Advances in Large-Scale Robotic 3D Printing with Plastic Pellets

Abstract

3D printing with robotics is reaching an unprecedented level of maturity both in the market and at a technological level. This paper discusses current applications of large-scale robotics applied to 3D printed real-scale final parts for the construction and product design industries, including state-of-the-art methodologies, technologies and applications. Furthermore, an in-depth view of technologies and applications developed by the author will be provided, including robot-end effector integration and the automated generation of machine code through an ad-hoc computer aided design to computer aided manufacturing (CAD-CAM) integration. This integration accounts for parametric capabilities and design-time feedback. Consequently, advances in the seamlessly integrated design-to-manufacturing workflow will be presented: (i) design (by means of employing parametric modeling software), (ii) geometry analysis (by means of ad-hoc machine manufacturing process simulations), (iii) CAD-CAM integration (by means of automated geometry processing and machine code generation), and (iv) manufacturing and testing (by combining 6-axis robots and large-scale plastic extruders).

1. Introduction

Additive manufacturing has exponentially increased its adoption and widened its use in the industry in the last 10 years after a short but significant decay in its implementation relative to former adoption expectations [1]. Its use focused on small-scale parts and rapid prototyping mainly, while allowing a decent amount of market space aside in terms of final parts for the automotive, aerospace, product design or construction industry, to name just some. One of the main issues behind this shortfall was high, unstable production costs, deriving simultaneously from both raw printing material and manufacturing costs—including operating and entry—level ones, such as high machinery prices. Besides, most industries have traditionally operated on metal materials—mostly aluminum or steel—and Computerized Numerical Control (CNC) based techniques, such as milling, drilling or similar.

Furthermore, the main drivers at a technological level for the adoption of additive manufacturing techniques, in general, and 3D printing, in particular, showed a bottom-up pattern [2]. These were generally pushed by technically-oriented workforce but barely precepted by top-management implementations, often unable to carry out the necessary innovations due to a series of impediments, namely: (i) market constraints, (ii) misleading and conflicting market interests or (iii) lack of capacity to prove the feasibility of the investment in 3D printing technologies [3]. Finally, 3D printing technology has proven unable to scale appropriately in terms of size, production costs and time, despite the efforts made to bridge this gap [4].

On the material side, despite the fact that many metals have gained presence and microstructural stability in the 3D printing market, especially in small-scale applications related to jewelry, biomedicine or high-end parts [5] are still barely affordable for most end-use products. Besides, the precision and manufacturing tolerances of 3D printing techniques with titanium, for example, are far from those reachable by milling or CNC machining techniques. Therefore, mixed manufacturing technologies [6] have augmented their importance, thus allowing firms to gradually adhere to those in order to optimize their production processes while refining their product quality and expanding to otherwise unapproachable market opportunities [7].

As a consequence, there is plenty of room to implement other materials and large-size printing technologies with wide potential use in sectors such as the automotive industry or the construction and engineering fields, where large parts are needed to replace their more factory-like, traditionally-made counterparts at an affordable price while maintaining a well-balanced material performance and consumption [8, 9]. Moreover, automation processes are nowadays commoditized, pushed by the power of Supervisory Control and Data Acquisition and Artificial Intelligence (SCADA-AI) integrative applications [10] that allow to obtain large sets of end-effector data and produce behavioral patterns for a variety of practices and routines. For instance, data pattern analysis enables firms to control the overall performance of their production lines and tools in integrated user interfaces, favoring the use of “intelligent devices” able to provide data in real time. Robots, as fully customizable and programmable machines, gain further momentum over simpler CNC-driven architectures.

This research combines the two aforementioned opportunities into a single solution while providing affordable hardware and software solutions in an attempt to democratize technology and widen its use and implementation in predominantly the engineering and construction sectors. Thus, a solution for plastic pellet extrusion is presented, combined with a 6-axis robotic arm through seamless software and hardware integration.

