A gravity-independent powder-based additive manufacturing process tailored for space applications
Olfa D'Angelo, Felix Kuthe, Szu-Jia Liu, Raphael Wiedey, Joe M. Bennett, Martina Meisnar, Andrew Barnes, W. Till Kranz, Thomas Voigtmann, Andreas Meyer
AA gravity-independent powder-based additive manufacturing processtailored for space applications
Olfa D’Angelo a, ∗ , Felix Kuthe a,b , Szu-Jia Liu a,1 , Raphael Wiedey d , Joe M. Bennett e , Martina Meisnar f , AndrewBarnes f , W. Till Kranz g,a , Thomas Voigtmann a,h , Andreas Meyer a a Institut f¨ur Materialphysik im Weltraum, Deutsches Zentrum f¨ur Luft- und Raumfahrt (DLR), Linder H¨ohe, K¨oln, 51170, Germany b Labor f¨ur Regelungstechnik und Mechatronik, Institut f¨ur Produktentwicklung und Konstruktionstechnik, Technische HochschuleK¨oln, Betzdorfer Straße 2, K¨oln, 50679, Germany c Department of Materials Science and Engineering, University of Toronto, Toronto, ON M5S 3E4, Canada d Institute of Pharmaceutics and Biopharmaceutics, Heinrich Heine University, Universit¨atsstraße 1, D¨usseldorf, 40225, Germany e STFC-UKRI, Rutherford Appleton Laboratory, Didcot, OX110QX, United Kingdom f ESA-RAL Advanced Manufacturing Laboratory, European Space Agency, ECSAT, Fermi Avenue, Didcot, OX110FD, United Kingdom g Institut f¨ur Theoretische Physik, Universit¨at zu K¨oln, K¨oln, 50937, Germany h Department of Physics, Heinrich-Heine-Universit¨at D¨usseldorf, Universit¨atsstraße 1, D¨usseldorf, 40225, Germany
Abstract
The future of space exploration missions will rely on technologies increasing their endurance and self-sufficiency, includingfor manufacturing objects on-demand. We propose a process for handling and additively manufacturing powders thatfunctions independently of the gravitational environment and with no restriction on feedstock powder flowability. Basedon a specific sequence of boundary loads applied to the granular packing, powder is transported to the printing zone,homogenized and put under compression to increase the density of the final part. The powder deposition process isvalidated by simulations that show the homogeneity and density of deposition to be insensitive to gravity and cohesionforces within the DEM model. We further provide an experimental proof of concept of the process by successfully 3Dprinting parts on-ground and in weightlessness, on parabolic flight. Powders exhibiting high and low flowability are usedas model feedstock material to demonstrate the versatility of the process, opening the way for additive manufacturingof recycled material.
Keywords:
Additive manufacturing, 3D printing, powder handling, powder-bed fusion, DEM simulation, spacetechnology, weightlessnessAs human reach into space expands, need arises for ma-chines that work under extreme conditions – notably, inabsence of gravity. Space exploration missions are severelyconstrained by payload capacity, and relying upon ground-support would largely increase the risk of failure of suchmission [1]. As long endurance missions must be able tosolve unexpected problems autonomously, a sustainableapproach is the only valid alternative for human space-flight to non-low Earth orbit: missions’ self-reliability willbe a key to their success [2].A vision for space exploration is in-space manufactur-ing (ISM): fabrication, assembly and integration of smallto large structures directly in space [1, 3]. ISM has thepotential to significantly enhance the self-sustainabilityof missions, as it could support space exploration mis-sions by maintenance, repair and production of objectswithout depending on ground-support [4]. Having au-tonomous manufacturing capabilities in space also opensthe possibility to adapt the design of structural systems ∗ Corresponding author
Email address: [email protected] (Olfa D’Angelo) to their final function in zero-gravity environment, insteadof over-engineering them to resist terrestrial gravity andlaunch. Approximately 30% of the structural mass of pay-load shipped to space today could be saved if the launchload constrains could be avoided [5], representing high eco-nomical and ecological gains.Additive manufacturing (AM), also known as three di-mensional (3D) printing, encompasses technologies thathave two essential advantages for space applications: first,compared to subtractive technologies, they reduce thequantity of waste material produced. Second, they openthe possibility to access virtually any geometry, renderingobsolete the geometrical constraints of classical manufac-turing techniques. The possibility to recycle former ob-jects into new feedstock material would optimize payloadall the more by up-cycling waste to minimize the necessaryraw material mass.Strictly speaking of manufacturing, AM already is a per-manent tool in space: extrusion-based 3D printers havebeen on-board the International Space Station (ISS) since2014 [6, 7]. This so-called Additive Manufacturing Facility(AMF) has produced over 200 parts in orbit to this day, in- a r X i v : . [ c ond - m a t . m t r l - s c i ] F e b luding spare parts and tools [7], highlighting AM as an es-sential tool for future space missions. However, extrusion-based technologies suffer inherent limitations. Firstly, theyare restricted to materials showing continuous viscositydecrease with increasing temperature, which makes suchtechnique most adapted to thermoplastics. Moreover, spe-cially manufactured filament feedstock is necessary, whichhas to be carried along at the cost of large storage volume.Besides, filament-based technologies have limited resolu-tion, restrained by the diameter of the deposited filament;parts produced are typically prone to delamination andhighly anisotropic in their mechanical and physical prop-erties [8].As on Earth, different manufacturing technologiesshould be available for space in order to respond to thevariety of needs. Among AM technologies available on-ground, powder bed fusion (PBF) technologies offer thehighest resolution [9] and most versatile techniques [10–12]. The difficulty to handle powders in reduced grav-ity [13–17] has hitherto been an obstacle to further devel-opment of powder-based technologies for ISM. A recentbreakthrough showed the possibility to adapt SelectiveLaser Melting (SLM) of metal powders to weightlessness( µg ). The method proposed by Zocca et al. [18–20] con-sists of stabilizing the powder bed by applying a pressuredifference between the bottom and the top of the powder-bed using a suction pump. Tested in µg between 2017and 2019, it enabled to produce parts from ceramic andstainless steel powders [20] while depositing the powderin weightlessness. Despite the tremendous achievement ofproducing the first parts manufactured from powder de-posited in µg , this method suffers specific drawbacks, de-tailed by Zocca et al. [20]: as large closed surfaces wouldprevent the air flow from going through the parts and ac-cessing the next deposited powder layer, closed horizontalsurfaces cannot be printed. Using open structures con-nected by vertical walls, the thickness of those walls islimited to approx. 2 mm. Moreover, the required pumppower increases with the powder bed height, necessitatinga large quantity of hardware. Finally, powders which in-clude many fines cannot be processed because the fillingof interstitial volume becomes too high and annihilatesthe effect of the air flow. It is also noteworthy that asfor all powder-based AM processes used on-ground, thepowder deposition step is based on the high flowabilityof the powder feedstock [21–23]. This implies strict re-quirements on the manufacturing and storage of the pow-der, difficult to provide in remote, extreme environments.Furthermore, it complicates directly reusing material fromprevious batches and prohibits closed-loop recycling. Suchinherent drawbacks question the superiority of additivetechnologies for ISM, as the impossibility to neither reusenor recycle powder amounts to the production of largequantities of waste material.The contours of a technological gap appear: to be suit-able for space applications, an AM technology would com-bine the assets of PBF with the possibility to use powders regardless of their flow-properties, and be robust againstchanges in g -level. While powder handling remains animportant issue on-ground (1 g ) [24, 25], and in absenceof constitutive equations enabling large-scale predictionsof granular flows in any environmental conditions [26],powder handling technologies for space applications facespecific challenges. Considering the requirement to fulfillgravity-independence, the body force created on each grainby gravity can indeed not be used as transport mechanism.Besides, normal pressure applied on top of the granularpacking can also not be used to induce powder flow, sinceit is reoriented horizontally [27]. Hence, in our process,robustness against gravity-variations is achieved by de-positing powder solely using driving mechanisms shownto induce similar response regardless of the gravitationalenvironment – namely, shear [28] and shaking [29] of thegranular material.Versatility in raw material is required to ease powderstorage and recycling. Besides the higher stress requiredto overcome friction and mechanical locking between par-ticles for low flowability powders, a jammed phase [30, 31]also appears at lower packing density for particles showingangular shape and rough surface state [32]. The appear-ance of a jammed region in a larger packing is a calamity inpowder handling, and can draw complete industrial pro-cesses to a halt. Recent studies show that by changingthe force balance acting on each grain, µg also appearsto decrease the mobility of the grains’ spatial configura-tion [33], ergo facilitating jamming. Actively avoiding theappearance of a jammed phase hence becomes yet anotherrequirement to ensure reliable functioning of powder han-dling and 3D printing for space applications.We propose a method to 3D print powders, regardlessof the rheological properties of the feedstock raw material,and independently of the gravitational environment. Af-ter describing the AM method in Sec. 1, discrete elementmethod (DEM) simulation will be used in Sec. 2 to modelthe powder handling process. An experimental implemen-tation of the AM process will follow in Sec. 3, providing aproof of concept on-ground and in weightlessness throughparabolic flight campaigns (PFCs). Parts manufacturedfrom materials of variable flow-behaviors, under gravityconditions of 1 g and µg , will be analyzed in Sec. 4, en-abling to assess the performances of the AM process.
1. Additive manufacturing method
AM generally amounts to multiple iterations of two mainsteps: material deposition followed by material solidifica-tion. In the case of PBF, powder deposition consists increating a thin and homogeneous layer of granular mate-rial, which will then be selectively solidified. The presentpowder-based AM method aims to conduct this materialdeposition step without relying on the gravitational en-vironment, nor imposing constrains on the raw materialflow-properties.2 igure 1: Principle of the AM method. The feedstock material isconfined in a closed container, inside which the entire process takesplace. The stepwise process consists of: I. incremental platform rise,II. powder deposition and III. selective solidification of the newlydeposited layer. The powder deposition step encompasses the fol-lowing powder movements: 1. vertical downward transport towardsthe bottom of the container, 2. horizontal homogenization to createevenly distributed layers under the platform, and 3. compression ofthe newly deposited layer.
The approach proposed here consists in confining theraw material in a closed container inside which the en-tire process takes place (whereas in conventional PBF,deposition and solidification happen on an open powderbed [9, 20]). The deposition step amounts to controllingthe powder flow inside the closed space to force material tothe desired location. Inside the container is a platform orprinting substrate on which the object will be 3D printedupside down; the desired location for each new layer tobe deposited is the horizontal space under this printingplatform. At the beginning of the manufacturing process,the platform is placed at the bottom of the container. Itthen moves up in discrete steps, each iteration allowingone new layer to be 3D printed underneath the platformand portion of object already printed. Fig. 1 illustratesthis method.Handling the confined raw material is achieved by mov-ing the container itself to force the material to flow to-wards the desired location. The powder displacement canbe divided into two types of movement: 1. the vertical,downward powder transport and 2. the horizontal, planarmovement to create homogeneous layers under the plat-form. Once deposited, the powder can be selectively so-lidified. The solidification also takes place inside the con-tainer: an energy input is provided from outside to thematerial inside the container through the bottom wall ofthe container, transparent to the type of energy used tosolidify the raw material.
