][42][43][44], these complex relationships and correlations are still poorly understood.In this work we delve into the local deformation behavior within refractory (T m > 4000F) cold sprayed deposits.We specifically focus on the deformation mechanisms at the intersplat boundaries and within the bulk of the impacting particle splats, unveiling how these micro-level interactions influence the macro-scale mechanical properties of the cold spray deposited refractory material.Additionally, we establish an understanding of the critical particle size threshold necessary to attain a dense deposit of the refractory material, thus providing crucial insights for optimizing the CS deposition process for refractory metals.To that end, we show that macroscopic samples cold sprayed from nominally identical refractory metal powder batches (henceforth referred to as feedstock 1 and feedstock 2) deposited under identical cold spraying conditions may exhibit vastly different mechanical responses.While this observation is expected to apply to virtually any metallic system, here we demonstrate it with a model system consisting of a single-element refractory metal.Fig. 1 depicts 3-point bend test results for the two cold sprayed deposits, produced from feedstock 1 and feedstock 2 -henceforth referred to as deposit 1 and deposit 2. Two specimens were prepared and tested from each deposit.While deposit 2 showed good strength and ductility, deposit 1 failed in a brittle manner at ~40 % lower stress and ~55 % lower strain.In order to understand the underlying mechanisms responsible for this unexpected difference in mechanical behavior, we chemically, microstructurally and micromechanically investigated both feedstock powders and post-mortem 3-point bend samples.The objective of this study has been to explore, isolate and differentiate possible reasons for the wide variability in observed mechanical behavior for coatings obtained from spraying deposits from different feedstocks using identical spraying procedures, and to correlate the observed variability in mechanical behavior with the properties of the feedstock powders or the microstructure obtained after spraying.The comparative and cross-correlating experimental studies demonstrate conclusively that differences in the volumetric content and spatial distribution of the deposit microstructural constituents (i.e.differences in the splat boundary network and spatial arrangement of ductile, work-hardened, and unbonded features), resulting primarily from differences in the feedstock powder particle size distributions, give rise to the observed variability in macroscopic mechanical behavior of the consolidated deposits.These results are believed to be general to any cold-sprayed refractory element or alloy.

Microstructural characterization of powder particles
The chemical compositions of the feedstock powders were analyzed with the Inert Gas Fusion (IGF) method to assess the nature and concentrations of impurities (hydrogen, oxygen, etc.) present in the as-Fig.1. 3-point bend test results for deposit 1 and deposit 2, along with an image of a portion of the post-mortem specimen.
M. Amiri et al. supplied feedstocks.The particle size distribution (PSD) of the two powder feedstocks were measured utilizing a Microtrac MRB SYNC laser diffraction and dynamic image particle size analyzer.
Powder morphology and individual-particle grain structure were studied via scanning electron microscopy (SEM) using secondary electron (SE) and backscattered electron (BSE) imaging modes on a FEI Magellan 400XHR SEM (Thermo Fisher Scientific Inc.).Powders were embedded in a conductive resin/epoxy mount (Struers EpoFix + 20 wt% graphite flakes) and were polished with diamond suspensions (Allied High Tech Products Inc.) down to 1 μm, and final-polished with 0.05 μm colloidal silica for BSE analysis.The crystal structures of powders were characterized by X-ray diffraction (XRD).XRD was performed on a Rigaku Ultima III X-ray diffractometer, operated at 40 kV and 30 mA with Cu K α radiation.High resolution transmission electron microscopy (HRTEM) was utilized for powder particle surface oxide analysis.Thin lamellae comprising the passivating surface oxides and near-surface region of the metal particles were prepared by lift-out from single particles using focused ion beam (FIB) milling on a FEI Quanta 3D FEG dual beam SEM/FIB (Thermo Fisher Scientific Inc.).Particles were dispersed on carbon tape attached to the top of an SEM stub.A thin carbon protective layer was deposited by permanent marker to improve contrast during TEM imaging, and to protect specimen surfaces during subsequent platinum deposition.Platinum protective layers were deposited on top of the carbon film, first using an electron beam deposition followed by an ion beam deposition.

Cold spray deposition
CS deposits of both pure elemental refractory powders (feedstock 1 and feedstock 2) were produced at Solvus Global using a high-pressure cold spray system (Gen III Max, VRC Metal Systems).A spraying pressure of 3.79 MPa (~550 psi, 38 bar) and a temperature (at the applicator) of 610 • C were used.The main powder accelerating gas flow through the nozzle was 1150 SLM (standard liter per minute), powder carrier gas flow 120 SLM, nozzle traverse speed 500 mm/s, and nozzle standoff distance 20 mm.