2. Current research and technological development

While most large-scale 3D printing technologies rely typically on 3-axis gantry-like machines, these have proven to have relevant limitations in terms of usability and scope: (i) these require large structures and initial investments, (ii) the crane-like structures must be used off-site and present constraints when located on-site, (iii) they are restricted to a horizontally layered logical structure, which (iv) restricts the geometrical capabilities of the systems, such as its use in cantilevers or inclined geometries in general. As a result, the use of robots and thereby adapted 3D printing technologies is growing steadily, activating research projects and technological advancements worldwide. Much research has been conducted by ETH Zurich, a pioneer in the development of advanced industrialized methods through an architectural lens leading the research in the field. Numerous initiatives have been carried out by this institution in the last 10 years since the creation of the Gramazio and Kohler research group [11]. ETH’s block research group presented in 2021 the first-ever pure unreinforced concrete offsite 3D printed bridge alongside Zaha Hadid Architects’ computation and design group and other industrial partners, showing a unique dry-assembled construction stabilized solely by its geometry through shape-based topological optimization [12], thus minimizing material usage [13] through a compression-only, computationally pre-optimized design. This bridge was displayed at the Venice Biennale of Architecture in 2021, allowing visitors to see the lightweight, single-layered shelled assembled panels generated by a fusion of FDM 3D printing with casting methods [14].

Further mentioned worthy efforts have been realized by market players and private companies alike. Aectual, a Netherlands-based company producing furniture and architectural products, focuses on the use of terrazzo, bioplastics and plastic pellets. The firm implements an ad-hoc setup employing 6-axis robotic arms and a regular-size pellet extruder controlled by Siemens PLCs, thus externalizing a meaningful part of their printing technology [15]. Furthermore, Aectual uses a 9-meter-long track to extend its maximum buildable volume capacity to 170cubic meters in order to create interior design elements.

Also located in the Netherlands, MX3D claims to introduce the advantages of 3D metal printing to new high-impact industries. MX3D uses a robot-mounted wire arc additive manufacturing (WAAM) system [16]. WAAM allows the use of conventional welding filler materials, which may reduce material costs drastically when compared with SLS manufacturing in a ratio of 10:1 [17], despite the high energy requirements of the system. The MX3D bridge, built with that technology at an early stage of development, took 6 months to print and required extensive testing in order to test the material’s mechanical properties and calibrate the calculus of the structural section. Printed steel’s properties and characteristics proved to differ significantly from regular steel in the elastic modulus, which affected the overall stability and stiffness of the bridge [18].

3. 3D printing design-to-manufacture. A comprehensive software and hardware solution for plastic pellet extrusion

Optimizing the manufacturing capacity is one of the most substantial aims of every production industry. These intend to achieve mass customization without sacrificing efficiency or the benefits of economy of scale in terms of return on investment. As a result, the ability to respond to customer requirements in a quick and flexible manner while keeping high version numbers at low batch sizes must increase [19]. The implementation of a customizable, programmable and fully integrated large-scale production system is of use precisely to tackle this issue at (i) software, (ii) hardware and (iii) material levels. This concept is a great advancement in the design industry, due to the time-saving factor and the subsequent cost-effectiveness component. This design method also challenges sustainability goals. It enhances the material reduction in the manufacturing process and subsequent effect of generating less waste at the end of the product lifetime.

Therefore, a comprehensive 3D printing design-to-manufacture technology is proposed, which entails: (i) A parametric or-feature based modelling tool that grants the designer the possibility of modifying an established geometry in a matter of instances by easily generating algorithms grounded on the Rhino-based visual programming interface Grasshopper or others, which allow for single- or multi-solutions-based algorithmic design [20] and receive intense optimization attention [21]; (ii) a quick-response, integrated slicing simulation algorithm capable of dealing with complex and intricate geometries of various topologies including boundary representations (BReps) and meshes [22]; (iii) a CAD-CAM translator for a variety of robot models; (iv) a visualization interface whereby production can be simulated and robot signals set and programmed; and (v) a fully-functional end-effector comprising an extruder and low-cost, reliable control electronics for (vi) pellets obtained from thermoplastics including Polylactic Acid (PLA), polyethylene terephthalate (PET) and polyethylene terephthalate glycol PETG.