Motion is imposed on the powder exclusively by move-ments of the container itself, i.e. only through boundaryforces. The container is cylindrical, axially symmetricalabout the axis along which the platform rises. The pro-cess is schematized in Fig. 2 through the motion of eachpart. Since direct compression of the powder might lead to afully jammed phase, another transport mechanism mustbe sought. Shear stress applied to a granular packingcreates a primary flow independent of the gravitationalfield [28, 34], and it can be applied to powders regardlessof their flow-properties. Therefore, shear will be the pre-ferred mechanism to trigger controlled granular motion.As shear can also lead to shear-jamming [35], a superpo-sition of shear forces in different spatial directions is usedto avoid the creation of stable force chains and thus topreempt jamming.Also in an effort to avoid direct compression, the print-ing platform moving through the container is not a plat-form but a tube; hence, no powder can remain compressedbetween the platform and the upper wall of the containeras the platform moves upward to give space to the printedpart. This rising tube on which the printing substrate isinstalled is labelled inner tube .The rise of the inner tube at each new iteration increasesthe volume available for the powder under the printingplatform, but also in the container in general. To main-tain the total powder volume fraction constant throughoutthe process, the volume gain is compensated by loweringthe part closing the container on its upper section. La-belled closing disc , this part links the outer wall of thecontainer to the inner tube; it descends on the feedstockarea, to push downward the raw material stocked there.Again to avoid normal compression, the closing disc de-scribes oscillatory rotation at a frequency of 1 Hz whiledescending. It is equipped with paddles penetrating thepowder bed. This oscillatory motion forces the material toreorganize regularly, destroying and reforming the “frag-ile skeleton” [31] of force chains supporting the downwardpressure. At each reorganization, the particles are pushedto a position of (temporary) stability lower along the z -axis than their previous one. It is ensured in this mannerthat the material is periodically pushed downwards andenters the next step of the powder handling system: thetransport area.Vertical transport of granular material in a closed con-tainer has been widely studied on-ground, for instance inthe case of silo discharge, showing that normal pressureapplied on top of a granular packing is reoriented hori-zontally [27]. The present application poses a supplemen-tary requirement: the body force created on each grainby gravity can also not be used as transport mechanism,as it would render the powder handling method gravity-reliant. Therefore, a screw conveying system is used totransport the powder vertically: the rotating outer con-tainer (labelled outer tube ) is equipped with helical bladesthat shear the material downward as they rotate. Thismechanism allows to handle a wide range of powders re-gardless of their physical or rheological properties, as willbe demonstrated below.During granular shear, force chains form oblique to thedirection of shear [36]. As force vectors, such fragile net-works of force chains is desired; if they percolate into a3 eedstockTransportscrew-conveyorVibrating platePower sourceVibrating plateOuter tubeClosing discInner tube Figure 2: Schematic representation of the different stages of the AMprocess. The powder container is divided in four parts, each movingindependently: the closing disc descends while describing oscillatoryrotation; the inner tube rises stepwise, rotates and descends to com-press the newly deposited layer; the outer tube rotates to activatethe screw conveying system; the vibrating plate produces horizontalshaking. The apparatus can be divided in four stages (from top tobottom): the powder feedstock, the material transport by screw con-veyor, the homogenization by horizontal vibration and the selectivesolidification. stable configuration, they might transfer the load directlyfrom the screw conveyor to the inner tube, creating ajammed phase that is stable against further motion. To en-sure that force chains forming are intermittently destroyed,a secondary force field is superposed by rotating the innertube simultaneously with the screw conveyor. The mecha-nism used to defuse the force chains is illustrated by Fig. 3;superposition of perturbations in different directions havebeen used previously to tune jamming in dense shear thick-ening suspensions [37]. In the present case, the inner tubeas it rotates imposes a torque on the particles in contactwith it, which “elongates” the chain, thereby destabilizingit by rolling out of the main stress direction [31, 38]. Su-perposing a secondary flow direction forestalls jamming bydefusing the long force chains as they appear, constantlyimposing plastic deformation to the packing. It is note-worthy, that if surface friction increases stability of theforce chains, both mechanisms used in the superpositionof directions of drive are enhanced by such surface frictionincrease, as it also renders contacts between particles andcontainer’s surface more stable.Once the grains have been brought to the bottom of thecontainer, the powder needs to be spread homogeneouslyon the entire bottom surface. To do so, the inner tube risesby more than one layer-height, leaving under the printingplatform an empty volume greater than that of a powderlayer. Then, homogenization is realized by applying hor-izontal shaking to the bottom of the container. Granularhomogenization by planar shaking is a well-know mecha-nism on-ground [39, 40]. In µg , shaking a confined granu-lar sample leads to the formation of a large cluster bounc-ing around the middle position of the axis along which (a) Section view (b) Top viewFigure 3: Illustration of the superposition of force chains (in darkgrey) imposed by the screw conveyor motion (white arrow), and mo-tion created by rotation of the inner tube (black arrow), disruptingthe end-components of those force chains (red particle). shaking is applied. This has been shown through simula-tion [29, 41, 42] and experiments [43–45]. Using alternatedshaking along x - and z -axis, powder is shifted towards themiddle position. Powder coming from the sides is addedas it reaches the lower part of the container. Shaking con-tinues until powder completely fills the bottom layer. Theduration needed to reach this state is monitored in-situ toensure that the entire printing surface is filled with powder(see Sec. 1.3).After having been transported down and homogeneouslydistributed horizontally, the newly deposited powder layerundergoes normal compression by the platform descendingonto it. The powder layer is hence compressed between thebottom wall of the container and the previously solidifiedlayer sitting on top of the platform, increasing the pack-ing density up to close packing. The compression ratio,expressed as a function of the layer height, is a printingparameter.Finally, the newly deposited powder layer can be selec-tively solidified through the bottom wall of the powdercontainer, by the energy source placed outside. Since thiswall must be transparent to the type of energy used tosolidify the material, it is labelled solidification window . Quality and repeatability have been identified as theAchille’s heel of AM [46]. The problem of defects appear-ing in printed parts constitutes a major obstacle for AMin industrial applications, as layer-wise material deposi-tion increases the risk of defects appearance; yet it alsoenables a direct insight into the bulk of the object whileit is manufactured. Using this specificity for in-situ mon-itoring would enable to spot defects and hinder their ap-pearance [47, 48] by continuing the deposition procedure.The present process is designed to allow for closed-loop control by in-situ monitoring. Primarily, the torqueneeded to rotate the inner tube mono-directionally duringmaterial transport is recorded. The inner tube is equippedat its bottom with blades to enhance powder-powder con-tact. The torque developed during oscillatory rotation ofthe closing disc as it descends is also recorded, providing a4econd source of information on the raw powder’s rheolog-ical behavior. The adaptive control loop allows to reactto changes in flowability upon changes in environmentalconditions. It must be noted, that here flowability doesnot refer to an inherent property of the material, but tothe flow exhibited by a powder in the given conditionsand environment in which it is processed, which includes– but is not limited to – the gravitational environment.Hence, the torque developed during powder transport iscompared to a scale established a priori , giving the typi-cal duration needed to deposit a material as a function ofits flow response. This adaptive closed control loop allowsto optimize powder deposition without being limited tosituations formerly encountered.In parallel, a quality assurance system is implementedto monitor the appearance of defects during material ho-mogenization. The solidification window is transparentnot only to the solidification energy but also to visiblelight; hence, live imaging captures the progression of thepowder layer homogenization from below. An image anal-ysis procedure spots heterogeneities in the powder layerby monitoring changes in variability metrics, continuingthe material deposition procedure as long as the chosenmetrics have not dropped under a threshold.Proposing a step toward autonomous manufacturing,concurrent use of those two “probe-and-adapt” systemsdoes not only provide traceability of defects, but their au-tomated correction, ensuring constant and reliable manu-facturing quality. The full printing procedure, including in-situ monitoring mechanisms, is schematized in Fig. 4.
1. Transport
Torquesensing
Assess feedstockrheology
II. DepositionI. IncrementControl time oftransport andhomogenizationIII. Solidification
2. Homogenization3. Compression
Defect spotting(image analysis)
Monitor powderdepositionRawmaterialPrintedobject
Figure 4: Manufacturing procedure, including closed control loopused to optimize the duration of powder deposition and ensuredefect-free powder layers.
This process was developed to be independent of feed-stock flowability and variations in g -level. To observe theeffect of varying those parameters on the powder depo-sition efficiency, for the sake of comparison, the processshould be tested in those different situations with the sameprinting parameters. Printing parameters include powderdeposition time, rotation speed of the different parts, so-lidification time and compression ratio. In an effort to limit the µg -time necessary, the experimental campaign ispreceded by a preliminary simulation study. Besides therotation speeds optimization (not shown here), it aims tominimize empirical parameter screening µg -time, by veri-fying if the same parameters can be used for manufacturinghigh and low flowability materials, in 1 g and µg .
