Abstract
Desktop three-dimensional printing (3DP) with the fused filament fabrication technique is widely employed for the manufacture of small-scale horizontally layered elements with a uniform striated appearance. What remains a challenge is developing printing processes that can automate the construction of more complex large-scale elements with a distinct fluid surface esthetic for architectural design applications. To address this challenge, this research explores the three-dimensional (3D) printing of multicurved wood–plastic composite panels that have the appeal of natural timber. It compares six-axis robotic technology and its ability to rotate the axes to print smooth curved layers in complex shapes with a large-scale, gantry-style 3D printer that is predominantly used for creating fast, horizontally aligned linear prints typical of 3D printing toolpathing. The prototype test results demonstrate that both technologies can produce multicurved elements with a timber-like esthetic.
Keywords: aesthetics, additive manufacturing, curved surfaces, fused filament fabrication, wood plastic composite
Introduction
Three-dimensional printing (3DP) with fused filament fabrication (FFF) is a well-known technique whereby a thermoplastic filament (or composite thermoplastic filament) is fed through a heated three-dimensional (3D) printer head that melts and deposits the material in horizontal layers to build an object.1 The 3D printing process converts a digital 3D model into movement instructions for the printer (referred to as G-code), using a model slicing software (such as Cura or Simplify 3D).
However, the horizontal printing process limits design possibilities and results in the striated appearance of 3D printed elements. For desktop scale printing, striation is acceptable as it is used as a low-cost method to create models and prototypes quickly for general consumers or used where speed and cost are of greater importance than print quality and appearance.2
Curved layer fused filament fabrication (CLFFF) not only achieves a better and less striated appearance but is also limited in scale. It creates sloped or curved rather than stepped or striated layers by varying the vertical z-axis, in addition to the horizontal x- and y-axis, coordinates and thus generates a series of layers that more closely resemble a homogeneous surface.
Several different 3D printing mechanisms have been explored that use variations of CLFFF, including the Stewart mechanism that rotates the base of the print, delta-style printers that utilize a tool that can move rapidly in the vertical axis, and open-source syringe extrusion using thixotropic materials that only flow during extrusion processes.3–6 However, all these setups have the disadvantage of a limited machine working size, which restricts the opportunities to produce larger scale elements for architectural building elements such as wall panels.
Building scale can be achieved with big area additive manufacturing (BAAM), which scales up the production of 3D printed products, using a variety of polymer powders, pellets, and additives to create the desired composite material. BAAM printers omit the process of filament fabrication—they melt the mixed raw material in an extruder and directly deposit it through a horizontal layering process.
Both large-scale gantry-style and robot systems have been demonstrated in the additive manufacture of concrete walls at building scale, a technique known as contour crafting.7,8 BAAM systems have the potential to construct any object unbounded by size. The concept of this system is comparable to an interconnected group of robots working on different areas of an additively manufactured vehicle, with the ability to coordinate on how to deal with overlapping regions9; however, this concept is still in its infancy.
The main challenge in using either the traditional FFF technique or large-scale BAAM system is eliminating the stepping effect that is created as subsequent layers are laid to create angled or inclined surfaces. This effect is amplified when the production scale is increased.10 The resulting surfaces are rough and marked with striae and require additional surface cladding or sanding to hide the layers.
To make large, 3D printed elements more marketable, a smooth curvature and surface pattern is required to address esthetic criteria, facilitate ease of cleaning, and eliminate the need for additional cladding or postprocessing. Wood composites are ideally placed to achieve this.
The expression of warmth from wood products11 and their versatile structural applications have made natural wood a popular material for building envelopes.12 Under heat, wood changes color and this can be controlled through the 3D printing process to create patterns. Wood substitute products enhanced the marketability of components by utilizing a thin veneer of wood from a timber log, plastic laminated boards embedded with a wood grain printed on paper, or a combination of both veneer and laminates, with many patents spanning several decades detailing the techniques of creating a wood surface finish.13–16
These techniques are mainly limited to flat or single-curvature elements. Recently, an innovative technique using voxel (volumetric pixel) printing with a Stratasys PolyJet printer enabled the replication of both the exterior and interior of a wood sample by mapping the texture of wood on each cross-sectional layer and printing all the horizontal layers17; however, this requires a specific printing technology in a limited volume, which may not be applicable for large-scale structural applications.