3.1 Parametric modeling and geometrical analysis

Parametric modeling is a rather restrictedly employed term that defines the ability to design parts and products based on implicit geometrical relationships rather than via explicit dimensioning. This associative way of modeling and depicting solids and other topologies—such as BReps, meshes, and others—relies on the ability of 3D CAD platforms to define geometry through either (i) a history-based hierarchical object-dependency tree, (ii) customizable parameters and equations, (iii) programmable functions or (iv) a combination of the above. Each 3D modeling software offers a variety of interfaces, including but not limited to (i) equation editors, (ii) visual programming interfaces, (iii) application programming interface (API) accessibility through programming interfaces or a (iv) combination of any of those. McNeel Rhinoceros is used for testing and programming purposes, as well as a platform to program and extend its built-in modeling capabilities. Rhinoceros is a relatively lightweight NURBS-based software able to produce and handle all sorts of geometry, which constitutes an ideal system for advanced users that like to generate their own custom scripts, create generative systems [23], or create logic models.

For the purpose of this research, a set of different geometries comprising numerous variable conditions are tested, such as those displayed in Figures 1 and 2. These display entirely parameterized sets of sizes, curvature settings, cantilevers and overhangs, among others, allowing the designer to modify and adapt the design according to manufacturing results and analyse optimal printing setup and material results in the printed part. Besides, three main 3D printing set-up-related parameters are implemented in the design of the parts: (i) nozzle diameter, (ii) shell overlap (which define the overall wall thickness for the part), and total part length, which relates simultaneously to printing speed and temperature cool down—the latter crucial to the successful layering and cohesion of the parts. Parts were designed hollow and single or two-cycled (this is, employing just one or two outlines per layer). Figures 3 and 4 show the analysed printed configuration and the variable parameters used throughout the design process that defines the non-uniform rational Bezier spline (NURBS) model. Please note that the single-cycle chair design considers variable separations between contours in order to test the actual part’s thickness as compared to the nominal size and adjust material flow.

Figure 1.

3D printing of a Striatus bridge block. Studio Naaro.

Figure 2.

Software and hardware layers. Integration of electronics.

Figure 3.

Test parts (I). Single-cycled chair-like design showing dimensioning.

Figure 4.

Test parts (II). Hopper showing dimensioning constraints.

Table 1 shows the relationship between material flow, layer height setup, resulting part thickness (for a single wall pass) and qualitative results at a printing speed of 40 mm/s. These affected the design tolerances to a great extent, showing that speed could be adjusted to ranges of 60mm/s to 200mm/s if desired. Optimal material flow was found at 375.5 to 392 mm3/s at speed rates of 40 mm/s to 60mm/s, although further research must be conducted to reduce data dispersion. As seen in Table 1, layer heights could also be modified accordingly.

Material flow adjustments
Flow (mm3/s)	Layer height (mm)	Part thickness (mm) real/nominal	Results
300	0.8	8.6–9.8/1.25–1.8	Increase layer height; Increase flow; Appropriate adhesion
325	0.8	9.5–11.2/1.25–1.8	Increase layer height
350	1	8.1–9.6/1.25–1.8	Appropriate flow; Appropriate adhesion
375	1	8.8–10.1/1.25–1.8	Increase layer height; Appropriate flow; Appropriate adhesion
400	1	9.4–11.4/1.25–1.8	Regular adhesion; Increase speed
450	1	10.56–12.2/1.25–1.8	Regular adhesion; Increase speed
400	1.25	7.3–8.6/1.25–1.8	Regular adhesion; Excessive flow; Undesired overhangs
450	1.25	8.4–9.7/1.25–1.8	Regular adhesion; Excessive flow; Undesired overhangs

Table 1.

