Five-Axis Printing of Continuous Fibers on the Mold

Abstract

This paper explores a five-axis printing method designed to improve the fabrication of continuous fiber-reinforced thermoplastic composites (CFRTPCs), essential for producing lightweight, complex structures in advanced manufacturing. Traditional CFRTPC placement techniques often face challenges with precision, scalability, and optimal fiber orientation, especially in customized, small-scale applications. The proposed five-axis printing technique overcomes these issues by enabling precise fiber orientation and the production of robust spatial structures using 3D-printed molds compatible with CFRTPCs. Validation through three-point bending and surface quality tests revealed that five-axis printed cylindrical-lattice samples, with fibers oriented at 45°, exhibited superior mechanical properties and surface quality. The five-axis printed samples achieved a load-to-weight ratio 27% higher than traditional samples and maintained their shape even under significant deformation. Surface quality improved significantly, with roughness values reduced from 37.63 µm to approximately 12 µm. This method advances CFRTPC applications in industries requiring complex, lightweight components.

1. Introduction

In the realm of advanced manufacturing, the creation of lightweight yet robust structures is paramount, especially within sectors such as aerospace [1] and orthotic devices [2,3]. Continuous fiber-reinforced thermoplastic composites (CFRTPCs) have become the material of choice due to their exceptional strength-to-weight ratio and durability [4,5]. However, fabricating structures that incorporate CFRTPCs presents significant challenges, particularly in achieving optimal fiber placement [6] and realizing complex geometrical shapes [7,8]. Traditional methods of CFRTPC placement, whether manual or automated, often fall short in precision, consistency, and scalability [9]. While conventional automated methods are efficient for large-scale production, they do not adequately cater to customized solutions and small-scale fabrication needs [10]. The potential of 3D printing to overcome these challenges is increasingly apparent, enabling precise material placement and the creation of complex, lightweight structures with high strength [11].

1.1. Hardware in Multi-Axis Printing

One of the challenges in manufacturing with CFRTPCs is related to the hardware required to achieve high design freedom. Multi-axis printing has demonstrated significant advancements in additive manufacturing, offering solutions to many limitations of traditional printing methods. In three-axis printing, the nozzle moves along three linear axes (X, Y, and Z), building layers horizontally, which often results in a staircase effect [12]. This method, while simplifying printer design and kinematics, necessitates support structures for complex shapes and overhangs, leading to increased material use, post-processing and lower surface quality [13,14,15].

To address these issues, the introduction of additional rotational axes in multi-axis printing allows the nozzle to approach the structure from different orientations, producing parts with smooth surfaces and intricate designs without the need for support material, thereby enhancing material efficiency and structural strength [16]. The first significant step in this direction was the development of the five-axis machine by Milewski et al. in 1998, which employed a high-powered laser as the energy source [17]. Since then, researchers have developed five-axis printers for various techniques, including direct metal deposition, multi-material printing, and wire mesh model printing [18,19,20]. Additionally, the development of five-axis CNC hybrid machines that combine additive and subtractive processes allows for both material addition and precise finishing in a single setup. However, these machines, which were not initially designed for AM, often increase in size and cost [21,22].

Other approaches to overcoming hardware limitations include the additive-lathe method, which converts a conventional print bed into a rotating metal shaft to produce cylindrical samples [23], and its extension to create other solids of revolution using inflatable elastomers [24]. Zhang et al. [25] explored the fabrication of curved shapes with CFRTPCs printed onto their surfaces, enhancing non-foldable printed base structures by implementing CFRTPCs [26]. Chu et al. [27] developed a system using a robotic arm and a rotary table to print continuous fibers onto previously printed structures, specifically for actuators. Kipping et al. [28] mounted the print bed on a robotic arm and utilized two fixed, rotated printheads on the machine frame to print non-planar continuous fibers and various filaments. Fang et al. [29] employed multi-axis printing with two robotic arms and support material to freely create complex structures with optimally placed continuous fibers.