Prior research demonstrated that it is possible to achieve a timber-like pattern in flat 2D surfaces and create varying tones by manipulating the toolpath and temperature of the hot end in FFF 3D printing with a wood-plastic composite (WPC) filament.18–20 The aim of this study is to overcome the limitations of current processes and materials, by developing production processes for additively manufactured WPC elements with complex multicurvature geometries, and incorporate a timber esthetic.
There is growing interest in creating alternate wood products that resemble the texture and appeal of wood while also preventing the presence of imperfections and flaws in wood, such as splinters, warping, and knots.21 These undesirable features are absent in WPC products such as macroextruded decking or screening products.22 In contrast to natural wood, WPC products are better suited for outdoor applications due to their resistance to weathering and water absorption. Additive manufacturing of WPC provides an additional benefit of reducing waste compared with traditional wood manufacturing processes and allows for greater customization.
This project looks at developing a manufacturing protocol to obtain a surface finish that is representative of a curved timber surface, which currently can be achieved through subtractive manufacturing, but not through conventional additive manufacturing technologies with horizontally layered, FFF, 3D printing techniques. This research will compare the use of a large-scale robot and FFF 3D printer to optimize the surface design of a fabricated WPC component.
Method: Overcoming Limitations of 3D Printing to Produce Complex Geometries
While printing 2D layers has been the norm in the 3D printing process, the challenge lies in the production of 3D curved layers. Prior studies have successfully printed curves at an angle perpendicular to the horizontal base, but most techniques are severely limited in their ability to align the print nozzle to be normal to the curved surface.4–6
Similarly, exploration of timber-like patterns has been limited to 2D layering processes.20 Currently, there are no studies on how to produce a timber-like esthetic on multidimensional surfaces. Hence, the aim is to reproduce the appearance of wood in 3D printing by using an FFF toolpath, which is applied as a curved 3D printed surface.
Two methods used to produce curved layer prints will be compared. The first method uses a six-axis robot with a custom FFF tool that can be rotated to any orientation. The second method uses a large-format, FFF 3D printer with a traditional gantry-style operation controlled by movements predominantly in the Cartesian coordinate system.
Principle of curved layer printing
The objective of robotic 3D printing is to print a curved layer by aligning the nozzle so that it is normal to the printing surface at any time. By aligning the nozzle to follow the surface during the printing, a surface free of steps is created, avoiding the stepped layer effect from a conventional, FFF 3D printer that prints horizontal layers only, as shown in Figure 1. Aligning the nozzle such that it is normal to the surface ensures that the nozzle is at a similar height away from the surface across the entire extrusion width, which theoretically creates a more homogeneous surface.
FIG. 1.

(A) Nozzle aligned vertically, which deposits the filament unevenly; (B) nozzle aligned normal to the surface, which distributes the filament evenly on the surface; (C) red layers show the stepping effect from horizontally layered 3D printing, while the gray dashed curve shows the ideal curve that can be printed from the curved layer technique. 3D, three-dimensional. Color images are available online.
Toolpath generation
To produce the 3D printed product from a model, a script is generated in Grasshopper, which is a visual scripting program that is associated with the computer-aided design (CAD) modeling program, Rhinoceros. Grasshopper utilizes easy-to-manipulate connecting blocks, as opposed to typical lines of text, allowing changes to the script to be visualized in a straightforward manner. Visualization of the toolpath aids in performing a digital validation of the machine's movements to determine whether there are any collisions of any component during the 3D printing process.
Toolpath generation for robotic 3D printing
To generate the script for robotic 3D printing, the initial surface is converted to a series of toolpath lines, which represent the path the tool on the robot will take while extruding the filament. The toolpath lines are then divided into a series of Cartesian coordinates at appropriate spacing to provide an accurate representation of the surface while not overloading the system with unnecessary commands. At each point, the toolpath is divided at ∼5-mm intervals. The normal of the surface is calculated and added to each point as a series of x-, y-, and z-axis rotations.
The programmed movement of the robot arm needs to correspond to the extrusion of filament through the extrusion head. Since this is performed externally to the robot system, a signal is placed at regular intervals, which will act as an instruction for extrusion of the filament. All these data are saved into a file that the KUKA robot can read, using the Grasshopper add-on KUKA PRC to generate the resultant script with the src file format. This process is shown in Figure 2.
FIG. 2.