Microstructural characterization of deposits
Microstructures of deposit cross-sections were characterized by scanning electron microscopy (using both SE and BSE imaging modalities) on a FEI Magellan 400XHR SEM (Thermo Fisher Scientific Inc.).CS deposits were first cut using a diamond saw (Buehler) and were then mounted in a conductive resin/epoxy mount (Buehler EpoxiCure2 + 20 wt% graphite flakes).Cross-sections were subsequently prepared by grinding and polishing with SiC abrasive papers and with diamond solutions (DiaDuo-2, Struers Inc.), starting with 6 μm and progressing down to ¼ μm for the diamond polishing.Cross-sections were finally polished with 0.05 μm colloidal silica suspension (Allied High Tech Products Inc.) prior to SE and BSE analysis.Additionally, thin lamellae from regions containing intersplat boundary domains were extracted from the as-received deposits via FIB milling.These lamellae were investigated utilizing high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) to analyze the microstructural features and the grain structure formed during CS deposition.
A robust measurement of residual porosity was made utilizing crosssectional images of the deposit microstructure.BSE images from randomly selected areas, captured at length scales that elucidate the relevant porosity (in this case, images with a horizontal field of view of 150 μm), were used and processed through a MATLAB image filtering and analysis routine.Each image, with 150 μm HFW represented an area of ~0.016 mm 2 .A contrast threshold along with a suite of image filtering and pixel-based quantitative image analysis algorithms were applied to each image in order to extract the total porosity.
To investigate differences in the morphology of the splat boundary networks in deposits produced from the two feedstock powders, a quantitative approach (akin to the line-intercept method used for grain size analysis) was utilized to evaluate two microstructural aspects: (1) the distances between pockets of relatively less deformed material originating from a single impacting particle and largely retaining the grain microstructures of the feedstock powder particles; and (2) the distance between intersplat boundary crossings, where extensive grain refinement is observed.Both measurements were obtained by drawing a set of randomly oriented lines across the SEM images (keeping the set of lines and the imaged field of view constant across the microstructures evaluated) and quantifying the distances between undeformed pockets and between splat boundary interfaces.With the set of lines used with each region imaged and analyzed, a total intercepting line length of ~25 mm was analyzed for each deposit material (representing over 200 feature intercepts for each deposit).

Mechanical characterization of deposits and powder particles
Nano-indentation was performed on powder particles mounted in a conductive resin/epoxy.Powder cross-sections were polished with the same procedure as described in section 2.1.Nano-indentation tests were carried out in an Agilent G200 nanoindenter with a Berkovich tip and a maximum load of 20 mN with a 10 s hold time at maximum load, and a strain rate of 0.1 mN/s.Samples were unloaded to 10 % of the maximum load and held for thermal drift correction.For each feedstock, 60 measurements (on 60 different powder particles) were performed.An algorithm incorporated within the built-in software was used to calculate hardness values based on the Oliver-Pharr [45] method.
Wedge indentation tests for characterizing the deformation and cracking of deposits [46] were conducted on an Instron 5985 load frame equipped with a 250 kN load cell.During this test, a WC wedge indenter with an included angle of 90 • was used to apply load at a constant displacement-rate of 0.001 s − 1 .
Micro-mechanical tests were performed on individual powder particles, as well as samples extracted from specific locations within the deposits, to analyze the intrinsic feedstock powder properties and local mechanical properties of the deposits.Micro-tensile specimens with an aspect ratio of ~2:1 were prepared from individual particles of the powders via a semi-automated FIB lathe milling technique on a FEI Quanta 3D FEG dual beam FIB/SEM (Thermo Fisher Scientific Inc.).The technique was initially developed by Uchic and Dimiduk [47] and allows for creation of taper-free specimens to obtain a homogeneous stress state along the gauge length [48], which otherwise results in mis-estimation of the properties [49].Additionally, micro-tensile specimens with an aspect ratio of ~2.5:1 and micro-compression pillar specimens with an aspect ratio of ~2:1 were created from the two cold sprayed deposits using the FIB lathe milling technique referenced above (see Video S.1 for the fabrication procedure).The specimens extracted from the deposits were oriented such that the loading direction was aligned along the uniaxial strain direction under macroscopic 3-point bending, i.e. perpendicular to the particle impact direction.Two specimens were prepared from each of the two deposits to investigate the mechanical behavior at locations including an intersecting intersplat boundary and locations within the relatively less deformed core of a single splat.Before the micro-mechanical specimen extraction, powder particles and the deposits were mounted in conductive resin/epoxy, and ground and polished from two perpendicular sides with the same polishing procedure described in section 2.3.All micro-mechanical tests were carried out using the FemtoTools FT-NMT03 nanomechanical testing system in displacement control mode to obtain a constant strain rate of 10 − 3 s − 1 .
Microforce sensing probes with ±200,000 μN force range were FIB machined into either a flat punch with 25 × 20 μm cross-section or a gripper with ~7 μm opening and ~55 • grip angle for compression and tension tests, respectively.SEM images of the flat punch and gripper structures are shown in Fig. 2 alongside examples of prepared compression and tension specimens.A custom MATLAB script was used to accurately assess the strain values from the displacements of two Pt fiducial markers deposited on the sample surfaces at the top and bottom of the specimen gauge sections, as measured from SEM images captured during the micromechanical tests.
Fracture surface analysis of the post-mortem mechanical test samples was performed on a FEI Magellan 400XHR SEM (Thermo Fisher Scientific Inc.).Scanning transmission electron microscopy (STEM) was employed to understand the microstructural features of the micromechanically tested samples and to analyze correlations between the micro/nano-structure (grain structure, presence of intersplat boundaries, highly deformed and/or recrystallized regions around the boundaries) and the micro-mechanical stress-strain behavior.HRTEM on the as-received powders and STEM imaging on the as-received deposits and post-mortem mechanical specimens were performed on the JEOL JEM-2800 TEM operated at 200 kV.