Material flow adjustments and results.

4. Pre-processing for manufacturing. CAD-CAM integration

Geometrical processing entails a three-step software-based workflow: (i) robot model and overall setup, (ii) model slicing and (iii) target generation. The software, an extension of Rhinoceros’ Grasshopper visual interface, as indicated above, is capable of processing both mesh-like and surface-like topologies, including solids. From a mesh point of view, a wide range of extensions are allowed, including but not limited to *.stl, *.igs, *.obj, *.ply, *.msh and similar mesh-compliant file formats. This functionality is provided by the platform itself. Surface-like and solids can also be processed as BReps, which support ASCII encoding; some formats allowed are *.3 dm (native Rhinoceros format), *.brep, *rle, *.step and *.sat, widely used in engineering and product design. Solids or surface-like geometries may be translated into meshes to the user’s intent, who will decide fabrication tolerances or further geometrical affections (Figure 5).

Figure 5.

Programming workflow diagram.

The workflow comprises the following stages:

1. Robot model and overall setup. Employing the built-in functionality of the robot’s extension on Grasshopper, a robot model is provided. Figure 6 depicts the selected ABB IRB 1200 for the development and experiment’s case studies. The setup includes also: (i) the design and input of an end-effector, including its gripper or adaptor to the robotic arm and the (ii) configuration of printing parameters, namely target’s speed and precision zones. For the purpose of the present experiment, a range of 40 mm/s to 200 mm/s was used as maximum linear printing speeds and a redefined z0 as a precision zone for the pass-by toolpath points. Further robot parameters are also set up, such as elbow configuration and digital output (DO) naming.

2. Model slicing. The imported or natively generated model is sliced by equally distanced planes according to the manufacturing requirements. A range of 1 to 2 mm is used. The slicing operation results in polylines, which are then divided either by length or by curvature to obtain the points that will become the targets’ origin by assigning them an orientation plane.

3. Target generation. A horizontal plane in each point is generated, providing the actual orientation for the tool (in this case, vertical). The movement interpolation type for the targets is set to linear. Additional home or safe positions are programmed for the purpose of safety and further program control. Finally, the electronic control output is fed into the desired targets, allowing to turn the extruder either on or off when needed. In this case, continuous printing is tested, thus demanding a single turn-on and turn-off cycle after completion of the part’s printing.

4. Program generation. Targets are combined through simple list-management python-written routines. These targets are stored in a linear array and then fed as a single data stream into the program generator routine.

5. Program simulation. The program is simulated showing errors or warnings in execution time. Warnings may include robot configuration issues, while errors prevent the program from running, and might refer to (i) out-of-boundaries errors, (ii) “close to singularity” instructions and (iii) collisions or similar faults related to the mechanical unit.

6. Program export. The program is exported and saved to a user-defined folder. In the case of the current example, a RAPID program (*.pgf) is generated, comprising XML-written data that refers: (i) a MAIN module (*.mod file) including variables and overall setup and (ii) additional modules containing movement instructions and other commands (additional *.mod files). Due to the used IRC5 limited RAM size, modules are split into 32,000-long instruction modules. These contain as well dynamic memory loading and unloading. Further considerations must be implemented in order to control the extruder’s behavior while the movement modules are loaded or unloaded from memory, in order to prevent material overflow at stop points.

7. The program is loaded onto the controller and runs normally.

Figure 6.

Test part (II). Hopper in a simplified simulation. Printing bed width 700 mm.

Figure 6 above demonstrate the software’s flexibility and accuracy in depicting the system’s behavior via an ad-hoc geometry processing algorithm. Figure 7 depicts the corresponding printing stages on the 700 mm-wide printing bed proving the workflow’s feasibility.

Figure 7.