2. Simulation of powder flow
The following simulation study of material deposition isused to validate the printing parameters prior to experi-ment, for both 1 g and µg . Lacking precedent on which torely on for comparison and in an effort to narrow the possi-ble sources of variations, the same printing parameters areused on all the situations presented (therefore, the in-situ probing is not automated in this series of experiments).The simulation study allows to apply this principle whileminimizing the risk of failure, notably for µg experiments. DEM simulation [49] is used to validate in principle thepowder deposition process. It is implemented in the open-source package LIGGGHTS [50] (version 3.8.0), a molecu-lar dynamics (MD) variant suitable for granular materials.The system modelled encompasses N = 76 000polystyrene (PS) deformable 3D particles of diameter d = 2 mm, surrounded by an aluminium (Al) container.Each point particle i is represented by a sphere, and over-laps with a particle j by a distance of δ ij . The Hertz-Mindlin contact model [51–53] is used for the force calcu-lations: each particles pair interacts through a non-linearspring-dashpot viscoelastic mechanical response. Theforce F ij resulting from a collision is expressed as a func-tion of the overlap δ ij and relative velocity through itsnormal v n ij and tangential v t ij components: F ij = (cid:16) k n δ n ij / − γ n v n ij δ n ij / (cid:17) n ij + (cid:16) k t δ t ij δ n ij / − γ t v t ij δ n ij / (cid:17) t ij − κ A ij n ij . (1)The first two terms of Eq. 1 are the normal and tangentialcomponents of the force governed by the stiffness parame-ters k n,t and viscoelastic damping parameters γ n,t . Theyrepresent the mechanical properties of the material consti-tuting the particles. While their numerical value can belinked to true material properties (elasticity modulus andPoisson ratio), they are bounded by numerical constraints.Notably, optimization of simulation time requires to fix asufficiently large time step dt ; but deeming dt too largewould result in overlooking certain collisions, thus inval-idating the simulation. It is customary to consequentlyadapt the value of k n,t to remain at the lower end of thepermissible spectrum, hence reducing computational ef-fort while maintaining expected effects on the large scale.In the light of previous studies [54–56], we estimate that k n,t (cid:38) N m − / is sufficient to obtain stiff particle5 umerical parameters PS-PS PS-Al k n Normal elastic coef. 1 . · . · k t Tangential elastic coef. 2 . · . · γ n Normal viscoelastic damping coef. 4 . · . · γ t Tangential viscoelastic damping coef. 4 . · . · Table 1: Simulation parameters k n,t and γ n,t for the two types ofinteractions present in our model. Elastic coefficients k n, t are inN m − / and viscoelastic damping coefficients γ n, t in kg m − / s − . behavior using a time step dt = 5 · − s. To produce re-alistic effects, during a collision, most of the energy shouldbe dissipated by viscous damping or through friction be-tween the particles, which is obtained in the over-dampedregime, once γ n,t (cid:29) (cid:112) k n,t m , which is by far the casewith γ n,t ∼ kg m − / s − . The exact parameters usedare given in Tab. 1. The tangential term is curbed to re-spect | F t ij | ≤ µ | F n ij | , where µ is the friction coefficient(here µ = 0 . − κ A ij n ij , where A ij is the disk-shapedcontact area between spherical particles i and j , and κ thecohesion energy density, provides an extra cohesive compo-nent to the force calculation, in the form of the SimplifiedJohnson-Kendall-Roberts (SJKR) model [58]. It appendsan additional attractive normal force contribution to theforce calculation at each collision: as two particles enterinto contact, this supplementary force tends to maintainthe contact proportionally to the contact area, calculatedfrom the overlap. The cohesion energy density κ repre-sents all the non-mechanical cohesive forces between theparticles, due to the reduction in surface free energy whenparticles are in contact. It encompasses many possiblemechanisms responsible for cohesion in granular materi-als. Values from κ = 10 − to 10 N m − / are used tostudy the effect of variable cohesion in the granular mate-rial (see Fig. 8).The boundary conditions are embodied by contact sur-faces following the stepwise powder deposition process de-scribed in Sec. 1 (Fig. 2), with the dimensions of the ap-paratus used for experiment (see Fig. 9). The consecutivemotions of each of those parts are listed in Tab. 2. Thenumber of particles is calculated to fill the volume of theexperimental container with a packing fraction ϕ = 0 . Figure 5: Snapshots of the simulation results: on the left, at thebeginning of the process, before the inner tube rise; in the middleat the end of move 5, corresponding to the end of powder transport;on the right at the end of the deposition of one layer. The particles’color indicates their velocity magnitude, made explicit by the scalebar on the right (in m s − ). The critical region regarding powder deposition qual-ity is the centre of the powder layer at the bottom of theprocessing container, which will be selectively solidified.In simulation, this region is a disk of 15 particle’s diame-ters d as radius, and height 5 d after the compression step.Analysis of the particles’ distribution within this region iscarried out by finding the local packing fraction ϕ associ-ated to each particle from a Vorono¨ı tessellation [60]: eachgrain is assigned a unique Vorono¨ı flat-faced polyhedronrepresenting the region of space closer to the centre-pointof each particle than to the centre of any other particle.The ratio of particle volume and polyhedron volume repre-sents the local packing fraction [61]. The mean local pack-ing fraction (cid:104) ϕ (cid:105) can hence be found for any region of space.Vorono¨ı tessellation is performed by the Python packageSciPy [62] and verified with the Voro++ open source soft-ware library [63]; ambiguous Vorono¨ı cells (e.g. at the sys-tem’s boundaries) are discarded. To visualize the simula-tion results, the bottom layer of the powder bed is dividedinto concentric rings, each equally spaced by 2 d . This di-vision of the powder bed bottom layer is represented inthe inset of Fig. 6(b), the printing region represented ingrey. The influence of the gravitational environment is stud-ied by tweaking the gravitational constant g = 9 .
81 m s − ,multiplying it by +1, 0 and − g , 0 g and − g ). Under 0 g , no mass-dependent external force field isapplied: particle flow is induced solely by boundary mo-tion and forces transmitted through surrounding particles.The +1 g environment promotes the fall of particles towardthe bottom of the process container, whereas the − g con-dition tends to pull the particles towards the container’stop, working against the desired flow direction.The evolution of (cid:104) ϕ (cid:105) at the end of each step of the pow-der deposition process, averaged over the printing regionin Fig. 6(a), shows that the gravitational vector strongly6 able 2: Description of the consecutive motion of each element of the powder container modeled in the DEM simulation, corresponding tothe 3D printing procedure described. Move 2.a and 2.b occur simultaneously. d is one particle diameter. Time steps are given in simulationunits, dt = 5 · − s. Time step(end of move) Move Part Movement Velocity1 . · Move 0 Closing disk Linear movement in − z -direction: close printing bed Speed0 .
23 m s − . · Particle settling (realized under +1 g )Adapt g -level1 . · Move 1 Inner tube Linear movement in + z -direction: incremental rise by 10 d Speed0 .
20 m s − . · Move 2.a Inner tube Clockwise rotation about z -axis Period 0 .
025 sMove 2.b Closing disk Linear movement in − z -direction Speed0 .
05 m s − . · Move 3 Closing disk Counterclockwise rotation about z -axis Period 0 .
05 s2 . · Move 4 Closing disk Clockwise rotation about z -axis Period 0 .
05 s2 . · Move 5 Outer tube Clockwise rotation about z -axis Period 0 .
025 s2 . · Move 6 Vibrating disk Shaking along x -direction Period 0 .
05 s3 . · Move 7 Vibrating disk Shaking along y -direction Period 0 .
05 s3 . · Move 8 Inner tube Linear movement in − z -direction: compression by 5 d Speed 0 . − . . . . . . . . . · . . . . κ = 1 · J m − (a) Time /time step M e a np a c k i n g f r a c t i o n 〈 ϕ 〉 − g g +1 g − − − Move number | 〈 ϕ g 〉 − 〈 ϕ g 〉 | 〈 ϕ g 〉 . . (b) Distance l from the centre /particle diameter d M e a np a c k i n g f r a c t i o n 〈 ϕ 〉 − g g +1 g Figure 6: (a) Mean packing fraction (cid:104) ϕ (cid:105) on the printing region aftereach move of the powder deposition process (moves 0 to 8), for grav-itational acceleration +1 g , 0 g and − g ( g = 9 .
81 m s − ). The localpacking fraction ϕ is obtained for each grain by Vorono¨ı tessellation,error bars represent the standard deviation of the ϕ distribution overthe entire printing region. The inset shows the normalized differencebetween +1 g and 0 g at the end of each move. (b) (cid:104) ϕ (cid:105) per ring ofthe bottom layer at the end of move 8 (end of layer deposition), for+1 g , 0 g and − g . The division of the powder bed bottom layer inconcentric rings, equally spaced by two particle diameters d , is rep-resented as an inset. The printing region is only the middle cylinder,represented in grey (and marked by a vertical dashed line). modifies the granular density after the rise of the innertube (move 1), before the container’s motion begins to shiftmaterial downwards: under +1 g , the particles’ weight cre-ates a collective motion towards the bottom, and then hor-izontally redistributes the particles as they slide on eachother, piling until reaching a slope corresponding to theangle of repose. The empty space under the printing plat-form is filled at (cid:104) ϕ (cid:105) = 0 .
08 solely under the effect of gravityat the boundary of the printing region, which can be ob-served in Fig. 7(a): at t = 1 . · the outer part of thebottom layer, at a distance l > d , is already filled at (cid:104) ϕ (cid:105) = 0 .
41, and particles have reached the outer diameterof the printing region, with (cid:104) ϕ (cid:105) (10 d ≤ l ≤ d ) = 0 . g – see Fig. 7(b). In contrast, under 0 g and − g , theparticles are not pushed towards this empty space, respec-tively due to a lack of mass-dependent force or a forcetowards the top of the container. With move 2 begins thepowder transport phase, which triggers collective granu-lar downward motion, regardless of the gravity-level. (cid:104) ϕ (cid:105) undergoes a steep increase, but the normalized differencein packing fractions between +1 g and 0 g – represented inthe inset of Fig. 6(b) – remains at 40%. Throughout thistransport phase (moves 2 to 5), material pushed down-wards slowly invades the printing region, already creatinga relatively homogeneous layer under +1 g , and remain-ing at the outskirts in 0 g , see Figs 7(a) and 7(b). InFig. 6(a), the large standard deviation in ϕ results fromthe variability throughout the printing region: as mate-rial is slowly pushed downwards, the centre remains atlow or null packing fraction – see Fig. 7(b). The normal-ized difference between +1 g and 0 g has dropped to 30%by move 4. During move 5, the screw conveyor trans-ports a large quantity of material downwards, which isforced towards the container’s bottom: some is pushedtowards the centre, reducing the difference in (cid:104) ϕ (cid:105) be-tween +1 g and 0 g to approx. 5%. The homogeniza-tion phase follows (moves 6 and 7): the granular density7 . (a) +1 g κ = 10 J m − (b) g κ = 10 J m − g κ = 10 − J m − (c) Distance l from the centre /particle diameter d M e a np a c k i n g f r a c t i o n 〈 ϕ 〉 . · – . · – . · – . · – . · – Figure 7: Mean packing fraction (cid:104) ϕ (cid:105) per concentric ring (equally spaced by 2 d ) as a function of the distance from the bottom layer centre (atabscissa 0) to its outer bound (at 20), expressed in particle diameter d . The color code corresponds to the time t given in simulation timesteps. becomes homogeneous throughout the printing substrate( l ≤ d in Figs. 7(a) and 7(b)), regardless of the gravita-tional environment (Fig. 6), culminating for all g -levels at (cid:104) ϕ (cid:105) ≈ .
55. Finally, compression of the newly depositedlayer (move 8) compacts the powder and erases the remain-ing difference between g -levels, with |(cid:104) ϕ g (cid:105)−(cid:104) ϕ g (cid:105)|(cid:104) ϕ g (cid:105) ≈ (cid:104) ϕ (cid:105) after move 8 is shown inFig. 6(b) throughout the bottom layer. The packing frac-tion within the printing region (for l ≤ d ) is very high,with (cid:104) ϕ (cid:105) ≈ .
87. The average packing fraction achievedis significantly higher than the expected close packing ofmonodispersed spheres, indicating that particles overlapdue to compression. It is noteworthy that the most cen-tral region, being composed of the smallest volume, is moreprone to statistical variability, which explains its strongervariation in (cid:104) ϕ (cid:105) as a function of the g -level. Besides, thecloseness between the 0 g and − g results is explained bythe fact that we look solely at the bottom layer centre,enforcing the importance of the horizontal motion (layerhomogenization).To summarize, the quality of the powder depositionshows no dependence on the gravitational environment,although the system simulated shows variations depend-ing on the g -level, suggesting robustness of the processagainst changes in gravitational environment. The other relevant parameter for the AM process is de-pendence on the flow-properties of the powder feedstock.The corresponding simulation variable is the cohesion en-ergy κ , which adds to the contact force an attractive termalong the axis defined by the aligned particles centers, pro-portional to κ times the area of contact A ij (see Eq. 1).Numerical values of κ are varied from 10 − J m − (fora very low cohesion, hence highly flowable powder) to10 J m − (for a highly cohesive powder). (cid:104) ϕ (cid:105) through-out the powder deposition is presented in Fig. 8.For all κ ≤ J m − , our model shows very similar re-sults: such high flowability powder flows under the print-ing platform quickly, as it is already put into motion by . . . . . . . . . · . . . .
