While commercial additive manufacturing processes involving direct metal wire or powder deposition along with powder bed fusion technologies using laser and electron beam melting have proliferated over the past decade, inkjet printing using molten metal droplets for direct, 3D printing has been elusive. In this paper we review the more than three decades of development of metal droplet generation for precision additive manufacturing applications utilizing advanced, high-temperature metals and alloys. Issues concerning process optimization, including product structure and properties affected by oxidation are discussed and some comparisons of related additive manufactured microstructures are presented.
Discussions of new manufacturing paradigms usually invoke comparisons between more traditional subtractive manufacturing and additive manufacturing (AM), having historical roots which can go back nearly 150 years in the context of photo-sculpture, topography, and lithography in various forms . Photolithography and stereolithography (SL) evolved as AM technologies using laser beams to cure (or solidify) photosensitive polymers leading to photolithography central to integrated circuit and multi-layer device fabrication which continues to evolve today. Simultaneously, powder spray and weld-metal overlay technologies evolved as a means to repair worn surfaces and associated surface degradation as well as surface (layer) modification or hardening using electron or laser beam melting of injected metal alloy or hard compound particles or powders [2–5]. A. Ciraud  in a 1972 patent, introduced the concept of metal layer fabrication by selectively melting powders using electron, laser or plasma beams. A decade later, Hikodama  described the first rapid prototyping (RP) system, while Herbert  almost simultaneously described the earliest 3D CAD-driven laser stereolithography system. This was followed by the founding of one of the first commercial AM companies by Charles Hull (ca. 1986) utilizing CAD-driven SL to build layer-by-layer solid structures. Other RP involving solid freeform fabrication (SSF) began to evolve in the late 1980 and early 1990 period, which utilized metal wire feedstock melted by laser or electron beams or similar schemes using powder feed delivery nozzles forming layer-by-layer solid objects as illustrated schematically in Fig. 1(a) and (b)[9,10]. Laser wire feedstock melting evolved as laser cladding-based technologies similar to weld surface cladding or direct metal deposition (DMD), a process referred to as laser engineered net shaping (LENS) of AM metal objects [11,12]. A similar process using electron beam melting of a feed wire in vacuum was also developed as electron beam free-form fabrication (EBF3) . Laser sintering of powder as shown in Fig. 1(b) evolved as direct metal laser sintering (DMLS) or selective laser sintering (SLS), and both wire and powder feed processes have been referred to as direct energy deposition (DED) processes.
Schematic comparisons of metal AM processes and systems. (a) Laser or electron beam cladding using wire feed process. (b) Laser or electron beam sintering based systems. System can incorporate multiple powder feeders. (c) Powder bed fusion processes using electron or laser beam selective melting. Powder is rolled or raked from supply container or cassettes. (d) Binder jet powder process which requires post sintering to permanently bind metal powder and expel binder. Unbound powder is recovered.