Test part (II). Hopper in actual printing process. Printing bed width 700 mm.

5. Manufacturing and testing

As introduced above, the manufacturing test setup conveys a series of components, entailing mechanical, logical and ad-hoc control units. The main components are shown in Table 2, including robot, end-effector, feeding system and printing bed.

Main components
Robot
Mechanical arm	ABB IRB 1200 H 7–70
Controller	ABB IRC 5 Compact
End effector	Adapted pellet extruder
Size	150 mm x 100 mm x 495 mm
Weight	5.25 kg
Extruder motor type	Stepper (Nema)
Extruder motor torque	3 Nm
Material flow rate	<500 mm3/s
Transmission type	Planetary gear 1:10 ratio
Heating	Direct coil heating
Coil number/specs	3220V coils w/ PT 100thermistors
Electronics	Ad-hoc designed control, Arduino MEGA boards, SSRs
Adaptor	Ad-hoc designed, waterjet cut adaptor
Cooling	Liquid cooling
Feeding system	Venturi valve with pressurized inlet
Air pressure	7 bars
Printing bed	Glass, 700 mm x 700 mm x 4 mm on aluminium profiles
Type	Non-heated

Table 2.

Main components of manufacturing unit.

The main component for the present setup is the robotic arm ABB IRB 1200 with the IRC 5 compact controller. Secondly, an adapted pellet extruder is used. Heating is delivered by three heating coils that work at 220 volts and are controlled by 3 PT100 type thermistors. An Arduino code developed ad-hoc for the purpose runs on an Arduino Mega 2560 board with an added ramps 1.4 to control the three solid state relays (SSR) that regulate the three 220 V heating coils. Thirdly, with the objective of controlling the extruded material flow a Nema 23 stepper is used, which supplies a maximum torque of 3 Nm. Fourthly, an air-refrigerating system is mounted on the nozzle to cool down the printed material, comprising a 4 mm aluminum pipe curved around the tip of the nozzle. The extruder body is cooled by liquid refrigerant pumped from a tank placed aside the robotic arm. The printing bed consists of a 4 mm thick glass surface covered with a single layer of Scotch 3D printing tape, which enhances the adhesion of the part to the bed (Table 3).

General printing parameters
Printing speed	Per part and layer (below/above 10 initial layers)
Printing type	Continuous
Nozzle diameter	1.5 mm
Layer height	Per part and layer (below/above 10 initial layers)
Support required	No, geometrically feasible
Cooling
Extruder refrigeration	Refrigeration liquid cooling
Layer cooling air pressure	3 bars
Layer cooling air temperature	18°C
Layer cooling air humidity	40–55%
Adhesion
Adhesion type	None, direct layering
Adhesion material	Adhesive tape on printing bed

Table 3.

Printing parameters.

Whereas the robot program includes the necessary DO-related commands, the wiring was set up to minimize the need for electronic components and coding. An Arduino MEGA board is used to control the overall behavior of the end-effector (this is, to turn the stepper motor on or off according to needs) while keeping track of the temperature regulation in each coil, as mentioned previously.

ABS was not tested due to its thermal instability while printing and in order to prevent warping. Additionally, ABS is not a suitable printing material in open spaces and in the absence of properly heated printing beds. Solid-state relays are employed as gates to power the coils, which cool down under regular atmospheric circumstances.

Table 4 shows the test part specifications including operating time for both specimens. Note that the single-cycled design presents tighter layer heights and a lower printing speed at the same material flow rate, thus expelling more material per length unit. A close-up view of the part’s layers is displayed in Figure 8, depicting geometry inaccuracies caused by intricate geometry and low precision zones (Figure 8 left). Furthermore, material burns can be seen in intermediate layers (Figure 8 left).

Figure 8.

Closeup of single-cycled design showing irregular geometry, material burns (left) and warping (right), scale bar shows a mark every 10 mm.

Figure 9.