81 [ κ ] = J m − g = 0 m s − Time /time step M e a np a c k i n g f r a c t i o n 〈 ϕ 〉 κ = 1 · − κ = 1 · κ = 5 · κ = 7 . · κ = 1 · Figure 8: Mean packing fraction (cid:104) ϕ (cid:105) on the printing region after eachmove of the powder deposition process (moves 0 to 8), for valuesof the cohesion parameter κ ranging from 10 − to 10 J m − , theextreme cases available in the present simulation. the concurrent inner tube rotation and closing disk de-scend (move 2). Non-cohesive powders are very sensitiveto collisions and tend to move freely within the container.The remaining steps of the process have little effect on thepowder repartition, with the exception of move 5, whichslightly increases powder density in the printing region bybringing some more material downward, and the compres-sion step (move 8) which significantly increases (cid:104) ϕ (cid:105) from0.52 to 0.85. Localized (cid:104) ϕ (cid:105) shown in Fig. 7(c) reveal thesame trend, as the transition between the initial state ofabsence of powder at l ≤ d and the post-transport stagewhere (cid:104) ϕ (cid:105) ≈ . l ≤ d happens in less than 0 . · time steps. The highest interparticular cohesion availablein our model is κ = 10 J m − . Such highly cohesive par-ticles tend to remain together, as outward forces resultingfrom collisions are minimized. This is particularly visibleon moves 2 to 5, where the difference between highest andlowest κ reaches its maximum ( ≈ (cid:104) ϕ (cid:105) ≈ . (cid:104) ϕ (cid:105) ≈ .
56, slightly higher than for high flowabil-ity powders. Again, the final layer compression completesthe deposition process by fixing the particles in a state ofhigh density, independently of their cohesive interactions:it erases all differences and brings the final (cid:104) ϕ (cid:105) to ≈ . (cid:104) ϕ (cid:105) ≈ .
87, with a stan-dard deviation among all experiments of 0 .
01 only, erasingboth internal and external variability factors. The sameparameters can hence be used in a µg experimental cam-paign, enabling acute assessment of the effect of feedstockflowability decrease and g -level change.
3. Experimental proof of concept
Having confirmed by simulation the working principleof the proposed AM process, it is implemented in two3D printers and tested on-ground and in weightlessness.Parabolic flight campaigns (PFCs) are used to conduct µg experiments: the 34 th DLR PFC in September 2019, usedfor testing the hardware in µg , and the 72 nd ESA PFCin November 2019 (in the context of the ESA EducationFlyYourThesis! GrainPower project), during which fivesamples were successfully manufactured fully in weight-lessness, providing a proof of concept for the AM process.PFCs allow, by flying an airplane describing parabolic tra-jectories, a period of free fall of about 22 s, which enablesto conduct experiments in weightlessness. This µg periodis surrounded by hypergravity phases as the plane risesand swoops. This maneuver is typically repeated thirty-one times per flight day, a campaign consisting of three tofour flight days. The final experimental rack used to pro-duce the weightlessness samples contains two 3D printers,used to each produce one sample per flight-day. A 3D printer built to implement the AM process de-scribed (and its digital counterpart) are shown in Fig. 9.The powder container is composed of two coaxial cylin-ders: (A) the inner tube and (C) the outer tube, respec-tively of diameter 65 mm and 120 mm. From above, it isenclosed by (B) the closing disc, filling the space betweenthe two cylinders and moving down to control the powderbed volume. On the bottom, it is closed by (D) the vibrat-ing disc, which contains the solidification window, through (A) Inner tube(C) Outer tube(D) Vibrating disc(E) Infrared lamp(B) Closing disc
Figure 9: (Left) photography and (right) Computer Assisted Design(CAD) of the 3D printer. The main components of the printer arelabeled. On the CAD, the fixed structure is represented in grey whilethe moving parts are colored, each color representing a movementblock.Table 3: 3D printing parameters, l being the layer height. Part of printer Motion Operating parameters (A) Inner tube Translation z steps: +1000 µ m, − µ m( l = 500 µ m, compressionratio / )Rotation Rotation speed 0 .
102 m s − (B) Closing disc Translation z step: − µ mRotation Rotation speed 0 .
126 m s − (C) Outer tube Rotation Rotation speed 0 .
226 m s − (D) Vibrating disk Shaking Amplitude 2 mmFrequency 5 Hz(E) Infrared lamp Lamp power 500 WSintering time 20 s which the raw material can be solidified by (E) the energysource, placed underneath the powder container. A thor-ough technical description is available elsewhere [64].The type of energy source and the solidification win-dow’s material determine the maximum temperature al-lowed, hence the adequate raw materials. In the demon-stration experiment presented here, polymer powders areused; an infrared (IR) lamp serves as energy source (Quat-tro IR emitter from Heraeus, Germany), and the solidifi-cation window is a doubled 5 mm thick fused silica plate(proQuarz, Germany). The general structure is formedby 30 ×
30 mm aluminum profiles on which the individ-ual parts are mounted. The printing volume available is acylinder of diameter 65 mm and height 50 mm. Motion ofeach part exactly follows the description given in Fig. 2 –specific motion parameters are given in Tab. 3.The 3D printing procedure is as follows: first, the inner9ube translates vertically to make space for the new layer.At the end of the powder deposition phase, it moves ver-tically downward to compress the powder underneath. Tomaintain a constant volume inside the powder contain-ment, the closing disc translates down to counteract themotion of the inner tube; meanwhile, it rotates in alter-nating direction to probe the packing and keep it fromjamming. Powder downward transport is carried out bythe screw conveyor placed inside the outer tube, rotatingto push the powder towards the bottom of the container.The shell of the inner tube also rotates (independentlyof its translation motion) to probe the powder flow fromwithin the container, providing a rheological characteriza-tion of the feedstock powder.As mentioned previously, in this first study the in-situ probing control loop is not automatized to allow betteroverview of the effects of our variable parameters (gravityand powder flowability) on the powder deposition. To demonstrate feasibility of AM from raw materials ofhigh to mediocre flowability, two exemplary demonstra-tor powders are used, which share all physical character-istics but surface roughness. This enables us to test solelythe effect of a decreased flowability on the powder deposi-tion and solidification process. The model substances arecrafted as follows. A monodisperse spherical polystyrene(PS) powder of main diameter 80 µ m is used. Produced bythe company Microbeads under the name Dynoseeds, thepowder as-received from the manufacturer is dry-coatedwith a sub-micron angular PS dust, as shown in Figs. 10aand 10b. This powder is labelled Rough Surface ( RS )in the subsequent text. This coating is removed by wet-sieving the powder batch and subjecting it to ultrasound ata frequency of 20 kHz for a duration of 8 hours per batch.The resulting particle surface state is shown in Figs. 10cand 10d: the asperities have been removed, leaving ex-posed the smooth surface of the spherical particles. Thispowder is named Smooth Surface ( SS ) in the subsequenttext.The resistance to wear of the rough coating is testedto ensure its persistence throughout the experiment. It isessential to validate the principle of this study that themodel powders used retain their flow-properties through-out the entire experiment, independently of the load towhich they will be submitted during powder deposition.Particles are subjected to shear in a Couette-Taylor shearcell, continuously for 20 hours at increasing shear rate˙ γ = 10 − to 10 s − ; SEM microscopies taken beforeand after are shown in Fig. 11. The surface is visibly un-changed by the long duration shear test: the rough coatingis still distributed on the entire particle surface, showingthat it remains despite frictional contacts.How to best characterize the rheology of powders forAM is an open question [21]. The term powder flowabil-ity is widely used and intuitively understood; however, itlacks a clear definition, as it does not rely on a normalized (a) RS grain (b) RS grain closeup(c) SS grain (d) SS grain closeupFigure 10: Scanning Electron Microscopy (SEM) images of thepolystyrene (PS) powder used as raw material (the microscopies aredone for particles of diameter 250 µ m), (a) for powder with roughsurface state ( RS ), at the scale of a particle and (b) at the scale ofits surface; (c) for powder with smooth surface state ( SS ), at thescale of a particle and (d) at the scale if its surface. The increasedsurface roughness in the sub-micrometer range is clearly visible in (a)and (b), while (c) and (d) exhibit a much smoother surface. SEMimaging is done at 1 keV. measurement method, nor on international system units.A powder is defined as flowable if it tends to plasticallydeform (i.e. flow akin to a liquid) under a certain stimu-lus – which may simply be its own weight. In contrast, a non-flowable powder resists flowing and tends to maintainits shape akin to a solid. If it is forced into flowing byan external load, it will do so in large chunks of materialthemselves preserving their shape, in an erratic mannerand showing higher tendency to block the flow by formingstable aggregates that can withstand a finite amount ofstress before yielding (jammed regions). In other words,contacts between grains tend to be more enduring [65].The terms high and low flowability are used throughout (a) Before testing (b) After testingFigure 11: SEM images of the rough surface polystyrene powder usedas raw material (a) before and (b) after a twenty hours long sheartest, with two minutes of air-fluidization before and after at flow rate5 L min − . The sub-micrometer surface roughness remains presentin the same quantity and homogeneity on both images. Imaging isdone at 1 keV.
50 10046810 (a) RS , d = 40 µ m RS , d = 80 µ m RS , d = 250 µ m 0 50 100 (b) d = 80 µ m RS SS (c) d = 10 µ m to 40 µ mTi-64 Penetration rotational speed v /mm s − Sp ec i fi c fl o w e n e r g y E * Figure 12: Specific flow energy E * as a function of the helix speed v for samples (a) of Rough Surface ( RS ) polystyrene powder of diame-ter 40 µ m, 80 µ m and 250 µ m; (b) of 80 µ m-diameter Smooth Surface ( SS ) and RS powders (model materials used for additive manufac-turing experiment); (c) of polydisperse Ti-64 powder with size in therange 10 µ m to 40 µ m, a typical material used in AM [70]. the present work following this phenomenological defini-tion.In absence of a universal definition, the procedure usedto characterize flowability of the SS and RS powders isthe so-called flow energy measurement available on theFreeman Technology 4 Powder Rheometer (FT4) [66, 67].It consists of extracting from a powder bed of height h an helix of angle α and radius r , at different speeds v ,recording the torque M and normal force F to calculatethe flow energy E . Following Wenguang et al. [68, 69], adimensionless flow energy E * is introduced by normalizingthe flow energy by the potential energy of the sample: E * = Em s h g = 1 m s h g (cid:90) h (cid:48) (cid:18) M ( h (cid:48) ) r tan α + F ( h (cid:48) ) (cid:19) dh (cid:48) , (2)where m s is the total mass of the sample and g the grav-itational acceleration. Qualitatively, an increase in flowenergy E ∗ corresponds to a decrease in powder flowabil-ity . Fig. 12 shows the specific flow energy E *, measuredat penetration rotation speeds between 10 and 100 mm s − for powders at high packing fraction ϕ = 0 .