Powder-bed fusion technologies also evolved in the 1990s in part as an extension of SLS. Two popular methodologies became commercialized as shown schematically in Fig. 1(c) . In Fig. 1(c), powder from a reservoir is rolled into a layer which is selectively melted using a CAD-driven laser beam, while alternatively in Fig. 1(c), powder is gravity fed from cassettes which is racked into a layer and selectively melted by a CAD-driven electron beam. Fig. 1(c) uses an inert gas (Ar or N) environment for laser melting while electron beam melting is in vacuum. The laser melting process is referred to as selective laser melting (SLM), while the corresponding electron beam melting process is referred to as electron beam melting (EBM). A process devoid of laser or electron beam sintering or melting uses a powder bed, which is selectively spread in a layer from a movable powder nozzle. This is followed by selective dropping of a suitable binder from an ink-jet printer head directed by a CAD program to create a metal/binder product which is sintered at high temperature to remove the binder and sinter (solidify) the metal powder. This process, shown schematically in Fig. 1(d) is variously referred to as binder jetting , powder bed/inkjet printing, drop-on-powder printing, etc. The binder/powder product is extracted from the building process, and after removal of excess or unbound powder, is sintered at high temperature as the binder is vaporized. A variance of this process deposits a metal powder and binder precursor “ink” aggregate on a substrate which is sintered or melted by a laser or electron beam as the binder is expelled.2Examples of powder bed melt 3D (AM) products and microstructures in contrast to wire and powder feed technologies
It is interesting to examine the fundamental schematic views for AM processes shown in Fig. 1 in the context of process variances and parameter variations necessary to achieve optimized product fabrication. For example, in Fig. 1(a), representing LENS or EBF3 processes, the layer (or deposit) thickness will depend upon wire diameter and feed rate relative to the laser or electron beam power. Similarly, in Fig. 1(b), the powder size and size distribution as well as the powder feed rate associated with the SLS or DMLS processes will have a similar effect. In both wire feed and powder feed processes, Fig. 1(a) and (b) respectively, it is possible to add multiple (different) metal or alloy wire or powder feed systems to allow for functional grading of the deposit or multiple metal or alloy component fabrication and integration in producing a product. This can include metal/composite wire or powder injection or the use of separate hard (ceramic) powder feed nozzles. Part resolution and surface finish will depend upon these variables, and part build rates for commercial systems utilizing these concepts have been observed to vary from around 70cm3/h for LENS systems to as high as 700cm3/h for EBF3 systems. Wire feed systems employing a laser can integrate an inert gas shroud to prevent excessive melt zone oxidation, and can effectively repair or resurface large configurations, while commercial LENS systems can build products measuring nearly 1m3; although LENS is generally a low-dimensional accuracy process.
In the direct metal sintering or melting AM processes which can be configured in Fig. 1(a) and (b), relatively complex geometries can be fabricated with no unmelted feed wire or powder to be removed . This is not the case for the powder bed fusion processes depicted in Fig. 1(c) and (d) where unmelted or unsintered powder must be removed. This requires open-cellular or open-porosity geometries to allow the efficient removal of the powder. Systems illustrated schematically in Fig. 1(a)–(c) can also be envisioned as building new melt layers on previously solidified layers, and the solidification of successive melt layers will be variously influenced by the previous crystal structure (and structure orientation) of the (solid) layer; providing varying degrees of epitaxy or epitaxial growth layer-by-layer as recently described by Basak and Das . As a consequence of layer-by-layer epitaxy, many AM products built from either wire or powder feed stock, as well as powder-bed fusion processes (Fig. 1(c)) can produce variations of directional or extended microstructures in the build direction (or Z-axis direction) illustrated in Fig. 1(a)–(c). However, LENS and SLS as well as SLM products usually require post-process HIPing of the AM product similar to contemporary cast products because of process induced strains as well process introduced porosities.
Although somewhat limited in process cost and flexibility in product development, as well as AM product size, powder bed fusion processes involving laser (SLM) and electron beam (EBM) melting of powders have proliferated worldwide as a consequence of commercial system availability. For example, in 2014, 42 Arcam EBM systems were purchased while EOS, SLM solutions, and Concept Laser sold a combined 210 SLM units worldwide; and overall sales increased more than 50% in 2015 . Correspondingly, metal and alloy powder sales have also increased in a similar fashion. But powder costs in contrast to cast or wrought precursor metals and alloys are more than 10 times the unit price. On the other hand, conventional, subtractive manufacturing of products utilizing these precursors often wastes as much as 90% while the required tooling and subsequent processing adds additional product costs.2.1EBM fabrication of Ti-6Al-4V powder products
Fig. 2 illustrates typical precursor, atomized powder for EBM processing along with a more generalized system view of an EBM machine. Ti-6Al-4V represents one of the most important powder materials used in both EBM and SLM processing over the past 15 years as a consequence of its importance in medical device and aerospace component fabrication. An important feature of powder bed fusion technologies, Fig. 1(c), involves the proper selection of powder size and size distribution, since a raked or rolled powder bed is optimized by space filling by smaller particles, and smaller particle sizes require less energy to melt. In addition, unlike the much larger technology of powder metallurgy, so-called green compacts do not require flowable, spherical powders which are essential in rolling or raking powder beds in SLM or EBM processes.