Process to create an src file for the KUKA robot (from top to bottom). Surface model; toolpath lines; Cartesian coordinates of divided toolpath lines; surface planes that are tangential to chosen Cartesian coordinates, to determine the rotation angle of the tool head; extrusion signal; and exported as a KUKA src file. Color images are available online.
Toolpath generation for the large-format, FFF 3D printer
The process to generate the G-code for a traditional 3D printer begins in a similar manner to the robot by generating a toolpath for the lines and dividing up the toolpath to obtain Cartesian coordinates of the desired accuracy. No rotational plane is calculated since the tool head will always be vertically aligned. Extrusion values are calculated based on the distance traveled between each coordinate and the required layer thickness. All these data are saved into a G-code file for the FFF 3D printer to read and print. This process is shown in Figure 3.
FIG. 3.

Process to create a G-code file for the FFF 3D printer (from top to bottom). Surface model; toolpath lines; Cartesian coordinates of divided toolpath lines; extrusion length; and exported as a G-code file. FFF, fused filament fabrication.
Equipment setup
Robotic 3D printer equipment setup
To perform the movements following the toolpath in the KUKA src script, a KR6 R900 KUKA six-axis robot was used from the facilities in the Design Modelling and Fabrication (DMaF) Lab at the Sydney School of Architecture, Design and Planning at the University of Sydney. The KUKA robot was installed with a custom-fabricated end-effector, consisting of an E3D hot end and a Titan extruder connected to an FFF 3D printed part assembly, as shown in Figure 4.
FIG. 4.

(A) Custom end-effector for FFF 3D printing installed onto a KUKA robot; (B) close-up of the end-effector on the KUKA robot, showing the E3D hot end and the Titan extruder. Color images are available online.
While movements are controlled by the KUKA robot, an Arduino Mega 2560 and RAMPS 1.4 setup is used for filament extrusion and temperature control of the hot end and heated bed. Since an open-source setup with the Arduino Mega 2560 and RAMPS 1.4 was utilized, such components could also be integrated as a part of the production system with any KUKA robot and to reproduce the printed product using another system in any location.
The interaction between the KUKA robot and Arduino setup is highlighted in Figure 5. The KUKA robot utilizes the src script to maneuver the robot arm. The src script in the KUKA system is programmed to send a digital signal through an Ethernet connection from the KUKA robot to a programming script on the computer, created using the open-source computer programming software, Processing.
FIG. 5.

Communication diagram between the KUKA robot and Arduino.
Upon receiving the digital signal from the KUKA robot, the Processing script sends a signal through a USB connection to the Arduino Mega 2560 to extrude the filament while the robot arm is moving. The Arduino setup provides feedback on the extrusion process; once the filament is extruded, the Arduino setup is ready to execute the next instruction from the Processing script, which in turn relies on a signal triggered from the KUKA src script. The Arduino setup also provides feedback on the temperature of the hot end and heated bed.
Large-format, FFF 3D printer equipment setup
To print the custom G-code, the 3D Platform Workbench FFF 3D printer was used. The printer can print a build volume of 1000 × 1000 × 500 mm3. The 3D printer was operated in a stand-alone format by reading an SD card inserted into the control panel. The x- and y-axis motions are operated through a high-angle threaded rod, while the z-axis uses a low-angle threaded rod. The printer executes the instructions from a G-code file in a similar manner to desktop, FFF 3D printers.
Mimicking properties of wood
A custom-built script was developed to generate a 3D printed surface that mimics the features of timber, such as the variation in wood grain distribution. To achieve this goal, a new algorithm was developed that seeks to replicate aspects of timber grain texture. The texture is controlled by adding regions of influence, using the Boid and Anemone plug-ins in Grasshopper to deform the print path at varying magnitudes and in multiple directions, as shown in Figure 6.
FIG. 6.

(A) Schematic of the multiple regions of influence, which vary in magnitude, to distort the originally horizontal toolpath; (B) multidirectional influencers produce the wavy texture that is reminiscent of wood; (C) a timber-like pattern that is printed using the KUKA robot with a Laywoo-D3 WPC filament. WPC, wood-plastic composite. Color images are available online.
The magnitude of the region of influence can be controlled to determine how far the lines deflect away from the center of influence. The ratio of inward versus outward deformation of the influencing regions, as well as their positions, determines the extent of waviness of the wood texture, which in turn is influenced by the natural waviness of timber products, including timber laminates.