Powder analysis
Results of the IGF chemical analysis and particle size distribution (PSD) measurements obtained from the two feedstocks are shown in Fig. 3. Feedstock 1 had an average particle size of 19.64 ± 10.2 μm, and a wider distribution with a tail of particles as large as ~60 μm, whereas feedstock 2 had an average particle size of 12.87 ± 6.26 μm, and no particles larger than ~35 μm.While feedstocks 1 and 2 had similar concentrations of oxygen (0.036 and 0.045 wt%, respectively), feedstock 1 was ~7 times richer in hydrogen (0.0058 wt%) than feedstock 2  (0.0008 wt%).
The morphology and grain structure of the powders are displayed in Fig. 4(a) and (b).Both powders had similar irregularly shaped morphology and equiaxed grain structures, albeit at different average particle sizes.HRTEM studies of the surface oxide layers in Fig. 4(c) revealed the presence of an amorphous oxide layer on the surface for both powders and nearly identical oxide thicknesses.The larger average particle size of feedstock 1 resulted in smaller surface to volume ratio, and hence a smaller fraction of total oxygen in the oxide layer, relative to feedstock 2. Subtracting the concentration of oxygen bonded in the oxide layer from the total amount measured by IGF (see Appendix for details) revealed that the two feedstocks had nearly the same concentrations of interstitial oxygen in the metal, viz.0.026 wt% for feedstock 1 and 0.031 wt% for feedstock 2. This small difference in interstitial oxygen content is not expected to significantly affect mechanical properties.XRD analysis obtained from the two feedstock powders revealed identical BCC crystal structures, with no measurable evidence of hydrides or oxides and no significant difference in the lattice parameters due to extensive incorporation of interstitial impurities.
The results of the characterization of the feedstock powders in the assupplied condition showed that the two powder feedstocks were nearly identical in terms of the key microstructural, morphological, and chemical composition characteristics, differing exclusively in particle size distributions and hydrogen content.
Previous studies have reported mixed data on the effects of PSD on properties of CS deposits [15,[17][18][19][20]. Zahiri et al. [17] reported that smaller average particle size and a narrower size distribution would result in higher impact velocity and a smaller volume fraction of residual porosity in CS titanium.Conversely, Marrocco et al. [18] analyzed the same material and reported that powders with larger particle sizes lead to lower porosity, an effect that was attributed to the peening effect from larger particles.In this case of PSDs with a larger average particle size and/or broader distribution, a larger proportion of particles will not reach the critical velocity, but carry significant momentum.The resulting non-bonding particles then contribute to a higher intensity peening, which results in lower porosity.Cinca et al. [15] observed more homogenous structure and properties for coatings produced from powders with narrower PSDs, but drew no clear relation between the average particle size and porosity percentage.Wong et al. [19] reported a mixed trend between PSD and porosity and microhardness.They also indicated that porosity, microhardness, and the ratio of particle impact velocity to critical impact velocity (v pi /v cr ) vary more significantly with powder morphology than with the PSD.
Hydrogen can change the mechanical properties of refractory elements and is well known to cause embrittlement in the refractory element systems through different mechanisms, mainly hydride formation and hydrogen enhanced decohesion (HEDE), and can deteriorate their mechanical properties [50,51].However, the effect of hydrogen on the mechanical properties of polycrystalline refractory metals and alloys depends upon the H concentration, loading conditions and strain rate, temperature, and presence of other elements [50][51][52].A study on the effects of H interstitials on yield stress and ductility of refractory elements vs. temperature showed that the presence of H (up to 0.0078 wt%) did not appreciably change the yield stress at room temperature [51].H strengthening occurred only below about − 23C (150 K).A decrease in ductility, however, was observed at room temperature and ductility continued to decrease dramatically as the temperature was lowered to − 48 C (230 K).As the temperature decreased further, ductility was observed to begin to recover [51].In the presence of other interstitials (N, C and O) with H, the contributions were in general additive, resulting in a further decrease of ductility and further increase of yield strength, with the same trend vs. temperature [51].
To further analyze the properties of the feedstock powder and their implications on the mechanical behavior of the cold sprayed deposits, nano-indentation and micro-tensile tests were performed on individual powder particles.Hardness values from nanoindentation showed ~10 % higher hardness for feedstock 1 (1.19 ± 0.44 GPa) than for feedstock 2 (1.09 ± 0.4 GPa), which might be attributed to the difference in the interstitial hydrogen concentrations in the two feedstocks.
Fig. 5(a) and (b) show uniaxial stress-strain curves from microtensile tests of specimens extracted from the individual powder particles as well as SEM images of the specimens at the end of the tests.Notably, both feedstock 1 and feedstock 2 powders showed very ductile behavior and withstood strains up to ~0.75 and 0.9, respectively.While the flow stresses in the two feedstock powder samples are quite different, it is important to note that while the micro-tensile specimens are polycrystalline, the sample diameter (5 μm) is only slightly larger than the average grain size (2-3 μm); as a result, we would expect significant variations in the flow stress with orientation of the individual grain(s) in which the dominant deformation and sample failure occurs.Both feedstocks exhibited flow stresses within the range of literature data for single crystals of the same refractory element at various orientations, suggesting that the difference in flow stresses observed during the micro-mechanical testing is more likely attributed to difference in grain structure and orientation of dominant slip systems relative to the loading direction, rather than to intrinsic differences in the properties of the feedstock powders.It has been reported that the orientation of the slip system relative to the loading direction can result in a factor two change in the yield strength [53,54].Despite this, the different strength values observed here compared to the reported values in other studies [53,54] could be partly attributed to differences in the interstitial elements concentrations.However, the observations of high ductility in testing samples from both feedstocks (Fig. 5) supports the key takeaway that differences in interstitial hydrogen content is not the primary contributor to the poorer mechanical performance of the deposit produced from feedstock 1 powder.
Fracture surfaces of the two samples, shown in Fig. 5(c), exhibited similar features, with both feedstocks showing that parallel slip systems were activated during plastic deformation.The traces of slip lines, which were clearly observed on the surface of both specimens (indicated with arrows in Fig. 5(c)) are consistent with previous studies on single crystals and large-grain polycrystals of ductile refractory alloys [55], confirming the important effect of dominant slip system on the strength.Videos of deformation and stress-strain curve evolution during the micro-tensile tests of the two powder specimens can be accessed in the supplementary materials, Video S.2 and Video S.3.