Composite view of Ti-6Al-4V powder (a) having nominal composition shown in energy-dispersive X-ray spectrum (b); typically used in electron beam melting (EBM) system shown schematically in (c): 1 – electron gun; 2 – focus and beam deflection; 3 – powder cassettes; 4 – layer raking system; 5 – specimen/product build and build table which drops down with each layer processing.
The attendant microstructures are also influenced by the beam heating or energy deposition which depends on the beam diameter, raster speed, and current density in the case of EBM. In addition, the necessity to preheat each layer prior to the melt scan, the pre-heat scan rate and beam current, as well as remelt strategies will selectively and often systematically alter the microstructure, and the residual product properties. The product volume (or size) will also influence the thermal issues, primarily cooling or rate of cooling, which can also have a significant effect on the microstructure. Fig. 3 illustrates a few of these features for EBM processed Ti-6Al-4V powder as represented in Fig. 2. The sequence of optical micrographs in Fig. 2(a)–(c) represents 1, 2, or 3 melt passes of the electron beam, respectively. Each melt pass increases the average thickness (width) of alpha-phase lenticular grains from 3μm in Fig. 3(a) to 4.5μm and 6μm in Fig. 3(b) and (c), respectively, causing a corresponding decrease in hardness from Fig. 3(a)–(c). In Fig. 3(d)–(f), the product cooling rate increases correspondingly, which also reduces the lenticular alpha grain width from Fig. 3(d)–(e) because of the heating and the annealing which results in larger-size product fabrication. Fig. 3(f) shows a transformation from alpha to alpha-prime phase plates having a 40% hardness increase from Fig. 3(a) as a consequence of rapid cooling in a foam ligament. Fig. 4 compares the common mechanical properties for as-fabricated EBM and SLM Ti-6Al-4V products with commercial wrought and cast-and-HIPed products.
Optical micrograph sequences showing microstructure variations for different thermal processing conditions for EBM-fabricated Ti-6Al-4V products. (a) to (c) show α-phase acicular grains increasing in thickness for 1 (3μm), 2 (4.5μm) and 3 (6μm) melt passes respectively for cm thick specimens. (d)–(f) show decreasing α-phase thickness for mm thick specimens for (d) (top) and (e) (bottom); and 1 pass in (f) creating <2μm spaced α¿-phase platelets as a consequence of rapid cooling in a foam ligament thin section.
Fig. 5 shows a vertical (plane) section view of the as-fabricated EBM microstructure for a Co-26Cr-6Mo-0.2C superalloy. In contrast to Fig. 3, there is a notable directional, columnar grain (GB) structure which contains columns of Cr23C6 carbide precipitate particles, generally parallel to the build direction (Z-axis) indicated by B. This microstructure has been described in detail by Gaytan et al. . Upon HIP and annealing (at 1200°C for 4h and 1220°C for 4h) the carbide precipitates were observed to dissolve as the structure became equiaxed grains (including annealing twins) containing extensive stacking faults. These features are illustrated in Fig. 6(a) and (b), respectively. This alloy, in the form shown in Fig. 6, is used extensively in cast and HIP-heat-treated commercial (contemporary) implants and other wrought and cast aerospace components. In this regard, Fig. 7 shows a variety of CAD models for open-cellular implant configurations which have been fabricated by EBM [19–21]. Fig. 7 illustrates to some degree the ability of AM processing like EBM or SLM to fabricate rather complex geometries which are impossible to fabricate by any conventional, subtractive or variously integrated manufacturing processes.
Microstructure features for HIPed, EBM-fabricated Co-Cr alloy in Fig. 5. (a) Optical micrograph showing equiaxed grain structure. (b) TEM image showing stacking faults within the fcc grain structure in (a).
Examples of complex, open cellular structure models which have been fabricated in various metals and alloys by SLM and EBM. (a) Porosity variances in cylindrical foam structures. (b) Cross-section view of (a). (c) Top view of (a). (d) Foam structure core surrounded by complex mesh-type cell structure.