While these methods offer varying degrees of design freedom, not all utilize continuous fibers, and some only enhance thermoplastic structures with continuous fibers rather than creating pure CFRTPC structures. Additionally, the use of robotic arms, while providing a high level of printing freedom, introduces greater complexity and expenses when compared to 3D printers. However, these drawbacks may be less significant for large industrial companies.

Challenges in printing with fiber materials include adhesion issues and the formation of voids caused by significant fiber deformation and printing speed [30]. Selecting an appropriate matrix is critical, as it must bond effectively to the fiber while also transferring compression forces. Abrupt directional changes during printing can result in fiber breakage, decreasing part strength [31]. Addressing these challenges requires strategies at both the design and slicer levels to reduce sudden directional shifts and enable smoother transitions.

1.2. Software in Multi-Axis Printing

The second major challenge in advancing CFRTPC manufacturing lies in the software required to control complex multi-axis printing processes. As multi-axis printing systems become more advanced, the generation of precise toolpaths and G-codes for controlling nozzle orientation and material deposition becomes increasingly complex. The complexity of five-axis machines, particularly in planning the toolpath and nozzle orientation, has hindered the development of fully capable slicing software.

Several attempts to address these software challenges include the use of robotic arms with six degrees of freedom for 3D printing parts without support structures [32,33]. To further enhance control over the printing process, researchers have explored advanced toolpath generation algorithms. For instance, Chen et al. introduced an algorithm that utilizes vector and scalar fields to adaptively generate toolpaths based on stress distributions, enhancing the mechanical properties of 3D-printed parts [34]. Liu et al. emphasized the need for design approaches that fully consider material, process, and structure to exploit the potential of continuous carbon fiber composites [35]. Moreover, Zhang et al. reviewed various fiber and matrix combinations, examining their effects on mechanical properties and highlighting the challenges and opportunities in material extrusion [13].

Despite these advancements, the lack of dedicated software tailored specifically for controlling five-axis printing processes remains a significant obstacle. Issues such as nozzle collisions, calibration needs to maintain accuracy and repeatability, and the overall complexity of multi-axis kinematics continue to challenge the development of robust and user-friendly software solutions.

1.3. Motivation

The motivation for employing a five-axis printing method to create CFRTPC structures stems from the necessity to overcome the previously mentioned limitations of traditional manufacturing processes. The unique capability of five-axis printing to manipulate and orient materials in five different directions allows for greater freedom in continuous carbon fiber placement. This freedom is crucial for harnessing the full potential of CFRTPCs’ mechanical properties, enabling the fabrication of structures that are not only customized, lighter, and stronger but also more complex than those produced through conventional methods. This innovation can hold significant value for industries requiring high quality and optimized weight-to-strength ratios in complex components.

In response to the dual challenges of hardware and software, this paper proposes a method aimed at enhancing the design freedom of CFRTPC spatial structures. Furthermore, a dedicated software tool has been developed to facilitate and support this innovative approach, ensuring precise control over the complex printing processes required for advanced manufacturing.

2. Materials and Methods

2.1. Manufacturing Concept

The integration of CFRTPCs into spatial structures presents a significant challenge in additive manufacturing. Traditional printing methods encounter limitations regarding fiber orientation and the feasibility of printing in mid-air, particularly within lattice structures. To address these challenges, a five-axis printing method is proposed to improve fiber placement and achieve superior mechanical properties in printed spatial objects.

The proposed method (Figure 1) initiates with the printing of a mold specifically designed to support the desired spatial shape. Utilizing a material compatible with CFRTPCs, the mold offers sufficient support during the printing process while allowing later clean detachment of the printed CFRTPCs structure.

After the completion of mold printing, the process of depositing CFRTPCs onto its surface begins. First, the distance of the nozzle to the surface of the mold is calibrated using 0.1 mm thick Feeler gauges. Thanks to the utilization of five axes, control over the orientation of the nozzle is possible, ensuring it remains perpendicular to the mold’s surface throughout the printing process. This facilitates consistent and uniform deposition of CFRTPCs onto the mold, enhances the adhesion of fibers both to the mold and to each other and improves fiber alignment. Upon completion of the printing, the printed CFRTPCs part is detached from the mold. The detachment process can be performed either by manually breaking or separating the mold or by dissolving it—all resulting in spatial object characterized by a high degree of freedom in CFRTPCs orientation.