6. In an effortto contextualize those demonstrator powders, the readeris provided with material for comparison: E * is measuredfor RS powders of much smaller and much larger particlediameter (respectively d = 40 µ m and d = 250 µ m), andfor a Ti-64 metal alloy powder, typically used in AM.It has been shown that grains of smaller diameter tendto exhibit higher cohesion, due to the predominance ofvan der Waals interactions [65, 71–73], hence undergoinga flowability decrease. This decrease is indeed captured bythe specific flow energy test shown in Fig. 12(a): as thediameter d doubles (40 to 80 µ m), the flow energy neededto make the powder flow decreases (by 5.7% on average),showing a better flowability of the powder with largergrains. The grain diameter is then increased to 250 µ m,inducing a further decrease of E * (by 9.6% in average),again showing that larger particles amount to a powder of higher flowability. The specific flow energy increases withdecreasing particle size; this trend is preserved over all the rotation speeds measured.In Fig. 12(b), E * is presented for the SS and RS µ mdiameter powders used in our experiment. In the densepacking of rough particles, surface friction is activated asasperities on the grain’s surfaces interlock: the flow en-ergy is 15% higher for the RS powder. The effect of sur-face roughness is clearly visible: increased friction begetshigher stress necessary for particles to slide along eachother, thus higher stress to trigger flow; the RS powderexhibits lower flowability than the SS powder. This effectis at least comparable to the one induced by a change ofthe particles’ diameter by more than a factor 6.Finally, a comparison is provided in Fig. 12(c) to a com-mercial 3D printing metal powder: the polydisperse Ti-64 alloy powder [70], with particle diameter in the range10 µ m to 40 µ m. Its flowability is slightly higher than the RS powder but much lower than the SS powder, placingour two demonstrator powders as boundaries framing thetypical materials used in AM. For sake of comparison, all samplesare obtained using the same printing parameters; in par-ticular, the same deposition and solidification time of 20 senable the full manufacturing procedure to be carried outfully in weightlessness for the µg samples. The layer heightis 500 µ m (corresponding to ∼ d ) and compression rate50% – i.e. the printing platform rises by 1000 µ m beforeeach deposition step, then descends by 500 µ m to realizecompression. To provide as much information as possibleon the powder deposition, no compression is applied dur-ing solidification, and sintering is preferred over melting,as it maintains possible heterogeneities of the depositedpowder layer. Samples preparation.
Samples are prepared to faithfullyreflect the powder deposition at its most challenging po-sition: in the centre of the printing volume. The energysource for solidification (IR-lamp) provides a homogeneousheat-surface, enabling to solidify the entire sample-sectionwithin less than 20 s. However, the IR-lamp providesslightly stronger heating on two regions of the printingbed of approx. 10 mm by 30 mm, represented in blue inFig. 13(a)-(b). The lamp is placed accordingly to ensurethat the centre of the printing bed be under one of the ar-eas of preferred heating, as shown in Fig. 13(a). Each 3Dprinted sample is hence cut to extract a square sample ofapprox. 10 by 10 mm, cut out of the centre of the printingbed – see Fig. 13(c).
Reference “known good” and “known bad” samples.
Asample is sintered on-ground under the best possible con-ditions to serve as a reference “known good” sample forcomparison with 3D printed ones. It is sintered from thehighly flowable SS powder in an oven at 200 ° C for onehour, under a weight of 3 kg to ensure continuous pressure during the solidification process. This aims to increase11 igh heatareas Selected sampleCentre ofdepositionzoneFullprinting area (a) Sample preparation rationale(b) IR-lamp high heat areas (c) Selected sampleFigure 13: Sample preparation rationale. (a) Superposition of highheat zones with the centre of the printing area, including delimita-tion of the selected sample. The sample represented here is manufac-tured on-ground from rough surface powder. (b) IR-lamp includingschematic representation of high heat zones (in blue on the picture),corresponding to the zones marked in blue on the full sample. (c) Re-sulting sample after cutting (the sample represented here is sample L
3, 3D printed in weightlessness from rough surface powder). material density, as the continuous pressure enhances de-gassing and porosity size reduction. (This “known good”exemplary sample is shown in Fig. 16(a).)Another sample, 3D printed in µg under wrongly tai-lored solidification parameters, serves as a “known bad”sample (see Fig. 16(d)). Instead of being sintered intohomogeneous powder layers, overheating resulted in par-tial melting. As the empty space between grains be-comes trapped into molten material, the volume loss isnot counteracted by reduction of the printing bed volume.Therefore, the supplementary volume transforms into largeporosities scattered along the sample. This sample is usedhere to validate the characterization procedure by showingthe results obtained for a “worst case scenario”. XCT specifications & data analysis.
In-bulk characteriza-tion of the samples is done by X-ray computed tomogra-phy (XCT). The machine and scanning parameters usedare presented in Tab. 4.Homogeneity of porosities distribution is found by twoautomatized image analysis procedures implemented inthe Python PoreSpy library [74]. First, images aremade binary by the automated procedure available in Im-ageJ [75] ( intermodes method, automated thresholding).Average density is calculated per “slice” of depth 8 µ malong the y -axis. The pore size distribution is found bydetermining for each pore the maximal radius of a spherethat fits inside. This method is adapted for samples show-ing relatively spherical porosities, homogeneous in shape,which is the case for most of our 3D printed samples. To capture the length of pores with “un-spherical” shapesand identify a possible anisotropy in pore size, the “chordlength” method is employed. It consists of drawing chordsthat span across each pore in a given direction; the appear-ance frequency of each chord length is extracted along x -and z -directions and compared, to detect large defects, inthe deposition direction ( xy -plane) or regarding interlayeradhesion ( z -direction).
4. Results & discussion
The primary result to report is the successful manufac-turing of samples 3D printed from the SS and RS pow-ders, under 1 g and µg (see Fig. 14). The resulting partsmaintain their shape, show homogeneous external appear-ance and smooth surface, without obvious defects, holes,nor heterogeneous powder repartition. Samples were putthrough further analysis to verify if this macroscopic as-sessment could be extended to the microscopic scale. (a) RS µg sample (b) SS µg sampleFigure 14: Samples 3D printed in µg , from (a) RS and (b) SS powders, as extracted from the printing bed. The average powder density of each sample is comparedfor all samples in Fig. 15, hence verifying the quantity ofmaterial effectively deposited. Typical XCT slices used forin-bulk characterization are shown in Fig. 16. Bare-eye ob-servation of those images show that all porosities display arelatively spherical shape, and homogeneous distributionthrough the sample, with neither a preferred direction norobvious signs of delamination between layers, for all sam-ples but the “known bad” one. Notably, the 500 µ m-highlayers cannot be distinguished with bare eye, although thesamples each comprise multiple layers.As the samples are sintered and not melted, the averagedensity of a good quality 3D printed part is expected tobe slightly higher than random close packing (rcp) for amonodispersed spheres packing, i.e. 64% [59, 76]. The av-erage densities of 3D printed samples are higher than rcp,showing that the powder is effectively deposited in the cen-tre of the printing bed. The “known good” sample sinteredunder compression and the SS samples 3D printed under1 g and µg reach ∼
70% density, with standard deviationsless than 1%. The RS samples show a density ∼
5% lower,12 able 4: Scanning parameters for X-ray computed tomography of sintered PS.
Computedtomography system Sourcevoltage Outputcurrent Projectionper scan Measurementsper projection Exposure time Voxel size
CT-ALPHA(ProCon X-ray, Germany) 80 kV 70 µ A 1600 10 1000 ms 8 µ m RS µgSS µg “Known bad” RS gSS g “Known good” 68 . . . . . . Figure 15: Average density of the samples. Error bars represent thestandard deviation on the per- xz -slice densities for all slices, in eachsample. regardless of g -level and of the fact that the same printingparameters are deployed for both types of feedstock ma-terial. Although higher porosity is reflected in the loweraverage density of the RS samples, very low standard devi-ations of 0.7% and 0.9% (respectively for 1 g and µg ) showthat mediocre flowability powder can be used as AM base-material without triggering major defects in the printedparts: porosities are distributed homogeneously through-out the samples. Comparatively, the “known bad” samplesshows surprisingly high average density of ∼ g and µg , but the RS powder is moredifficult to deposit. Simulation predicted that the homog-enization step would erase material-dependence, but ex-periments show that RS powder remains at lower den-sity after homogenization and compression. This discrep-ancy might be due to simulated particles being smooth inessence, leading to lower reliability of our model for grainswith roughened surface. Also, the simulation uses signifi-cantly larger particles, which increases powder flowability,as discussed in connection with Fig. 12. The pore size distribution is shown in Fig. 17. The XCTimaging resolution sets the minimum pore size detectableat 15 µ m. Comparison between SS samples in Fig. 17(a)shows that all have most pores in the range of 20 µ m to25 µ m. The sample sintered under weight has the largestamount of small pores (diameter 20 µ m), which shows, asexpected, the best powder repartition and compaction.The 3D printed samples also show a peak at small pores;however for both material-qualities, a second peak appearsat 30 µ m, increasing the mean pore size. Most importantly,large pores of 35 µ m to 60 µ m have a very low probability inthe “known good” sample, and low for the SS
3D printedsamples. The SS g sample has a high standard deviationfor 35 µ m pores, which shows that some of such pores, un-evenly distributed in the sample, slightly lower its globaldensity. Despite those differences, the three samples havemostly similar pore size distribution, and repeatedly showno large pore ≥ µ m.The RS samples in Fig. 17(b) show an obvious differ-ence in pores size for the “known bad” sample: its porediameter distribution has a much longer tail than all othersamples, with a low density probability of small pores un-der 50 µ m and some very large pores up to 700 µ m. Apartfrom this outlier experiment, RS experiments confirm thata clear difference between base materials can be made:even if the distribution is relatively close, with two peaksat pore diameters 15 µ m and 30 µ m, for the smooth-surfacesamples the smaller pore size is almost twice more likely toappear than the smaller one, while for the RS , the two porediameters are almost equally likely to appear. This clearlyshows that the lower average sample density encounteredon the RS samples is due to a higher amount of largerporosities of average diameter 30 µ m, albeit those porosi-ties being distributed evenly throughout the samples. Be-sides, it confirms that the samples printed on-ground andin µg from their respective base-materials show very sim-ilar pore size distribution: within experimental error, thegravitational environment in which the samples have beenmanufactured does not play a role in the quality of 3Dprinted samples.The RS sample has larger pores than the SS sample,showing that the lower average density unveiled in Fig. 16is the result of generally larger pores, similarly well scat-tered throughout the samples. Such larger porosities arelinked to the decreased packing efficiency undergone bymaterials showing increasing surface roughness: as thestress needed for grains to slide on each other is increasedby surface roughness, the powder’s ability to reorganize13 a) “Known good”, SS powder in 1 g (b) SS powder in 1 g (c) RS powder in 1 g (d) “Known bad”, RS powder in µg (e) SS powder in µg (f) RS powder in µg Figure 16: Extracts from XCT slices of samples 3D printed in all situations studied. xz -plane is shown, x to the right, z to the top, with z being the height (direction orthogonal to layer). · − (a) Porosity diameter / µ m P r o b a b ili t y d e n s i t y f un c t i o n / µ m − “Known good” µg g · − (b) Porosity diameter / µ m“Known bad” µg g . . µ m P d f · / µ m − Figure 17: Probability density function (pdf) per circle diameterfitted in each pore for (a) the “known good” sample, and the SS samples 3D printed under µg and 1 g ; (b) the “known bad” sample,and RS samples 3D printed under µg and 1 g . Points represent theaverage over each full sample, and error bars the standard deviation.The inset in (b) shows the “known bad” sample on appropriate scale. into a denser packing is decreased at