As we noted previously, the fabrication of complex or porous geometries as illustrated in Fig. 7 require unmelted powder removal. In addition, conventional powder bed fusion SLM and EBM systems have build volumes between about 0.1 and 0.2 cubic meters. Consequently, there are severe limits on the complexity and size of 3D printed parts, and the requirements for flowable powder feed increases the manufacturing costs in contrast to equivalent weights of more conventional feed stock, including wire forms. In addition, scale-up or enlargement of build volumes is not trivial since there are limitations on the ability to direct the beam to extended product dimensions (in particular an electron beam in EBM processing), and there are problems in maintaining level powder beds because roller or rake flexure can become serious over large areas or powder spans.
In this context, Zenon et al.  have noted that “Digital printing of metals is probably the single most important element missing from functional 3D printing, a technology that today still relies almost entirely on polymer materials”. Indeed, to put metal 3D printing; especially high-temperature materials represented by Figs. 5 and 6, in a truly 3D printing arena where large part manufacture would occur by directed metal droplet deposition from a classical ink-jet printer head will require novel technology design and development.3Metal droplet generator design and development
As Murr  has previously noted, Lord Kelvin developed the first printing device using “ink drops” emitted from an orifice, while Hewlett Packard Corp. (U.S.) introduced the first desktop inkjet printer in 1984. Fig. 8 shows a very general schematic not only representing a so-called drop-on-demand printer head which ejects single ink drops as required, but also conceptual views of ink drop loading and droplet deposition issues on contact with a substrate surface: forming hydrophilic droplet contact (or contact angles ≪90°) and hydrophobic droplet contact (or contact angles >90°). Applied to metal droplet generation and 3D product printing, the initial breaklength to form a droplet (distance from orifice exit surface to droplet formation in Fig. 8) is given by [23,24]:ρ is the drop density, γ is the surface tension, μ is the viscosity, vj is the jet (droplet) velocity, and dj is the jet (or droplet) diameter. c in Eq. (1) is a constant.
Schematic view of drop-on-demand ink jet generator. Droplets can be considered metal or alloy melt while ink “particles” can be hard (ceramic) nanoinclusions, etc. Droplet deposition on a substrate is shown in the context of contact or contact angle.
Generally, droplets or a droplet stream generated as shown in Fig. 8 involves droplet impact with a substrate where it “deforms” (or splats) and solidifies, i.e. the droplet makes contact, spreads, and solidifies. Some drops or fraction can rebound depending upon the size, velocity, and angle the droplet (or droplet stream) makes with the line of emission and formation from the orifice and the substrate plane. Fig. 9 illustrates the splat process or variances of the process, where the splat diameter, Ds, can be generally expressed by :
The final shape of each solidified droplet on the substrate surface is a complex issue affected by heat transfer and fluid dynamics occurring at the droplet collision point with the solid substrate.
Ideally of course the droplet must be in the liquid state (melt), and a metal droplet in a vacuum will cool only by radiation heat loss, which is described by the Stefan–Boltzmann Law :e is the emissivity, σ is the Stefan–Boltzmann constant, A is the droplet area (surface area=4пr2), T is the droplet temperature and Tc is the temperature of the surroundings. Correspondingly, the rate of droplet cooling (or cooling time, tc) will be governed generally byk is the Boltzmann constant, e is the ideal emissivity=1, N is the number of particles; N=mNA/M (where m is the mass of the object, NA=Arogadro's number, M=molten droplet mass). Eq. (6) assumes infinite thermal conductivity so that the temperature of individual metal droplets is equal to the droplet surface temperature. For high-temperature metals such as Ti-6Al-4V or Co-Cr alloys illustrated in Figs. 4 and 5, the surface tension is high, as implicit in Eq. (2), and for small radius (r) droplets, A in Eq. (6) will be small; allowing for slow droplet cooling.