A portion of a larger pattern was printed with a WPC filament, as shown in Figure 6, which demonstrates how the regions of influence affect the toolpathing and resultant texture. Toolpaths that are overlapping appear to be bunched up and rougher to the touch, which can be reminiscent of rough-sawn wood or wood with weathered grains.
Results: Creating Wood-Like Surfaces
Applications of manipulated toolpaths onto a complex surface will be demonstrated to show how a timber-like surface can be generated.
Timber-like complex surfaces
Using the techniques with optimized curved layer 3D printing that reduces travel movements, curved WPC layers were printed using a Laywoo-D3 filament, including algorithms to create waviness in the grain that mimics the esthetics of wood. Variations in the toolpath were programmed using deviations in the wood grains perceived from a variety of wood products as points of reference.
Timber-like complex surfaces with robotic 3D printing
Two types of 3D printed molds were used as the base for the surface that was created with Laywoo-D3: a double-curvature sine wave surface and a pseudorandom Perlin noise surface. A sine wave surface is mathematically defined based on circle geometry coordinates, whereas the Perlin noise surface does not have a definitive mathematical function that describes the surface. Both 3D printed molds were printed with a blue acrylonitrile butadiene styrene filament. On both molds, two types of timber patterns were projected to follow the mold surface shape, covering an area of 200 × 200 mm.
Toolpaths were generated and incorporated in the print using the KUKA robot setup described previously. The double-curvature sine wave form was printed with a thick, continuous extruded filament that followed the surface form, as shown in Figure 7. The resultant layer printed on the double-curvature sine wave surface can be a stand-alone product or it can be integrated into a finalized 3D printed product.
FIG. 7.

(A) KUKA robot 3D printing on a double-curvature sine wave surface with a Laywoo-D3 filament on an ABS base printed in advance using a conventional FFF printer; (B) close-up of the second double-curvature sine wave WPC print. ABS, acrylonitrile butadiene styrene. Color images are available online.
The two timber-like patterns printed on the pseudorandom Perlin noise surface exhibited signs of extruded filament breakage and more uneven surface texture when compared with the double-curvature sine wave surface, as shown in Figure 8. This can be attributed to the toolpath that is pieced together by assuming that the surface is made up of a series of planar surfaces when there is continuous curvature.
FIG. 8.
(A) KUKA robot 3D print on a pseudorandom Perlin noise surface with a Laywoo-D3 filament on an ABS base printed in advance using a conventional FFF printer; (B) close-up of the second pseudorandom Perlin noise surface WPC print. Color images are available online.
The flat end of the nozzle tends to spread the excess extruded filament differently when compared with the required distribution of material on the curved surface since the gap between the print surface and the nozzle is more varied on surfaces with higher curvature. The differences in the shape of the base mold that contributes to the degree of curvature in the final prints are outlined in Table 1.
Table 1.
Geometric Features of the Base Mold and Final WPC Three-Dimensional Prints
| Description | Sine wave surface | Perlin noise surface |
|---|---|---|
| Average height between local high and local low points | 30 mm | 7.5 mm |
| Average horizontal distance from local high point to local low point | 100 mm | 28.7 mm |
WPC, wood-plastic composite.
Overall, gentle variations in curvature appear to be more successful in portraying timber-like textures than a surface with large variations in curvature over short distances.
Timber-like complex surfaces with a large-format, FFF 3D printer
For comparison, a 3D print was performed on the 3D Platform printer. Only a Perlin noise surface was used since the curved surface of the double sine wave form has much larger differences in vertical height, which may collide with the hardware surrounding the tool head.
A 400 × 400 mm2 base was created with a Perlin noise surface on top using the Laywoo-D3 filament. A WPC filament was chosen for the base to provide more dimensional stability to accurately print the curved layer on top since pure thermoplastic filament prints of this scale would be subject to warping issues. The resultant surface is shown in Figure 9. In contrast to the toolpaths printed on the Perlin noise surface with the robot, the FFF 3D printer produced more consistent toolpath deposition with no breakages during the print.
FIG. 9.

(A) 3D Platform 3D print on a pseudorandom Perlin noise surface with a Laywoo-D3 filament on a Laywoo-D3 base printed on the same printer; (B) close-up of the second pseudorandom Perlin noise surface WPC print. Color images are available online.