constant compres-sion stress. The maximum packing density for each specifictype of powder is reached, in absence of other means im-plemented to increase material density (e.g. compressionduring solidification). Chord length analysis is used to capture anisotropy inthe pores’ shape and test whether the manufacturing pro-cess begets a preferred direction; it is presented in Fig. 18for the x - and z -directions. To allow precise compari-son of the different samples, each data set is fitted witha log-normal distribution. This distribution is plausiblebecause the chord lengths are not independent: each porethat contributes to chord length L , also contributes to allsmaller chord lengths, so that the addition of a long chordrescales the entire distribution at smaller chord lengths.Assuming that pore sizes are independent from each otherand randomly distributed, the central limit theorem hencesuggests a log-normal distribution. Previous work involv-ing the distribution of geometrical shapes enclosed withinrandomly distributed voids also found a log-normal distri-bution [77]. The fitting parameters, µ x, z and σ x, z , cor-responding to the mean and standard deviation of thelogarithm of the data, are presented for each sample in x - and z -directions in Tab. 5. The mean chord length (cid:104) l x, z (cid:105) = exp ( µ x, z + σ x, z / ) is also given for each fit toease interpretation.For the “known good” sample in Fig. 18(a), withoutsurprise the mean chord lengths are the lowest of all ex-14 − − − (a) “Known good” x fit x -dir. z -dir. (b) g SS x -dir. z -dir. (c) g RS x -dir. z -dir. − − − (d) “Known bad” x -dir. z -dir. (e) g SS x -dir. z -dir. (f ) g RS x -dir. z -dir. Chord length / µ m P r o b a b ili t y d e n s i t y f un c t i o n / µ m − Figure 18: Relative frequency of chord spanning porosities per chord length, for all samples studied, with chords extension done along x -and z -directions for each sample. The “known good” sample data along the x -axis is fitted to a log-normal distribution (solid black line),reproduced on all graphs for comparison.Table 5: Fitting parameters used to fit the chord length probabilitydensity obtained for each sample in x - and y -directions to the log-normal distribution, µ x, z and σ x, z , and mean chord length (cid:104) l x, z (cid:105) ,given in µ m.Sample reference µ x σ x (cid:104) l x (cid:105) µ z σ z (cid:104) l z (cid:105) “Known good” 3.6 0.57 44 3.7 0.63 48Smooth surface, 1 g g µg µg periments. They are also very close in size along x - and z -directions, with (cid:104) l z (cid:105) larger than (cid:104) l x (cid:105) by only 4 µ m. Thefit for this dataset ( x -dir.) is shown on all panels of Fig. 18(solid black line) to provide a comparative baseline. Com-parison of the 3D printed samples to this baseline confirmsthat all samples have slightly larger pores than our ref-erence sample. The comparative sample is sintered con-tinuously under weight, applying a constant pressure toallow porosities to close during solidification, while the3D printed samples are all compressed to a fixed heightrather than under a constant pressure: the compressionis not maintained constant during solidification. This isnecessary for assessing the powder repartition, but a con-stant compression pressure should be maintained duringthe solidification step in further manufacturing campaignsto decrease the size of porosities, thereby increasing theprints’ quality.Comparing the 3D printed samples by base-material de-pending on the g -level during manufacturing, a slight elon-gation of the pores in the z -direction is remarked for the 1 g samples. Precisely, σ z / σ x is always larger for the 1 g samples than for the µg samples, showing that 1 g samples have astronger anisotropy than µg samples. This gravity-relatedmild anisotropy is hypothesized to be due to the wave-likepowder homogenization under 1 g : during the step of theprocess consisting of shifting the powder towards the cen-tre of the printing area by horizontal shaking, under 1 g grains tend to crystallize layer-wise in the xy -plane. Suchself-ordering has been shown previously [78]; it results ina superposition of high-density, crystallized grain-like re-gions, surrounded by lower density boundaries. As thisphenomenon occurs along the shaking direction in the xy -plane, it creates voids elongated in the z -direction. This isstrongly visible in Fig. 16(c) for the RS g sample. Thisis generally compliant with the presence of a preferred di-rection under gravity; in µg the layer homogenization hap-pens through the motion of clusters of material, and thefinal 3D printed samples have higher isotropy.The µg SS sample has the highest isotropy of all 3Dprinted samples. The µg RS sample also exhibits highisotropy, despite slightly larger porosities. Generally, theraw material’s flowability ( SS or RS ) does not begetanisotropy in the printed samples.The “known bad” sample’s chord length presented inFig. 18(d) is notable. The distribution shows a longtail at large chord lengths l x, y ≥ µ m in x -direction,which does not appear along z -direction, indicating thelong horizontal porosities due to delamination betweenlayers, clearly visible in Fig. 16(d). This shows an exem-plary anisotropic sample with preferential direction of poregrowth along the x -direction, populated by many elon-gated porosities of length l x ≥ µ m.Apart from a minute elongation of porosities along the z -axis in the ground-manufactured samples, the pores’ chordlength distributions are very similar to the comparative15known good” samples: all 3D printed samples show highisotropy. The decrease in density of the RS samples islinked to an increase in pore size, but not accompanied bythe appearance of defects in the deposition. The quantityof material brought to the printing area, as well as thehomogenization time (representing the amount of materialbrought down and then to the printing-bed centre) aresufficient.
5. Conclusion and outlook
To be used for in-space applications, AM methods willneed to evolve into more versatile technologies with in-creased reliability [5, 9]. We proposed an AM process thatallows to produce parts from powder independently of thegravitational environment. This process emancipates fromthe current limitations on granular feedstock, as it doesnot rely on highly flowable powder for material deposition.Besides, it places no further geometrical constrains com-pared to ground-based PBF, while remaining superior toextrusion-based processes by allowing a wider range of ma-terials. Tested through DEM simulation and on parabolicflights, a proof of concept was provided for 1 g and µg ,using as feedstock material a 80 µ m-diameter PS powdermodified to obtain a good flowability powder (smooth sur-face, SS ) and a low flowability powder (rough surface, RS ).The RS powder was shown to have lower flowability thanthe typical 3D printing metal powder Ti-64. Analysis ofsamples that were 3D printed under gravity and in ab-sence thereof shows that the powder deposition is realizedequally efficiently under both g -levels. High reproducibil-ity is found between samples manufactured from the samebase-material in different gravitational environments, witha homogeneous pore size distribution and isotropy of thesamples. The ground-printed samples show slightly higheranisotropy, which we attribute to a layer-wise crystalliza-tion. A difference between samples realized with each basematerials persists, which is attributed to the lower abilityto pack densely for rough surface powder. Delaminationis observed in none of the 3D printed samples.All the samples analyzed in-bulk ( SS and RS , under1 g and µg ) were manufactured using the same printingparameters. The mild differences obtained in material de-position, show that the use of different material is reflectedin the process; yet all powders could be deposited and 3Dprinted. This rises the expectation that the RS powdercould be more densely packed by using better suited print-ing parameters. In particular, while the deposition processmight be ineffective in increasing packing density if thatis a material-dependent variable [32], this difference couldbe amended during solidification. Ideally, the compres-sion ratio could be increased as the solidification is takingplace, by maintaining a constant vertical pressure, trigger-ing degassing and thereby increasing part’s density, akinto hot isostatic pressing. This simple amelioration wouldallow to drastically decrease samples’ porosity, and in turn to control printed parts density (including to allow densitygradients throughout part additively manufactured).The possibility to compress the material layer after de-position also has a specific drawback: it implies that thematerial of the solidification window and the printing ma-terial are in contact before and most importantly duringsolidification. For the polymer powders used as modelsubstances in the work presented here, quartz glass wasnot reactive at the sintering temperature of polystyrene( ∼ ° C). However, if solidification happens at highertemperature – as would be the case to melt metal pow-der – the material of the solidification window would haveto be chosen accordingly, to minimize material exchangeand avoid the powder layer remaining stuck to the window.For example, the stability of a single-crystal sapphire glassplate should be investigated when exposed repeatedly tomolten metal alloys. Ensuring that no heat-induced chem-ical reaction happens will be the challenge to adapt thisAM process to metal and ceramic powders.Besides controlling parts’ density by compressing thenewly deposited powder layer, defects appearance is mit-igated by constant in-situ monitoring during powder de-position. The next step will be to automatize the controlloops (torque sensing during powder transport and imageanalysis during layer homogenization). On the one hand,homogenization time could be optimized through the im-age analysis procedure proposed, to minimize fabricationtime. On the other hand, assessment of print quality andlive-correction during manufacturing will allow AM to ac-cess a wider range of applications by increasing stabilityin prints quality, including application for space and ISM.Using those strategies to ensure high printing qualitywithout requirement on the rheology of the feedstock ma-terial will facilitate the use of recycled materials for AM.First tests on-ground have shown that powder produced byclosed-loop recycling (i.e. by grinding former 3D printedparts) can be directly used as raw material in the pro-cess presented here [64]. The possibility to not only reusematerial from previous batches, but also recycle formerobjects into new feedstock, would drastically reduce costsassociated with AM, on Earth as well as in space.Focusing on the powder handling aspect, the powder de-position process was developed to allow for the use of pow-ders regardless of their physical and rheological properties,meaning that it functions for any base-material (polymers,metals, ceramics. . . ). The powder handling method couldhence be adapted for granular transport even beyond 3Dprinting in reduced gravity environments, regardless of thematerial’s flowability. Notably, on sand-covered planetarysurface (e.g. the Moon, Mars or certain asteroids), powderhandling technologies will be necessary to process regolith,the main in-situ resource and a powder of notoriously poorflowability [79].As technological progress and space explorations will gohand-in-hand in the coming years, the authors hope thatthe AM process presented will be part of a movement tospur the development of in-space manufacturing in general,16ltimately enabling long-term human presence in space.
Author contributionsOlfa D’Angelo : conceptualization, investigation,methodology, data curation, and formal analysis, fund-ing acquisition and project administration, writing – orig-inal draft.
Felix Kuthe : investigation, methodology,data curation.
Szu-Jia Liu : investigation, data cura-tion.
Raphael Wiedey : resources.
Joe M. Ben-nett : resources.
Martina Meisnar : resources.
AndrewBarnes : resources.
W. Till Kranz : supervision, writing– review & editing.
Thomas Voigtmann : methodology,data curation, validation, supervision, writing – review &editing.
Andreas Meyer : funding acquisition, projectadministration, resources, supervision, writing – review &editing.
Acknowledgments
ODA gratefully acknowledges financial support from ESAEAC through the NPI contract 4000122340 on “Phys-ical Properties of Powder-Based 3D-Printing in Spaceand On-Ground” supported by Aidan Cowley, and of theDLR/DAAD Research Fellowship 91647576, as well as theESA Education Fly Your Thesis! 2019 flight opportunityand sponsoring through the GrainPower project, in par-ticular Nigel Savage. The entire GrainPower team is alsoacknowledged: Merve Se¸ckin, Abeba Birhane and TolgaBast¨urk, thank you. Many thanks to the Novespace teamfor their friendly assistance, and in particular Thomas Vil-latte. ODA also acknowledges Fanny Schaepelynck forher help with the schematic in Fig. 2, and the ESA-RALAdvanced Manufacturing Laboratory for making the FT4rheological measurements possible. WTK acknowledgesfunding from the DFG through grant number KR4867/2.ODA and SJL acknowledge support from the DAAD RISE2019 program under the grant number 57467143 for theproject “Characterizing powder flow in a prototype mi-crogravity powder-based 3D printer”.