Closeup of single-cycled design showing test part (left) and regular layering (right). Scale bar shows a mark every 10 mm.

Printing parts specifications
	Single-cycled chair-like design	Hopper
Size LxWxH	440.4 mm x 387.4 mm x 200 mm	334 mm x 334 mm x 329.8 mm
Layer type	Single continuous, 2 walls	Single discontinuous, 2 walls
Layer height (<10)	0.6 mm	0.65 mm
(>10)	0.8 mm	1 mm
Layers	252	333
Printing speed (1)	40 mm/s	60 mm/s
Printing speed (2)	40 mm/s	80 mm/s
Printing time	18377.72 s	3932.08 s
Printing length	735109.04 mm	311119.67 mm
Material	PLA	PLA
Material temp.	200 °C / 220 °C / 220°C	200°C / 220°C / 220°C

Table 4.

Test part specifications.

In addition to material burns, an excess of remaining latent heat in a layer can result in undesired material fluidity, causing the part to collapse partially or completely. More importantly, this may cause the nozzle to overheat, melting already printed parts or layers, thus affecting the overall stability and geometry in undesired ways. This effect may be seen in Figure 8 right above.

Figure 10.

Collapse of hopper’s tip due to heat accumulation and part’s layer quality (detail and overall).

In addition to adjusting printing speed per layer perimeter, a layer cooling system was developed and mounted (Figure 9). Also, geometry needs to be reviewed to minimize the amount of material at that height (Figure 10).

Tested materials include PLA and PETG presented in spherical pellet form and flake shape. Sizes are constrained to a maximum diameter of 6 mm for the former and up to 6.25 mm in any direction for the latter. The main characteristics of the used materials are displayed in Table 5 for clarity purposes.

Material specifications (spherical and flake pellets)
	Polylactic acid (PLA)		Polyethylene terephthalate glycol (PETG)
	Typical value	Test method	Typical value	Test method
Material density	1.24 $\mathrm{g}/{\mathrm{cm}}^3$	ISO 1183	1.27 $\mathrm{g}/{\mathrm{cm}}^3$	ASTM D792
Glass transition temp.	60 $°C$	D3418	85$°C$	D3418
Tensile strength at break	50 MPa	D882
Tension yield strength	60 MPa	D882	50 MPa	ASTM D638
Rockwell hardness (R Sc.)			108 R	ASTM D785
Tensile modulus	3.5 GPa	D882
Tensile elongation	6%	D882
Notched Charpy Impact	5 $\mathrm{KJ}/{\mathrm{m}}^2$	ISO-179-1eA
Notched Izod Impact			105 J/m	ASTM D256
Flexural strength	83 MPa	D790	69 MPa	ASTM D790
Flexural modulus	3.8 GPa	D790	2100 MPa	ASTM D790
Heat distortion temp.	55$°C$	E2092	70$°C$	ASTM D648
Pellet (spherical)—new
	Nominal	Actual	Nominal	Actual
Shape	Spherical	Irregular ellipsoid	Spherical	Irregular ellipsoid
Size	ϕ 2.5 mm	Variable ϕ 2 -ϕ 4 mm	ϕ 2.5 mm	Variable ϕ 1.5 -ϕ 3 mm
Pellet (flake)—reused
Shape	Irregular planar polygonal flake	Irregular non-planar polygon	Irregular planar polygonal flake	Irregular non-planar polygon
Size	< 10 mm	<7.5 mm	<10 mm	<7 mm

Table 5.

Printing material parameters.

Two main material arrangements are employed: (i) the use of purely spherical pellets alone and (ii) a combination of spherical and flake-like pellets. These tested material combinations include a series of mixes that allow checking for the feasibility of employing recycled and reused materials alongside new ones. Spherical pellets are provided as new, while flake ones are recycled from a variety of untraceable sources.