Discussion
Curved layer printing provides an opportunity to generate a timber-like surface finish that does not require postprocessing, which can be further expanded through upscaling. For larger print jobs, a robotic arm provides a less restrictive volume to produce complex shapes in a flexible manner. However, the use of traditional 3D printing technologies may be considered the more stable option to provide consistent results.
Comparison of 3D printing technologies
3D printing using a robot arm can overcome the limitations of a gantry and restricted volume, allowing complex surfaces to be achieved. This means the height and angle of the curved surface can be greater without colliding with the 3D printing machine, as would be the case with conventional, FFF 3D printers.
For example, a 3D printer with a print head fixed at a vertical position relative to the gantry would not be able to print models with a difference in height greater than the vertical distance between the tip of the nozzle and the underside of the gantry. A similar 3D printer would struggle to print past an angle of about 45° as the extrusion nozzle and surrounding printer parts would interfere with the printed model.
While newer, delta-style, FFF 3D printers can move vertically at higher speeds than a conventional FFF 3D printer, the robot has a much larger printing area and has the additional advantage of being able to rotate the nozzle from the vertical position to match the normal of the surface it is printing on. As a result, 3D printing with robots provides the opportunity for the technology to be scaled up to larger applications, such as BAAM. This will result in greater efficiency in the production of paneling, for example.
However, six-axis robots may struggle with printing complex surfaces with a high degree of curvature changes over a short distance, especially if the nozzle needs to be perpendicular to the point that is being printed. This can be attributed to the requirement of all six axes of the robot having to move quickly to accommodate large changes in curvature, which alters the movement speed of the nozzle and causes uneven deposition.
Singularity points (where one coordinate in space has multiple solutions for the positions of the six axes that control the overall positioning of the tool head) during robot arm movement also cause large rotational movements within the segments of the arm while trying to print a line in a linear direction, which also slow down the overall movement of the robot and tend to bunch up the extruded filament.
As a result of the limitations of the robot, the Perlin noise surface is printed with a substantially higher quality using a vertically aligned print head on the large-format, FFF 3D printer.
Integration of the esthetic surface with other applications
The Grasshopper script used the region of influence method to manipulate toolpath lines to mimic the texture of wood. This concept can be expanded by using voids to optimize the amount of light that passes through a wall element when used as a building component for an office scenario. This timber pattern can also be projected onto any surface, as shown in Figure 10. This would allow the originally flat timber pattern to follow the curvature of any surface.
FIG. 10.

Diagrammatic representation of a generated timber-like pattern projected onto any surface, which can then be printed with curved layer printing. Color images are available online.
By integrating all these concepts together, it will become possible to fabricate a wall panel that combines both esthetics and functionality.
Conclusions
The robot in combination with an extruder and hot end enables the successful creation of a timber esthetic using curved layer FFF. The use of the six-axis robot arm enables fabrication of large-scale objects that can be created in a timely manner while still achieving a timber esthetic that is comparable or even superior to patterns achieved on 2D surfaces if the 3D form has gradual changes in gradient. The use of a robot for 3D printing can be achieved through integration of the FFF toolset with an existing robotic system. Conversely, a traditional FFF 3D printer may be more suited to small but frequent changes in height in the curved layer.
This study demonstrates novel methods of creating a timber-like surface from FFF 3D printing that can be customized to suit a variety of applications, such as architectural building façade elements. Not only is the result an esthetically pleasing surface that can form the outer skin of any element, but the originally flat timber pattern can also be projected onto any surface and printed as a curved layer, achieving a complex form that does not require additional cladding.
Acknowledgments
The authors would like to thank the assistance provided by the staff from the Design Modelling and Fabrication (DMaF) Lab in the Sydney School of Architecture, Design and Planning at the University of Sydney. The authors acknowledge the support for the research by a grant from Forest and Wood Products Australia (FWPA). FWPA is funded by industry levies and the Australian Commonwealth Government.
Author Disclosure Statement
No competing financial interests exist.
Funding Information
The authors acknowledge the support for the research by a grant from Forest and Wood Products Australia (FWPA) (H6264-20601). FWPA is funded by industry levies and the Australian Commonwealth Government.