References [1] R. S. Skomorohov, C. Welch, A. M. Hein, In-orbit spacecraftmanufacturing: Near-term business cases, in: Proceedings ofthe 67 th IAC, Guadalajara, Mexico, 2016.[2] A. Owens, O. D. Weck, C. Stromgren, K. E. Goodliff, W. Cir-illo, Supportability challenges, metrics, and key decisions forhuman spaceflight, in: Proceedings of the AIAA SPACE andAstronautics Forum and Exposition, Orlando, FL, USA, 2017. doi:10.2514/6.2017-5124 .[3] National Research Council of the National Academies, 3D Print-ing in Space, The National Academies Press, Washington, DC,2014.[4] T. Prater, C. Hill, F. Ledbetter, M. Fiske, J. Edmun-sson, M. Meyyappan, P. Hall, C. Roberts, L. D. Huebner,N. Werkheiser, NASA’s in-space manufacturing project: Up-date on manufacturing technologies and materials to enablemore sustainable and safer exploration, in: Proceedings of the70 th IAC, Washington DC, USA, 2019. [5] A. E. Trujillo, M. T. Moraguez, S. I. Wald, A. C. Owens, O. L.de Weck, Feasibility analysis of commercial in-space manufac-turing applications, in: Proceedings of the AIAA SPACE: Spaceand Astronautics Forum and Exposition, Orlando, FL, USA,2017. doi:10.2514/6.2017-5360 .[6] T. Prater, Q. Bean, R. Beshears, T. Rolin, N. Werkheiser, E. Or-donez, R. Ryan, F. Ledbetter III, Summary report on phase Iand phase II results from the 3D printing in zero-g technologydemonstration mission, volume I, Tech. rep., Marshall SpaceFlight Center, Huntsville, Alabama (2016).[7] T. Prater, N. Werkheiser, F. Ledbetter, D. Timucin,K. Wheeler, M. Snyder, 3D printing in zero g technologydemonstration mission: complete experimental results and sum-mary of related material modeling efforts, International Jour-nal of Advanced Manufacturing Technology 101 (2018) 391–417. doi:10.1007/s00170-018-2827-7 .[8] G. D. Goh, Y. L. Yap, H. K. J. Tan, S. L. Sing, G. L. Goh, W. Y.Yeong, Process–structure–properties in polymer additive man-ufacturing via material extrusion: A review, Critical Reviewsin Solid State and Materials Sciences 45 (2) (2020) 113–133. doi:10.1080/10408436.2018.1549977 .[9] T. DebRoy, H. L. Wei, J. S. Zuback, T. Mukherjee, J. W.Elmer, J. O. Milewski, A. M. Beese, A. Wilson-Heid, A. De,W. Zhang, Additive manufacturing of metallic components –Process, structure and properties, Progress in Materials Science92 (2018) 112–224. doi:10.1016/j.pmatsci.2017.10.001 .[10] M. Schmid, K. Wegener, Additive manufacturing: Polymersapplicable for laser sintering (LS), in: International Confer-ence on Manufacturing Engineering and Materials (ICMEM),Vol. 149, Nov´y Smokovec, Slovakia, 2016, pp. 457–464. doi:10.1016/j.proeng.2016.06.692 .[11] C. Y. Yap, C. K. Chua, Z. L. Dong, Z. H. Liu, D. Q. Zhang, L. E.Loh, S. L. Sing, Review of selective laser melting: Materials andapplications, Applied Physics Reviews 2 (2015) 041101. doi:10.1063/1.4935926 .[12] S. L. Sing, W. Y. Yeong, F. E. Wiria, B. Y. Tay, Z. Zhao,L. Zhao, Z. Tian, S. Yang, Direct selective laser sintering andmelting of ceramics: a review, Rapid Prototyping Journal 23 (3)(2017) 611–623. doi:10.1108/RPJ-11-2015-0178 .[13] J. Blum, G. Wurm, S. Kempf, T. Poppe, H. Klahr, T. Kozasa,M. Rott, T. Henning, J. Dorschner, R. Schr¨apler, H. U. Keller,W. J. Markiewicz, I. Mann, B. A. Gustafson, F. Giovane,D. Neuhaus, H. Fechtig, E. Gr¨un, B. Feuerbacher, H. Kochan,L. Ratke, A. El Goresy, G. Morfill, S. J. Weidenschilling,G. Schwehm, K. Metzler, W. H. Ip, Growth and form of plan-etary seedlings: Results from a microgravity aggregation ex-periment, Physical Review Letters 85 (12) (2000) 2426–2429. doi:10.1103/PhysRevLett.85.2426 .[14] R. Weidling, C. G¨uttler, J. Blum, Free collisions in a mi-crogravity many-particle experiment. I. Dust aggregate stick-ing at low velocities, Icarus 218 (1) (2012) 688–700. doi:10.1016/j.icarus.2011.10.002 .[15] S. G. Love, D. R. Petit, S. R. Messenger, Particle aggregation inmicrogravity: Informal experiments on the international spacestation, Meteoritics & Planetary Science 49 (2014) 732–739. doi:10.1111/maps.12286 .[16] R. A. Wilkinson, R. P. Behringer, J. T. Jenkins, M. Y. Louge,Granular materials and the risks they pose for success on themoon and mars, in: AIP Conference Proceedings, Vol. 746,2005, pp. 1216–1223. doi:10.1063/1.1867248 .[17] O. D’Angelo, T. Voigtmann, Rheology of dense granular flowsin microgravity: A Review, unpublished manuscript (2021).[18] J. G¨unster, A. Zocca, C. M. Gomes, T. Muehler, Method forStabilizing a Powder Bed by means of Vacuum for AdditiveManufacturing, United States Patent 9533452 B2 (2017).[19] A. Zocca, C. M. Gomes, T. M¨uhler, J. G¨unster, Powder-bed stabilization for powder-based additive manufacturing, Ad-vances in Mechanical Engineering 2014 (2014) 491581. doi:10.1155/2014/491581 .[20] A. Zocca, J. L¨uchtenborg, T. M¨uhler, J. Wilbig, G. Mohr,T. Villatte, F. L´eonard, G. Nolze, M. Sparenberg, J. Melcher, . Hilgenberg, J. G¨unster, Enabling the 3D printing of metalcomponents in µ -gravity, Advanced Materials Technologies 4(2019) 1900506. doi:10.1002/admt.201900506 .[21] S. Vock, B. Kl¨oden, A. Kirchner, T. Weißg¨arber, B. Kieback,Powders for powder bed fusion: a review, Progress in Ad-ditive Manufacturing 4 (4) (2019) 383–397. doi:10.1007/s40964-019-00078-6 .[22] V. Seyda, D. Herzog, C. Emmelmann, Relationship betweenpowder characteristics and part properties in laser beam melt-ing of ti-6al-4v, and implications on quality, Journal of LaserApplications 29 (2) (2017) 022311. doi:10.2351/1.4983240 .[23] P. Kiani, U. Scipioni Bertoli, A. D. Dupuy, K. Ma, J. M. Schoe-nung, A statistical analysis of powder flowability in metal ad-ditive manufacturing, Advanced Engineering Materials 22 (10)(2020) 2000022. doi:10.1002/adem.202000022 .[24] H. M. Jaeger, S. R. Nagel, R. P. Behringer, Granular solids,liquids, and gases, Reviews of Modern Physics 68 (4) (1996)1259–1273. doi:0034-6861/96/68.4/1259.15 .[25] Y. Forterre, O. Pouliquen, Flows of dense granular media, An-nual Review of Fluid Mechanics 40 (2008) 1–24. doi:10.1146/annurev.fluid.40.111406.102142 .[26] I. S. Aranson, L. S. Tsimring, Patterns and collective behaviorin granular media: Theoretical concepts, Reviews of ModernPhysics 78 (2006) 641–692. doi:10.1103/RevModPhys.78.641 .[27] H. A. Janssen, Versuche ¨uber Getreidedruck in Silozellen [Ongrain pressure in silos], Zeitschrift des Vereins Deutscher Inge-nieure 39 (35) (1895) 1045–1049.[28] N. Murdoch, B. Rozitis, S. F. Green, T. L. De Lophem,P. Michel, W. Losert, Granular shear flow in varying gravita-tional environments, Granular Matter 15 (2) (2013) 129–137. doi:10.1007/s10035-013-0395-y .[29] E. Opsomer, F. Ludewig, N. Vandewalle, Dynamical clusteringin driven granular gas, EPL 99 (2012) 40001. doi:10.1209/0295-5075/99/40001 .[30] C. S. O’Hern, L. E. Silbert, A. J. Liu, S. R. Nagel, Jammingat zero temperature and zero applied stress: The epitome ofdisorder, Physical Review E 68 (2003) 011306. doi:10.1103/PhysRevE.68.011306 .[31] M. E. Cates, J. P. Wittmer, J. P. Bouchaud, P. Claudin, Jam-ming and stress propagation in particulate matter, Physica A263 (1999) 354–361. doi:10.1016/s0378-4371(98)00491-9 .[32] R. Mari, R. Seto, J. F. Morris, M. M. Denn, Shear thickening,frictionless and frictional rheologies in non-Brownian suspen-sions, Journal of Rheology 58 (2014) 1693. doi:10.1122/1.4890747 .[33] O. D’Angelo, A. Horb, M. Sperl, W. T. Kranz, Granular piston-probing in microgravity: powder compression, from densifica-tion to jamming, unpublished manuscript (2021).[34] N. Murdoch, B. Rozitis, K. Nordstrom, S. F. Green, P. Michel,T. L. De Lophem, W. Losert, Granular convection in micro-gravity, Physical Review Letters 110 (2013) 018307. doi:10.1103/PhysRevLett.110.018307 .[35] R. Seto, A. Singh, B. Chakraborty, M. M. Denn, J. F. Mor-ris, Shear jamming and fragility in dense suspensions, GranularMatter 21 (3) (2019) 82. doi:10.1007/s10035-019-0931-5 .[36] K. Mair, J. F. Hazzard, Nature of stress accommodation insheared granular material: Insights from 3D numerical model-ing, Earth and Planetary Science Letters 259 (3-4) (2007) 469–485. doi:10.1016/j.epsl.2007.05.006 .[37] R. Niu, M. Ramaswamy, C. Ness, A. Shetty, I. Cohen, Tunablesolidification of cornstarch under impact: How to make someonewalking on cornstarch sink, Science Advances 6 (2020) 19. doi:10.1126/sciadv.aay6661 .[38] S. F. Edwards, R. B. Oakeshott, Theory of powders, Phys-ica A 157 (3) (1989) 1080–1090. doi:10.1016/0378-4371(89)90034-4 .[39] S. Nadler, O. Bonnefoy, J. M. Chaix, G. Thomas, J. L. Gelet,Parametric study of horizontally vibrated grain packings, TheEuropean Physical Journal E 34 (2011) 66. doi:10.1140/epje/i2011-11066-y .[40] S. M. Khefif, A. Valance, F. Ould-Kaddour, Spreading of a gran- ular droplet under horizontal vibrations, Physical Review E 97(2018) 062903. doi:10.1103/PhysRevE.97.062903 .[41] E. Opsomer, F. Ludewig, N. Vandewalle, Phase transitions invibrated granular systems in microgravity, Physical Review E84 (2011) 051306. doi:10.1103/PhysRevE.84.051306 .