Efforts to develop metal droplet generators began in the context of additive manufacturing or 3D printing of metal products with the work of Orme and Muntz  published 3 decades ago; followed by a patent issued in 1990 . This work, focused on droplet generation and fabrication using aluminum alloys, was followed by extensive research spanning nearly 15 years [28–30]. As illustrated schematically in Fig. 10, this concept focuses on the use of controlled molten (aluminum or aluminum alloy) droplets for AM, and is very similar to the more general droplet generator in Fig. 8; with a notable exception being the droplet charging and deflection system beyond the emitting orifice. (Many ink droplet printers also have scanning electrodes.) In addition, Orme et al.  also noted that “remelting action of the previously deposited and solidified material will insure the removal of individual splat boundaries (Fig. 9) and result in a more homogeneous component”; where remelting thermal requirements were previously studied analytically by Orme and Huang  and indicated a minimum substrate temperature existed from a given droplet impingement temperature that results in remelting. This requirement is similar in effect to the substrate and pre-heat scan temperatures required for optimal melting in powder bed fusion systems (Fig. 1(c)). Furthermore, Orme et al.  observed droplet oxidation in forming grain structure (and at the solidified grain boundaries) for aluminum 2024 alloy AM product formation in an inert atmosphere as illustrated in Fig. 11. In addition, Fig. 11(b) and (c) illustrate that there is grain growth near the top of fabricated products similar to powder bed fusion processes (Fig. 3(d) and (e)) because of the build-up of heat in the AM process. In fabricating products as illustrated in Fig. 11(a), the build table (substrate) was moved in the x–y plane while the droplet print head (Fig. 10) directed the droplet stream at relatively small deflection angles. Droplet diameters were relatively constant at 190μm.
Aluminum 2024 product examples fabricated using a metal droplet generator shown in Fig. 10. (a) Examples of fabricated components. The largest square tube at left is 11cm in length. T and B indicate component top and bottom (or base) respectively in the building process. (b) Optical micrograph showing oxidized grains ∼11cm from a fabricated cylinder base. (c) Grain structure 0.5cm from fabricated cylinder base. Adapted from Orme et al. . (After Murr and Li in Ref. . Note magnification bars in (b) and (c) are 50μm.
Aside from the work of Orme and colleagues for aluminum alloy AM products in Fig. 11, there have been few examples of AM (droplet) product fabrication, while there have been numerous research programs involving other low-temperature metal and alloy droplet generation and deposition such as the work of Tseng et al. , Jiang et al.  and Chao et al.  dealing with Pb-Sn droplets, Cheng et al.  dealing with droplet-on-demand generation of Sn, Pb, and Zn; and related deposition strategies developed by Laso et al. , Chao , Lee et al. , and Bollinger and Abhari  utilizing tin. More recently, Harkness and Goldsmid  described a patent assigned to Boeing (U.S.) where “constituent features of the part are formed by 3D printing… and the part to be manipulated relative to one or more print heads”. This points out that not only is precision controlled molten droplet stream production imperative for advanced AM applications, but the computer control of such printer heads, the clustering of such heads to allow multimaterial deposition, or the enhanced quantity of material deposition is important for large structure and large component fabrication. In addition, the positioning of efficient droplet emitters and their movement and subsequent droplet stream direction will require novel gantry designs and other diverse deposition schemes, including robot arm or other robotic integration into gantry systems to allow large structure and subsystem or component manufacture.3.1Precision metal droplet generator and cluster design
Fig. 12 illustrates a compact, single droplet generator and complimentary droplet generator cluster arrangement recently developed by Johnson et al. . In this design concept (Fig. 12(a)) an inductive-coupled melt generator converts a continuous feed wire supply into a small melt pool in a suitable crucible and orifice arrangement while a traveling wave or other pulsing concept releases droplets forming a Rayleigh jet. The feed wire can be preheated prior to entering the melt regime and droplets can be heated well above the melting temperature. The metal feed wire is connected to a target potential or grounded, and a charging electrode (Fig. 12(a)) charges each droplet to a fraction of the Rayleigh limit as the drop forms on the end of the melt jet exiting the orifice according to :q is the charge, ¿ is the vacuum permittivity, γ is the droplet surface tension (which will be lower at temperatures above the melting point), and r is the droplet radius. Most systems work ∼44% of Rayleigh limit to minimize material and charge emission when the droplet deforms dynamically due to droplet emission dynamics.