References
- 1. Miguel F-V, Wilson C, Santiago F, et al. Effect of infill parameters on tensile mechanical behavior in desktop 3D printing. 3D Print Addit Manuf 2016;3:183–192. [Google Scholar]
- 2. Wittbrodt B, Pearce JM. The effects of PLA color on material properties of 3-D printed components. Addit Manuf 2015;8:110–116. [Google Scholar]
- 3. Song X, Pan Y, Chen Y. Development of a low-cost parallel kinematic machine for multidirectional additive manufacturing. J Manuf Sci Eng 2015;137:13. [Google Scholar]
- 4. Llewellyn-Jones T, Allen R, Trask R. Curved layer fused filament fabrication using automated toolpath generation. 3D Print Addit Manuf 2016;3:236–243. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Allen RJA, Trask RS. An experimental demonstration of effective curved layer fused filament fabrication utilising a parallel deposition robot. Addit Manuf 2015;8:78–87. [Google Scholar]
- 6. Singamneni S, Roychoudhury A, Diegel O, et al. Modeling and evaluation of curved layer fused deposition. J Mater Process Technol 2012;212:27–35. [Google Scholar]
- 7. Ghaffar SH, Corker J, Fan M. Additive manufacturing technology and its implementation in construction as an eco-innovative solution. Autom Constr 2018;93:1–11. [Google Scholar]
- 8. Khoshnevis B. Automated construction by contour crafting—related robotics and information technologies. Autom Constr 2004;13:5–19. [Google Scholar]
- 9. Holshouser C, Newell C, Palas S, et al. Out of bounds additive manufacturing. Adv Mater Processes 2013;171:15–17. [Google Scholar]
- 10. Lim S, Buswell RA, Valentine PJ, et al. Modelling curved-layered printing paths for fabricating large-scale construction components. Addit Manuf 2016;12:216–230. [Google Scholar]
- 11. Rice M, Kozak RA, Meitner MJ, et al. Appearance wood products and psychological well-being. Wood Fiber Sci 2006;38:644–659. [Google Scholar]
- 12. Mittal P, Chawla S, Gandhi AV. Building Material: Wood's Timeless Appeal. New Delhi, India: MGS Architecture, February 21, 2018. [Google Scholar]
- 13. Lutz JF. Wood Veneer: Log Selection, Cutting, and Drying. Washington, DC: US, U.S. Department of Agriculture, 1977; p. 137. [Google Scholar]
- 14. Hudson DV. Laminated Wood Products. USA. Patent No. 5968625. 1999. https://patentimages.storage.googleapis.com/45/88/3e/bb443908dd1c1e/US5968625.pdf Accessed April 2, 2020.
- 15. Kokrhanek V. Furniture Veneer and Method of Making Same. Italy. Patent No. 4743509. 1988. https://patentimages.storage.googleapis.com/4e/1f/d5/b7edd6ba9ad37a/US4743509.pdf Accessed April 2, 2020.
- 16. Bowns EE, Thornton DC. Plastic laminated structural board and method for making same US. Patent No. 5919575. 1999. https://patentimages.storage.googleapis.com/5b/1b/6f/00f11cc14e969f/US5919575.pdf Accessed April 2, 2020.
- 17. Fabian S, Joni M, Lewis C, et al. Digital wood: 3D internal color texture mapping. 3D Print Addit Manuf 2018;5:285–291. [Google Scholar]
- 18. Mai J, Girdis J, Proust G, et al. Shades of wood: the effects of temperature variation on the appearance and physical and mechanical properties of 3D printed wood-plastic composites. In: 5th Annual International Conference on Architecture and Civil Engineering (ACE 2017). Singapore: Global Science and Technology Forum, 2017; pp. 246–250. [Google Scholar]
- 19. Jovanovic I. 3D printing using wood composite materials. Mach Des 2015;7:23–30. [Google Scholar]
- 20. Mai J, Thistleton C, Löschke SK, et al. Towards a new techno-aesthetic paradigm: experiments with pattern, texture and colour in 3D-printed wood-plastic composites. In: Solid Freeform Fabrication Symposium. Austin, US: University of Texas at Austin, 2016; pp. 1868–1877. [Google Scholar]
- 21. Klesov AA. Wood-Plastic Composites. Hoboken, NJ: Wiley. [Google Scholar]
- 22. ModWood. ModWood Decking—The Sustainable Alternative. 2018. https://www.modwood.com.au/assets/Uploads/ModWood-Brochure-2018-Oct.pdf Accessed January 31, 2020.