Microstructural analysis of the as-processed deposits
Fig. 6 displays microstructural features for the two CS deposits obtained by BSE and HAADF-STEM analysis.An inhomogeneous microstructure and the presence of pores and unbonded intersplat boundaries were observed in both deposits, present to a much greater extent in deposit 1.In both cases, the microstructure was comprised of highly deformed areas featuring elongated and fine recrystallized grains near the intersplat boundaries, along with areas with less deformed coarse grain structures farther from the boundaries in the relatively undeformed interiors of individual impacting particle splats.Energy dispersive spectroscopy (STEM-EDS) revealed measurable oxygen content at certain boundaries within the deposit, as shown in Fig. 6(b), demonstrating the presence of residual oxides in those regions.A comparison of the microstructure of the two deposits in Fig. 6 also reveals that deposit 1 had larger areas with coarse-grained structure of low deformation, as well as larger individual pockets of porosity present at intersplat boundary intersections than deposit 2. Deposit 2 exhibited a higher volumetric density of domains with highly refined grains.
The results of the assessments of residual porosity are shown in the box and whisker plots of Fig. 7 point in the box and whisker plot, where ~90 regions were measured and analyzed for each of the deposit materials (totaling to about 1.5 mm 2 for each sample).It should be noted that (assuming average particle sizes of 13-20 μm), a total image area of 1.5 mm 2 for each sample accounts for an assessment of residual porosity resulting from well over 5000 particle impacts per feedstock materials.The results of the porosity measurements reveal that deposit 1 had both higher average porosity as well as a much less uniform porosity content distribution throughout the Differences in the morphology of the splat boundary networks in deposits produced from the two feedstock powders were also revealed using the quantitative line-intercept procedures outlined in Section 2.3.Two microstructural characteristics were evaluated to quantify differences in the structure and morphology of the splat boundary networks: (1) the distances between large, relatively undeformed pockets (representing the parts of impacting particles that do not undergo extensive plasticity and grain refinement); and (2) the distances between intersplat boundary interfaces (located within domains of extensive grain refinement).Important differences are observed in the distances between these two key microstructural features believed to play a  controlling role in the overall mechanical response of the two deposits.The distances between large, relatively undeformed pockets are quantified in Fig. 8(a), and the distances between intersplat boundary interfaces are quantified in Fig. 8(b).Significant differences are observed in the morphology and volumetric density of the splat boundary networks of the two deposits, revealing deposit 1 consists of a higher fraction of relatively undeformed material compared to deposit 2.An additional key observation is illustrated in Fig. 9, where it is shown that large particles within the deposit formed from feedstock 1 -with the broader PSDare correlated with large, unbonded splat boundary interfaces (Fig. 9(a)).Such microstructural features are absent in the sample formed from feedstock 2 -with the narrower PSD (Fig. 9(b)).Also, the uniformity of plasticity and bonding quality varies even within the same deposit, and is notably lacking around larger particles.Metallurgical bonding is effective adjacent to smaller particles within the same deposit, while larger particles are more frequently associated with the presence of cracks and pores (Fig. 9).In comparing the local microstructures within the two deposits, it is noted that the microstructural development (particle flattening ratio, residual porosity, and integrity of intersplat bonding) is indistinguishable in domains derived from impact of equivalent sized particles; the notable differences are associated with domains deriving from the incorporation of the larger particles mainly present in the PSD of feedstock 1.