Metal droplet generator design using wire feed system. (a) Single droplet jet head. (b) Cluster jet head design. Adapted from provisional patent application number 62308821 by Johnson et al. .
Because, in principal, the deposition scan range of a droplet generator as in Fig. 12(a) will be limited to a narrow angle of deflection using deflection plates below the charging electrodes (in Fig. 12(a)), as in Fig. 10, multiple emitters in clusters as shown in Fig. 12(b) can increase growth or deposition rates or deliver different materials (metals and alloys) or droplet sizes necessary for component or sub-component fabrication in large, integrated AM systems. As noted previously, clustered droplet generator (emitter) heads can be mounted on flexible gantry arrays, robot arms, etc. providing multi-material, multi-axial deposition as shown in the exaggerated cartoon in Fig. 13. It should be emphasized that large-scale AM, 3D printing systems envisioned in Fig. 13 are in high vacuum in order to eliminate or drastically reduce droplet oxidation or other contamination issues. In addition, the nozzle spacing to the printing surface would be considerably reduced from the exaggerated view provided in Fig. 13. Because the deposition occurs in relatively thin layers (assuming droplet sizes ∼20μm), the prospect for creating initial, amorphous structures (since cooling rates will be ≥107°C/s) or nanostructures, as implicit on comparing Fig. 9(a) and (f), is very good. In addition, large, integrated deposition systems implicit in Fig. 13 can also incorporate electron or laser beams to pre-heat or post-heat (and anneal) deposited layers or layer portions to control the residual microstructures and associated properties; or act as sources (S) along with selective detectors (D) for in situ, real-time process observation, analysis and diagnostics. It can be recognized that necessary CAD and related, integrated computer control for emitter head operation and metal droplet stream direction, as well as their orientation for optimal deposition will require a very large and sophisticated computer control platform as a major process component.
Exaggerated cartoon view showing large-scale AM of aircraft structure and components using wire fed, metal droplet generator clusters in vacuum enclosure. The metal or alloy wires are fed from spools (upper left). Modular analytical components which can be strategically placed during the build process are denoted, S, for diagnostic source (electron, X-ray beam, etc.) and, D, detector (secondary electron, energy-dispersive X-ray spectra, etc.). S can also represent electron or laser beam sources for thermal manipulation during layer building. The multiaxial, clustered droplet-emitter heads can be mounted on movable gantry arrays or movable robots or robot arms.
Fig. 13 can be visualized as epitomizing the smart factory where software (CAD) driven integrated advanced manufacturing concepts are combined with various levels of AM to fabricate large, complex structures. This includes 3D metal droplet printing of complex structures, such as those illustrated in models in Fig. 7, as well as closed cell structures, into a variety of structural members (including automotive, aerospace, etc.) to dramatically reduce weight and cost and increase strength and related performance. Using multi-wire metal and alloy cluster 3D droplet printers shown conceptually in Fig. 12(b), product functionality can be addressed in the integrated manufacturing process exploiting high-speed deposition and multi-metal 3D printing. Wire feed systems implicit in Figs. 12 and 13 can operate at reduced material cost, reduced waste, near-net shape fabrication, reduced or eliminated tooling, and product property and performance development through microstructure control as illustrated conceptually in Fig. 3. Our analysis is that parts built by this technique have cost of production significantly lower than obtained by subtractive processes.
It is apparent that in addition to prospects for large-scale AM using wire feed technologies implicit in Figs. 12 and 13, 3D droplet printer design can also be utilized in smaller-scale machines which could allow efficient customized product or component fabrication as well as scale up to larger production arenas. Such smaller-scale application would also benefit from lower precursor material costs and improved net shaping, as well as the elimination of material removal and recovery processes which currently limit powder bed fusion processes. Droplet printing implicit in Fig. 13 can also be combined with or integrated into other modular processes such as conventional EBM or SLM processes to fabricate components which can be integrated into larger modular manufacturing systems, including joining and finishing processes in an automated, CAD-driven manufacturing arena.Conflicts of interest
The authors declare no conflicts of interest.