Therefore, their specifications and applications may only be deemed as approximate values. Table 5 specifies the results and issues tested on different material mixes. As shown, most reliable results are obtained when combining new pellets and recycled ones in a proportion of 80–20% or higher due predominantly to two factors: (i) the substantial decrease in the reliability of the venturi feeding system while using flakes even at high air pressures and (ii) the affection on printing temperature requirements when combining different material sources, some of which might include unknown origins and previous physical or chemical treatments or pre-processing (Figure 11 right).

Figure 11.

Photography of ϕ2–ϕ4 mm PLA spherical and ϕ1.5–ϕ3 mm flake-shaped pellets used in the experiment.

6. Results and conclusions

In terms of size, the presented robot-based 3D-printing system is easily scalable to produce as-large-as-required parts simply by modifying or selecting robot models with a higher movement range, such as an IRB 1600, IRB 2600 or higher. Above mentioned parameters such as printing speed or layer air-cooling might need further adjustment, despite the fact that larger parts allow more time for layers to cool down, which plays a crucial role in the stability of the parts while printing. In other words, larger sizes should be easier to print.

Special attention should be paid to warping, which highly depends on the geometry of the parts. The single-cycle chair-like design presents warping, in every instance of the process, exactly where the geometry is more acute and the part’s printing length is higher, causing the part to be less condensed or present a lower level of adhesion to the printing bed. Warping, as opposed to layer cooling, will become a challenge with bigger parts.

Results show that a proper material mix combining new-to-used PLA and PETG is suitable for large-scale printing as shown in Table 6. Proportions of 80% new spherical pellets to 20% reused flake-shaped pellets and higher work properly. Whereas printing with pellets from equal thermoplastic types is achievable, mixing PLA and PETG has proven pointedly a more delicate task. Nonetheless, manufactured parts have proven to be stable and are well preserved through time, thus evidencing the feasibility and reliability of the entire system. The implications of this technology on the circular economy and environmental impact are left aside for future discussion.

Material combination results
	Spherical polylactic acid (PLA) pellets		Spherical polyethylene terephthalate glycol (PETG) pellets
	Proportion spherical*	Proportion PLA flakes*	Proportion spherical*	Proportion PLA flakes*
Proportion 1	100%	0%	100%	0%
Result	Satisfactory printing; Not satisfactory; Not printing		Satisfactory printing
Issue	None		None
Proportion 1	60%	40%	60%	40%
Result	Not printing		Intermittent printing
Issue	Feeder not fed Material not sucked into feeding pipe Venturi system failure		Material feeding interrupted Unpredictable feeding behavior Interrupted layer printing Unstable mixture-temp
Proportion 2	65%	35%	65%	35%
Result	Not printing		Intermittent printing
Issue	Feeder not fed Material not sucked into feeding pipe Venturi system failure		Material feeding interrupted Unpredictable feeding behavior Interrupted layer printing Stepper motor misses steps Unstable mixture -temp
Proportion 3	70%	30%	70%	30%
Result	Irregular printing		Intermittent/irregular printing
Issue	Material feeding interrupted Partially unpredictable feeding behavior Irregular layer printing		Material feeding interrupted Partially unpredictable feeding behavior Irregular layer printing Stepper motor misses steps Unstable mixture—temp
Proportion 4	80%	20%	80%	20%
Result	Regular, medium-quality printing		Low-quality printing
Issue	Flakes are dragged by spherical pellets Warping Occasional material defects		Unstable mixture—temp Increased warping Unreliable printing

Table 6.

Material combination results.

^*

In terms of weight.

Further research and funding will be required to challenge the thesis hereby presented and to be able to print bigger, more intricate geometries. Also, a significant part of future work will focus on the development of a more stable extruder, despite efforts made in that direction.

Acknowledgments

Special thanks are given to Archiologics and its staff for their invaluable support and for providing funding, materials, hardware and valuable knowledge. This research would have been impossible without them.

Conflict of interest

The authors declare no conflict of interest.