Micro-mechanical characterization of the deposits
To characterize the micro-mechanical response of the deposits both within a single splat and across intersplat boundaries, two locations of interest were identified from each deposit for extraction of micromechanical test specimens, as depicted in Fig. 10.The specimens extracted from inside splats of deposit 1 and 2 are referred to as deposit1-i and deposit2-i, respectively, and the specimens with intersecting boundaries are denoted as deposit1-b and deposit2-b.
Fig. 11(a) and (b) show stress-strain curves and SEM micrographs for the micro-compression tests of the two deposits, both within and across splats.Formation of shear bands was observed in all four cases.Pillars of both deposits which contained intersplat boundaries exhibited high strengths.This confirms that a well-bonded boundary, as chosen for these micro-compression tests, is not a weak region in the deposit (at least when loaded under compression), an observation that can be attributed to the highly deformed fine-grained microstructure formed along the boundary during the CS process, as observed in both BSE-SEM and STEM studies of the deposits.
The micro-compression tests did not reveal any hydrogen embrittlement in deposits created from either of the feedstocks, with the pillars made from deposit 1 (with higher H content) deforming in a ductile manner up to high strains.The micro-pillar of deposit 2 cutting through an intersplat boundary (Fig. 11(b)) failed early at strains around 0.15, along the boundary, followed by shearing of the top part of the pillar towards the back.After the onset of top shearing, further deformation was primarily localized in the separated top part of the specimen.The micro-pillar of deposit 1 cutting through a boundary, by contrast, showed a uniform strain-hardening behavior.The boundary region in this specimen was well-bonded and had not extended across the crosssection.This specimen initially formed plastic slip bands near the top of the pillar and eventually started to fail along the boundary at a strain ~0.4.Intersplat boundary locations for both specimens are shown in Fig. S1.Overall, both deposits showed ductile compressive response, both within single splats as well as across splat boundaries.
Fig. 11(c) and (d) display micro-tensile stress-strain curves and SEM micrographs of the specimens after deformation for the two deposits.Micro-tensile specimens initially failed at low strain values, below 10 percent, indicating a less ductile behavior in tension than in compression for both deposits.A less ductile behavior in tension vs compression for Fig. 9.A comparison of key microstructural aspects originating with the different feedstock powders.For the specimen with the broader PSD and largest powder particles (Deposit 1), evidence is observed of poor particle bonding and retention of large unbonded splat boundary domains.These features are absent in the specimens deriving from the narrower PSD and smaller maximum particle size (i.e., Deposit 2).cold sprayed systems was also reported by He et al. [56].In tension, after samples developed initial damage, parts of the material that were still connected started to undergo more extensive deformation, which resulted in larger elongations before final fracture/separation of the two sample pieces.
While the behavior of the two deposits, when characterized locally within a single splat, was generally ductile, the samples containing an intersplat boundary (with this domain extensively work-hardened) showed either low ductility or brittle failure through that boundary.A specific trend between the hydrogen content and micro-tensile response of the deposits could not be inferred.The properties varied based on the location of the samples in the deposits in relation to the splat boundary network and characteristics of the intersplat boundaries and adjacent deformed/recrystallized regionswith the specimens crossing an intersplat boundary generally showing low ductility or brittle behavior.Variations in the local mechanical properties are attributed to the grain structure and morphology of the defects at the intersplat boundaries, including the lack of full metallurgical bonding, trapped oxides, and the presence of residual pores and cracks.Consistent with this finding, defect-driven premature failures in cold sprayed deposits have also been reported for other material systems [56].
Initial BSE-SEM and STEM studies of the deposits (Fig. 6) revealed that the highly deformed fine-grained microstructure extended across each individual splat for the case of deposit 2, whereas for deposit 1, areas with larger and less deformed grains are seen inside the splats.The higher volumetric content of regions that had undergone extensive deformation (and effective work hardening) may have contributed to the higher yield strength of deposit 2.
One should note that, considering the high degree of plastic deformation experienced during the cold spray process and the presence of so many splat boundaries in the deposits, it is possible that splat boundaries might have been included within the samples that were nominally taken from within a single splat.Better understanding of this matter will be obtained through subsequent STEM and HRTEM studies of the tested specimens, as well as the fracture surface analysis under SEM.
For the videos of deformation and stress-strain curve evolution during micro-compression and micro-tension tests refer to the supplementary materials, Videos S.4-S.11.Snapshots of deformation phenomena and corresponding strain values for compression and tension are also included in Fig. S1 and Fig. S2.
Fracture surfaces of post-mortem micro-tensile specimens were characterized by SEM and the results are illustrated in Fig. 12. Fracture surface analysis showed that for the specimens with a crossing boundary, fracture initiated and progressed along the boundary and then continued along a path directed towards the inside the splat.Failure in deposit1-b initiated in the finer grain region closer to the boundary and continued towards the coarser grain region farther from the boundary (Fig. 12(a)).The fracture initiation site and some open areas along the boundary are shown in Fig. 12(a) and (b).The boundary region in a cold sprayed sample usually shows lower ductility due to the existence of the highly deformed (and work-hardened) structure with a high density of dislocations prior to the mechanical tests as well as the presence of defects along the boundaries, compared to the coarse grain area inside the bulk of the splat [57,58].It can be observed that, for this sample, fracture initiated at a defect in the intersplat boundary and propagated towards the inside of the splat, where further deformation took place and allowed for the material to exhibit higher elongation and more extensive ductility.The extent of the ductility depends on the cross-sectional fractional area of the relatively undeformed bulk of the splat relative to the heavily deformed and recrystallized boundary area.The fracture surface of deposit1-i (Fig. 12(c) and (d)) was similar to that observed for individual powder particles in Fig. 5(c).Parallel slip systems and slip steps on the surface can be clearly distinguished.
The presence of some undeformed inclusions and unbonded boundaries were observed on the fracture surface of the locally extracted sample from deposit2-b (Fig. 12(e) and (f)), which led to the early failure of this specimen.Several areas of non-bonding (shown with dashed lines in Fig. 12(e)) were detectable on the fracture surface which suggested several particles had come together at this location, similar to a triple junction, resulting in a weak region.In such a region, factors such as retained oxides or included pore space might cause a premature failure, which can be further studied via STEM and HRTEM.The fracture surface of deposit2-i consisted of a coarse grain structure region and an elongated and fine grain structure region shown in Fig. 12(g) and (h).This elongated grain structure is a characteristic of intersplat boundary areas.This suggested that this specimen was not extracted from only inside the splat bulk but also had cut through one or more boundary domains.The bulk of the splats in deposit 2 also consisted of larger grains compared to its boundary areas.However, the overall splat grain structure for deposit 2 was still smaller than deposit 1 and grains were generally more deformed.Also, due to the small size of the splat pockets, the fabricated specimen appeared to have included boundary regions.Fracture initiated from the right side of the sample shown by arrows in Fig. 12(g).Parallel bands emerged on the surface along the gauge length on the side with coarse grain structure (Fig. 12(h)), which indicated that plastic deformation happened by a dominant slip system and parallel systems were activated in the coarse grain region, similar to what was seen for the tensile tests of individual powder particles illustrated in Fig. 5. On the other hand, the presence of the fine grain structure in this specimen contributed to its high strength.Highly deformed fine and elongated grains along a well-bonded boundary increases the strength due to the high dislocation density and strain hardening in CS deposits, as seen in this specimen.The relatively higher ductility of this specimen compared to the other two specimens with boundaries can be attributed to both a well-bonded boundary and presence of the coarse-grain region (lacking extensive prior work hardening), which withstood further strain after the boundary region failed.
Overall, some samples exhibited brittle fracture when the boundary intersecting the gauge length lacked sufficient bonding and displayed cracks or openings along the interface, as observed in the instances of the deposit1-b and deposit2-b samples.In other instances, where deformation predominantly took place within the bulk of the splat, the samples underwent ductile fracture, with necking occurring prior to fracture.Additionally, the plasticity associated with particle impact and severe deformation at the splat boundaries during CS processing leads to extreme work hardening in these regionsgiving rise to less ductile behavior at these boundaries than in the bulk of undeformed material [57,58].This is consistent with the micromechanical testing results in Fig. 11.
Microstructural analysis of the tested micromechanical test samples was carried out via STEM/HRTEM examination (Fig. 13).The intent was to correlate and contrast the observed failure mechanisms and the evolution of plastic deformation with the microstructural differences and the intersplat boundary network characteristics.Thin lamellae were prepared from the post-mortem micro-tensile test specimens and were characterized via STEM and TEM.
The microstructure of the deposit1-i specimen (Fig. 13(a)) consisted of a few large grains with a low degree of plastic deformation, which resulted in the ductile behavior and lower strength observed for this specimen during the tensile deformation.The behavior of this specimen was found to be comparable with that of the specimens extracted from the individual powder particles examined in Fig. 5. Convergent beam electron diffraction (CBED) patterns from three locations near the fracture surface were collected, as shown in Fig. 13(a).The results showed the same crystal orientation for all three locations, indicating one large grain was located at the fracture surface and was mainly responsible for the ductile fracture during tension.Due to the large area imaged (several microns in each axis) and the fact that lamellae were produced by FIB milling, lamellae bending and thickness variations were inevitable, and more manifest in larger micron-sized grains as was observed with the deposit1-i specimen.
The STEM study of deposit2-i reveals that this specimen consisted of Elongated and recrystallized grain zones close to the boundary region as well as coarser grain areas farther from the boundaries are distinguishable in Fig. 13(b).The observed boundaries in this specimen appeared as well-bonded interfaces with no signs of cracks or large pores.STEM analysis of the post-mortem specimen of deposit2-b (Fig. 13  (c)) revealed that an intersplat boundary passed across the cross-section of the specimen nearly perpendicular to the direction of loading.The trace of this boundary is highlighted by a dashed line in Fig. 13(c).A higher magnification analysis of the area closer to the fracture surface revealed the presence of some openings (unbonded regions within the microstructure) along the intersplat boundary that were extended to the fracture surface (region inside the black box in Fig. 13(c)).STEM-EDS maps were collected from these open areas that led to the fracture surface.EDS maps showed the presence of oxygen all along the openings, indicating the presence of retained oxide.This denotes that these locations were unbonded intersplat boundaries within the microstructure, with retained or natively formed oxides.The presence of a poorly bonded intersplat boundary oriented normal to the loading direction is inferred to have resulted in the observed elastic fracture with negligible plastic deformation during the tensile loading.