[42] M. Noirhomme, E. Opsomer, N. Vandewalle, F. Ludewig, Gran-ular transport in driven granular gas, The European PhysicalJournal E 38 (2015) 9. doi:10.1140/epje/i2015-15009-4 .[43] A. Kudrolli, M. Wolpert, J. P. Gollub, Cluster formation dueto collisions in granular material, Physical Review Letters 78(1997) 1383–1386. doi:10.1103/PhysRevLett.78.1383 .[44] E. Falcon, R. Wunenburger, P. Evesque, S. Fauve, C. Chabot,Y. Garrabos, D. Beysens, Cluster formation in a granularmedium fluidized by vibrations in low gravity, Physical ReviewLetters 83 (2) (1999) 440–443. doi:10.1103/PhysRevLett.83.440 .[45] E. Opsomer, M. Noirhomme, N. Vandewalle, E. Falcon, S. Mer-minod, Segregation and pattern formation in dilute granularmedia under microgravity conditions, npj Microgravity 3 (2017)1. doi:10.1038/s41526-016-0009-1 .[46] G. Tapia, A. Elwany, A review on process monitoring and con-trol in metal-based additive manufacturing, Journal of Man-ufacturing Science and Engineering 136 (2014) 060801. doi:10.1115/1.4028540 .[47] M. Grasso, B. M. Colosimo, Process defects andin situmoni-toring methods in metal powder bed fusion: a review, Mea-surement Science and Technology 28 (4) (2017) 044005. doi:10.1088/1361-6501/aa5c4f .[48] S. K. Everton, M. Hirsch, P. I. Stavroulakis, R. K. Leach, A. T.Clare, Review of in-situ process monitoring and in-situ metrol-ogy for metal additive manufacturing, Materials and Design 95(2016) 431–445. doi:10.1016/j.matdes.2016.01.099 .[49] S. Luding, Molecular Dynamics Simulations of Granular Ma-terials, John Wiley & Sons, Ltd, 2005, Ch. 13, pp. 297–324. doi:10.1002/352760362X.ch13 .[50] C. Kloss, C. Goniva, A. Hager, S. Amberger, S. Pirker, Mod-els, algorithms and validation for opensource dem and cfd-dem,Progress in Computational Fluid Dynamics 12 (2012) 140–152. doi:10.1504/PCFD.2012.047457 .[51] H. Hertz, ¨Uber die Ber¨uhrung fester elastischer K¨orper, Journalf¨ur die Reine und Angewandte Mathematik 92 (1882) 156–171.[52] R. Mindlin, Compliance of elastic bodies in contact, Journal ofApplied Mechanics 16 (1949) 259–268.[53] R. Mindlin, H. Deresiewicz, Elastic spheres in contact undervarying oblique forces, Journal of Applied Mechanics 20 (1953)327–344.[54] S. Lommen, D. Schott, G. Lodewijks, DEM speedup: Stiffnesseffects on behavior of bulk material, Particuology 12 (2014) 107–112. doi:10.1016/j.partic.2013.03.006 .[55] D. Vescovi, S. Luding, Merging fluid and solid granular behav-ior, Soft Matter 12 (2016) 8616–8628. doi:10.1039/C6SM01444E .[56] K. F. Malone, B. H. Xu, Determination of contact parametersfor discrete element method simulations of granular systems,Particuology 6 (2008) 521–528. doi:10.1016/j.partic.2008.07.012 .[57] J. Ai, J.-F. Chen, J. M. Rotter, J. Y. Ooi, Assessment of rollingresistance models in discrete element simulations, Powder Tech-nology 206 (2011) 269–282. doi:10.1016/j.powtec.2010.09.030 .[58] K. L. Johnson, K. Kendall, A. D. Roberts, D. Tabor, Surfaceenergy and the contact of elastic solids, Proceedings of the RoyalSociety of London A 324 (1971) 301–313. doi:10.1098/rspa.1971.0141 .[59] C. S. O’Hern, S. A. Langer, A. J. Liu, S. R. Nagel, Randompackings of frictionless particles, Physical Review Letters 88(2002) 075507. doi:10.1103/PhysRevLett.88.075507 .[60] G. Vorono¨ı, Nouvelles applications des param`etres continus `a lath´eorie des formes quadratiques. deuxi`eme m´emoire. recherchessur les parall´ello`edres primitifs., Journal f¨ur die reine und ange-wandte Mathematik 1908 (134) (1908) 198–287. doi:10.1515/crll.1908.134.198 .
61] F. M. Schaller, S. C. Kapfer, J. E. Hilton, P. W. Cleary,K. Mecke, C. D. Michele, T. Schilling, M. Saadatfar,M. Schr¨oter, G. W. Delaney, G. E. Schr¨oder-Turk, Non-universal voronoi cell shapes in amorphous ellipsoid packs,EPL (Europhysics Letters) 111 (2015) 24002. doi:10.1209/0295-5075/111/24002 .[62] P. Virtanen, R. Gommers, T. E. Oliphant, M. Haber-land, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson,W. Weckesser, J. Bright, S. J. van der Walt, M. Brett, J. Wil-son, K. J. Millman, N. Mayorov, A. R. J. Nelson, E. Jones,R. Kern, E. Larson, C. J. Carey, ˙I. Polat, Y. Feng, E. W.Moore, J. VanderPlas, D. Laxalde, J. Perktold, R. Cimrman,I. Henriksen, E. A. Quintero, C. R. Harris, A. M. Archibald,A. H. Ribeiro, F. Pedregosa, P. van Mulbregt, A. Vijaykumar,A. P. Bardelli, A. Rothberg, A. Hilboll, A. Kloeckner, A. Sco-patz, A. Lee, A. Rokem, C. N. Woods, C. Fulton, C. Masson,C. H¨aggstr¨om, C. Fitzgerald, D. A. Nicholson, D. R. Hagen,D. V. Pasechnik, E. Olivetti, E. Martin, E. Wieser, F. Silva,F. Lenders, F. Wilhelm, G. Young, G. A. Price, G.-L. Ingold,G. E. Allen, G. R. Lee, H. Audren, I. Probst, J. P. Dietrich,J. Silterra, J. T. Webber, J. Slavic, J. Nothman, J. Buchner,J. Kulick, J. L. Sch¨onberger, J. de Miranda Cardoso, J. Reimer,J. Harrington, J. L. C. Rodr´ıguez, J. Nunez-Iglesias, J. Kuczyn-ski, K. Tritz, M. Thoma, M. Newville, M. K¨ummerer, M. Bol-ingbroke, M. Tartre, M. Pak, N. J. Smith, N. Nowaczyk, N. She-banov, O. Pavlyk, P. A. Brodtkorb, P. Lee, R. T. McGibbon,R. Feldbauer, S. Lewis, S. Tygier, S. Sievert, S. Vigna, S. Pe-terson, S. More, T. Pudlik, T. Oshima, T. J. Pingel, T. P. Ro-bitaille, T. Spura, T. R. Jones, T. Cera, T. Leslie, T. Zito,T. Krauss, U. Upadhyay, Y. O. Halchenko, Y. V´azquez-Baeza,SciPy 1.0 Contributors, SciPy 1.0: fundamental algorithms forscientific computing in python, Nature Methods 17 (3) (2020)261–272. doi:10.1038/s41592-019-0686-2 .[63] C. H. Rycroft, Voro++: A three-dimensional voronoi cell li-brary in c++, Chaos: An Interdisciplinary Journal of NonlinearScience 19 (2009) 041111. doi:10.1063/1.3215722 .[64] O. Lopez-D’Angelo, Powder-based additive manufacturing forspace – From granular rheology in varying gravitational en-vironment to the development of a gravity-independent pow-der handling method, Ph.D. thesis, RWTH Aachen University,Aachen, Germany (2021).[65] A. Castellanos, The relationship between attractive interparti-cle forces and bulk behaviour in dry and uncharged fine pow-ders, Advances in Physics 54 (2005) 263–376. doi:10.1080/17461390500402657 .[66] M. Leturia, B. Mohammed, S. Lagarde, I. Ronga, K. Saleh,Characterization of flow properties of cohesive powders: A com-parative study of traditional and new testing methods, PowderTechnology 253 (2014) 406–423. doi:10.1016/j.powtec.2013.11.045 .[67] H.-J. Yang, F.-Q. Wei, K.-H. Hu, G.-D. Zhou, J. Lyu, Compar-ison of rheometric devices for measuring the rheological param-eters of debris flow slurry, Journal of Mountain Science 12 (5)(2015) 1125–1134. doi:10.1007/s11629-015-3543-5 .[68] W. Nan, V. Vivacqua, M. Ghadiri, Y. Wang, Numerical analysisof air effect on the powder flow dynamics in the FT4 powderrheometer, EPJ Web of Conferences 140 (2017) 03036. doi:10.1051/epjconf/201714003036 .[69] W. Nan, M. Ghadiri, Y. Wang, Analysis of powder rheometryof FT4: effect of particle shape, Chemical Engineering Science173 (2017) 374–383. doi:10.1016/j.ces.2017.08.004 .[70] S. Liu, Y. C. Shin, Additive manufacturing of ti6al4v alloy: Areview, Materials & Design 164 (2019) 107552. doi:10.1016/j.matdes.2018.107552 .[71] I. Zimmermann, M. Eber, K. Meyer, Nanomaterials as FlowRegulators in Dry Powders, Zeitschrift fur Physikalische Chemie218 (2004) 51–102. doi:10.1524/zpch.218.1.51.25388 .[72] J. N. Israelachvili, Van der Waals Forces between Particles andSurfaces, 3rd Edition, Elsevier, 2011, Ch. 13, pp. 253–289. doi:10.1016/B978-0-12-391927-4.10013-1 .[73] L. Heitmeier, O. D’Angelo, T. Voigtmann, Rheological response of granular materials under varying gravitational conditions: acomparative study of discrete element method simulation andexperiment, unpublished manuscript (2021).[74] J. T. Gostick, Z. A. Khan, T. G. Tranter, M. D. Kok, M. Ag-naou, M. Sadeghi, R. Jervis, PoreSpy: A Python toolkit forquantitative analysis of porous media images, Journal of OpenSource Software 4 (37) (2019) 1296. doi:10.21105/joss.01296 .[75] C. A. Schneider, W. S. Rasband, K. W. Eliceiri, Nih image toimagej: 25 years of image analysis, Nature Methods 9 (2012)671–675. doi:10.1038/nmeth.2089 .[76] J. D. Bernal, J. Mason, Packing of spheres: Co-ordination ofrandomly packed spheres, Nature 188 (1960) 910–911. doi:10.1038/188910a0 .[77] E. Marakis, M. C. Velsink, L. J. Corbijn van Willenswaard,R. Uppu, P. W. H. Pinkse, Uniform line fillings, Physical ReviewE 99 (2019) 043309. doi:10.1103/PhysRevE.99.043309 .[78] O. Pouliquen, M. Nicolas, P. D. Weidman, Crystallization ofnon-brownian spheres under horizontal shaking, Physical Re-view Letter 79 (1997) 3640–3643. doi:10.1103/PhysRevLett.79.3640 .[79] O. R. Walton, C. P. D. Moor, K. S. Gill, Effects of gravityon cohesive behavior of fine powders: implications for process-ing lunar regolith, Granular Matter 9 (2007) 353–363. doi:10.1007/s10035-006-0029-8 ..