Macro-mechanical indentation testing and damage evolution
The macro-mechanical behavior and damage response implications of using the differing feedstocks, and correlation of these mechanical behavior differences to the specific microstructural differences in the two deposits was assessed via a wedge indentation technique.SEM (BSE) images were collected from the samples subjected to wedge indentation testing, before and after testing, which allowed for isolation and tracking of localized microstructural features and assessment of how these features respond under the induced deformation.The wedge indentation test setup and a schematic of the test method is shown in Fig. 14.
The two samples exhibited notably different damage responses, as summarized in Fig. 15.Under wedge indentation testing, deposit 1 did not uniformly accommodate the deformation and cracked through the thickness at a smaller wedge displacement compared to the deposit 2 sample.Fig. 15 shows the formation of regions within the microstructure of deposit 1 with highly localized deformation upon wedge indentation, and the opening of large voids at locations that presumably had unbonded or poorly bonded splat boundaries.These locally deformed regions and open voids generally occurred adjacent to the large, undeformed domains from the impact of the larger particles in the broader PSD.This behavior was generally absent in testing deposit 2, where the overall deformation was observed to be more uniform, and free of such heterogeneous zones of damage development and void formation.It is clear that the microstructure of deposit 2, which consisted of a closer packed network of nanograined boundaries, is responsible for the ability of this deposit to accommodate deformation (Fig. 15).In deposit 1, the relative inability to accommodate deformation within the microstructure, and the existence of large unbonded sections of splat boundaries, resulted in the formation of large defects and cracks that may link up and lead to overall fracture at lower bending strengths, as observed in Fig. 1.It can be seen in Fig. 15 that the large pockets of low deformation may act as crack initiation sites in deposit 1.
The fracture surfaces of the two samples were also analyzed, and further highlighted the microstructural differences between the two samples (Fig. 16).The fracture surface of deposit 1 showed a larger percentage of smooth surfaces corresponding to particle-particle boundaries of large particles that experienced little deformation during cold spray and some areas of localized deformation or opened cracks that happened during the wedge indentation.Conversely, the fracture surface of deposit 2 very clearly consists of smaller particles, which experienced a more uniform deformation and fracture upon wedge indentation.
The wedge indentation results further support the conclusion that poor bonding around the large particles (that likely did not reach the critical impact velocity during CS deposition) is responsible for the early failure and generally poorer mechanical behavior observed with deposit 1.
In summary, the significant difference in macroscopic and microscopic mechanical properties of the two deposits is clearly attributed to the deposit microstructure, intersplat boundary network morphology and attendant quality of bonding (presence of defects) at the intersplat boundaries.A well bonded boundary (as in the case for deposit2-i) results in high strength values due to the highly deformed and workhardened fine grain structure that is formed along the boundary region upon CS particle impact.On the other hand, a boundary in the same deposit with similar fine grain structure which is not well bonded and contains defects along the boundary can cause failure at very low strains and even lead to premature failure as in the case for specimen deposit2b.Nautiyal et al. [59] also found that intersplat properties are the important factors defining the modes of failure (ductile vs. brittle) in cold sprayed deposits.It is indeed the larger length scale differences in the volumetric content and distributions of the boundary regions (and associated recrystallized regions and defects) that lead to the macroscopic testing result differences.It is important now to investigate the mechanisms that have resulted in the microstructural and defect discrepancies between the two deposits.

Analysis of the effect of PSD variations
It is well known that PSD of the feedstock powders plays a role on the extent of deformation and bonding that takes place during the cold spray process [15,[17][18][19].It is surmised that the different microstructural features formed in the two deposits are related to the differences in the particle size distributions among the two feedstock powders.
When a particle is accelerated towards the substrate, its impact velocity (v pi ) can be expressed as a function of the particle size, according to Ref. [60]: where d p is the particle size (diameter), ρ p and ρ 0 are the particle and gas density, respectively, δ = 0.0007 m is the characteristic thickness of the bow shock boundary layer, f = 1.09 is a calibration parameter, v max represents the highest achievable particle velocity for a finite gas pressure and infinitely large gas temperature, and v 1 is the approximated gas velocity which depends on the nozzle properties.For details, the reader is referred to the original paper by Assadi et al. [60].The particle bonds to the substrate if its impact velocity v pi exceeds a critical impact velocity, v cr .While different models can be used to estimate v cr , Assadi et al. [60] suggest: As v cr is expected to depend on the particle strength [7], and our experimental results (Fig. 5) display differences in strength between the feedstocks of up to 200 MPa (an amount which likely intertwines differences in H content and crystallographic orientations of the oligo-crystals), we estimate a maximum uncertainty in v cr of ±20 m/s [7].
The PSD of both feedstock 1 and 2 powders along with the calculated particle impact velocity (v pi ) and critical impact velocity (v cr ) as a function of particle size are shown in Fig. 17, with the ±20 m/s uncertainty on v cr illustrated as a band.While v cr does represent the threshold velocity above which bonding will take place, particles accelerated to velocities in the neighborhood of v cr will often result in low quality (poor) bonding, as schematically indicated in Fig. 17.When superimposing these boundaries to the specific PSD of the two feedstock powders in this study, it becomes apparent that feedstock 1 had a significantly larger fraction of particles that could result in poor bonding or rebound without bonding.This would lead to a higher porosity, a higher density of unbonded intersplat boundaries and an overall larger defect networkresults consistent with the microstructural analysis performed above (Figs.6-9).Also, it should be noted that larger particles that achieve the critical impact velocity during deposit buildup will still have a smaller fraction of their particle volume undergoing deformation and recrystallization only near the particle surfaces (in regions that become intersplat boundaries) (Figs. 6, 8 and 9).Importantly, all these conclusions hold true regardless of the potential difference in strength related to the difference in H content between the two feedstocks.This clearly confirms that the observed differences in H content between the two deposits, albeit significant, are not responsible for the dramatic difference in mechanical response.Furthermore, in the case of wider distributions like the one exhibited by feedstock 1, there appears to be different competing mechanisms at play for the larger, slower particles that can result in either (a) poor bonding and therefore a potential source of porosity, as observed in this study and some previous works [17,20], (b) particles rebounding the surface and having a shot peening effect on the deposit, increasing deformation, and decreasing porosity [18], and/or (c) particles rebounding the surface and interacting with the incoming particles, causing a reduction in their impact velocities [5].These phenomena, and the circumstances enabling one to dominate over the other ones  must be explored further to fully characterize the effect of the specific PSD of the feedstock powders on the resulting cold spray deposit.

Conclusions
Cold sprayed deposits produced from two feedstock powders from a commercially pure refractory element, processed under identical spray processing parameters, showed significantly different macromechanical responses.Initial analysis of the powders showed they both had similar microstructures and similar surface oxide layers but different hydrogen concentrations and particle size distributions.Micromechanical and microstructural characterizations of the two feedstock powders and the CS deposits produced from them showed no significant effect of hydrogen content, instead clearly attributing the vastly different macro-mechanical response of the deposits to significant differences in deposit microstructure and defects: i.e., the deposit made from the powder with larger average particle size and (more importantly) a wider particle size distribution exhibited a higher percentage of porosity with larger pores and more unbonded or poorly-bonded splat interfaces.We attribute these differences to the dependence of particle impact velocity and critical impact velocity on particle size, whereby larger particles fail to achieve the critical impact velocity and highquality bonding: feedstocks with larger particle size distribution tails result in a larger network of poorly bonded intersplat boundaries, which are responsible for limited ductility and premature failure upon loading.
Overall, a set of independent experimental approaches have been employed to characterize differences in the characteristics of both the powder feedstocks prior to cold-spray deposition and the components of the resulting heterogeneous microstructures of the consolidated materials after spray deposition.The microstructural observations and measurements and the mechanical test results provide a comprehensive set of cross-validating results illustrating the key features that correlate with important differences in the mechanical performance of the resulting refractory element cold spray deposits.Collectively, all the micromechanical and microstructural investigations conducted in this study present compelling experimental evidence in support of the final Fig. 17.Effect of the particle size distribution (PSD) on impact velocity and bonding quality.PSD for (a) feedstock 1 and (b) feedstock 2 along with the calculated impact velocity and critical impact velocity curves, and highlighted ranges for bonding and rebounding are shown.The grey band captures uncertainties on the critical impact velocity, related to potential strength differences between the deposits.conclusion of the study: that the difference in particle size distribution between the two deposits resulted in significant differences in their splat boundary networks, which in turn is responsible for dramatic differences in the strength and ductility of the two coatings.Coating deposition and additive manufacturing of refractory materials systems (from refractory elements to high entropy alloys) remains extremely challenging due to the extreme melting temperature of refractories and susceptibility to oxidation.Cold spray additive manufacturing offers unique advantages for processing of refractory coatings and components, yet a full mechanistic understanding of the complex processing-microstructureproperties relations in cold spray of refractory alloys is not currently available.This effort improves our understanding of these relations, unveiling the interplay between particle size distribution, processing parameters, splat boundary network formation and strength and ductility of the coating.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 2 .Fig. 3 .
Fig. 2. SEM micrographs of the flat punch and gripper used for the micro-mechanical tests.

M
.Amiri et al.

Fig. 5 .
Fig.6displays microstructural features for the two CS deposits obtained by BSE and HAADF-STEM analysis.An inhomogeneous microstructure and the presence of pores and unbonded intersplat boundaries were observed in both deposits, present to a much greater extent in deposit 1.In both cases, the microstructure was comprised of highly deformed areas featuring elongated and fine recrystallized grains near the intersplat boundaries, along with areas with less deformed coarse grain structures farther from the boundaries in the relatively undeformed interiors of individual impacting particle splats.Energy dispersive spectroscopy (STEM-EDS) revealed measurable oxygen content at certain boundaries within the deposit, as shown in Fig.6(b), demonstrating the presence of residual oxides in those regions.A comparison of the microstructure of the two deposits in Fig.6also reveals that deposit 1 had larger areas with coarse-grained structure of low deformation, as well as larger individual pockets of porosity present at intersplat boundary intersections than deposit 2. Deposit 2 exhibited a higher volumetric density of domains with highly refined grains.The results of the assessments of residual porosity are shown in the box and whisker plots of Fig.7(a), and examples of the binary images produced by filtering and thresholding (albeit at smaller HFW) are illustrated in Fig. 7(b).Each individual imaged area represents one data

Fig. 6 .
Fig. 6.(a) BSE micrographs of cross-sections of the deposits, (b) Microstructural features observed via DF-STEM in the intersplat boundary regions, (c) High magnification DF-STEM images of unbonded areas with corresponding EDS maps showing segregation of oxygen to intersplat boundaries.

Fig. 7 .Fig. 8 .
Fig. 7. (a) Area percentage porosity of deposits, with (b) examples of the probed areas from the deposits and the filtered image with a grey-scale threshold applied to quantify area percent porosity.In (a), the solid red and dashed red horizontal lines inside the boxes show the median and mean values of all data collected from each deposit, respectively.The upper and lower values of the boxes are the medians of the upper and lower halves of the data.Lower and upper whiskers represent the minimum and maximum values that are not outliers, with outliers indicated by the data points beyond the whiskers.(For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

M
.Amiri et al.

Fig. 10 .
Fig. 10.Locations of interest for micromechanical tests of the cold sprayed deposits.Dashed lines represent the intersplat boundaries going across the test specimens.

Fig. 11 .
Fig. 11.Stress-strain curves for; (a) and (b) micro-compression, (c) and (d) micro-tension tests of the two deposits obtained from inside the splat and across the intersplat boundary.SEM micrographs on the right side of each plot show the corresponding specimens at the end of the tests.

Fig. 12 .
Fig. 12. Fracture surfaces of the post-mortem micro-tensile specimens of the deposits; (a) and (b) deposit1-b, (c) and (d) deposit1-i, (e) and (f) deposit2-b, (g) and (h) deposit2-i.Dashed yellow lines represent unbonded boundaries or openings, and arrows point to the features observed on the fracture surface.Note: Side view images were taken at a 45 • tilt angle.(For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 13 .
Fig. 13.Bright-field (BF)-STEM micrographs for micro-tensile specimens; (a) deposit 1 inside splat, (b) deposit 2 with a well-bonded boundary, (c) deposit 2 with a poor-bonded boundary alongside with STEM-EDS maps showing the presence of oxygen in the openings leading to the fracture surface.Insets in (a) represent the CBED patterns, showing the same crystal orientation for the three locations across the fracture surface indicating one large grain was located at the fracture surface and was mainly responsible for the ductile fracture during tension (ZA denotes zone axis).
with T p the particle temperature, T m the melting temperature of the material, v ref cr an experimentally known/measured critical impact velocity for a specific particle size (d ref p ).For the current material, we estimate v ref cr ~460 m/s for a particle with d ref p = 25 μm.C d is the drag coefficient of the particle (considered constant here), and c 2 (0.42 for N 2 and 0.45 for He spray gasses) a fitting parameter.

Fig. 15 .
Fig. 15.SEM images of the coatings' microstructure before and after wedge indentation, revealing unbonded interfaces around larger particles act as the crack initiation sites in deposit 1.

Dividing
Eq. A.1 by particle volume and density of the metal will result in Eq.A.4 for interstitial oxygen concentration in the lattice of metallic particle (