2.2. Material Selection for Mold

In order to effectively implement the proposed method, it is important to ensure the stability of the printed CFRTPC structures; therefore, careful consideration must be given to the selection of appropriate materials for the molds. The chosen material must exhibit compatibility with the CFRTPC material to facilitate good adhesion between the two, thus enabling the successful printing of CFRTPC structures. However, it is necessary to keep a balance, as excessive adhesion might impede the detachment of the CFRTPC structure. Therefore, a series of tests were conducted to evaluate the adhesion of CFRTPC material to various mold’s materials.

The potential mold material samples were fabricated using an FFF printer (Mark One, MarkForged, Watertown, NY, USA). Each sample consists of two parts. The lower part utilized different mold materials and measured 100 mm × 20 mm × 1 mm. It was produced with eight layers, each 0.125 mm thick. A hardened nozzle with a diameter of 0.4 mm was used, with the extrusion temperature set according to the manufacturer’s recommended value. The infill pattern was rectilinear with 100% density, and the filament tracks were oriented parallel to the longer dimension of the part. In total, 22 different mold material samples were printed.

Subsequently, rectangular CFRTPC shapes with dimensions of 80 mm × 10 mm × 0.5 mm were printed directly onto the previously fabricated mold material samples. The material used for the CFRTPCs consisted of a nylon matrix reinforced with 1K carbon fiber roving (Carbon Fiber, MarkForged). The material exhibits a flexural strength of 540 MPa and flexural modulus of 60 GPa, compared to the matrix’s 50 MPa and 1.4 GPa. The printing was performed with a layer height of 0.125 mm, utilizing a specialized nozzle designed by the manufacturer specifically for printing continuous fibers.

The adhesion between the two materials in the samples was assessed through manual separation, employing three criteria: weak adhesion (preventing printing of both materials together), moderate adhesion (allowing separation), and strong adhesion (resulting in sample breakage without material separation).

Additional details regarding the materials and printing parameters can be found in Appendix A. Based on the obtained results, two materials were selected for fabrication–flexible mold material (Istroflex, Nanovia) and a water-soluble material (Water-soluble, XYZ-printing).

2.3. Printing Process

In this study, the impact of the proposed five-axis printing method on part properties was assessed through three-point bending tests and surface quality measurements.

All CFRTPC samples and molds for the three-point bending test were printed using a 3D-printer (3D-1750L-B, Hage3D, Obdach, Austria), configured in five-axis mode with a single printhead. The print unit was manually interchanged between a Fused Filament Fabrication (FFF) unit with a standard 1 mm braze nozzle and a CFRTPCs unit. Both units have centralizing pins to ensure a precise fit. The molds were printed using either a flexible material or a water-soluble material.

CFRTPC samples, which are conventionally printed in a 2.5D manner, were produced on Kapton tape adhered to a glass bed coated with an adhesive stick. In this standard approach, the design is limited to stacking CFRTPC layers on top of each other in the Z-building direction, restricting the overall geometry to vertical layering.

The proposed five-axis printing process of CFRTPC samples introduces additional steps. Molds were printed in a vase mode with a dual-sided brim to enhance adhesion and stability, directly in the center of rotation of a glass print bed coated with adhesive stick. The print was completed using the FFF unit. These molds were sliced using conventional slicing software (Simplify3D, Version 5.1.2). After the mold was printed, the FFF unit was changed to CFRTPCs unit to print the sample. In the next step, the five-axis CFRTPC sample was printed onto the mold. When printing onto flexible mold, a rigid core was inserted to increase stability.

The five-axis CFRTPC samples afford greater design freedom, enabling fibers to be placed in any direction within the build volume, but are limited to being placed on the surface of the mold or on top of each other, thereby creating a local Z-building direction perpendicular to the part surface. All CFRTPC samples and molds were designed using Computer-Aided Design (CAD) software (Inventor Professional 2024, Autodesk, Mill Valley, CA, USA) and their geometries are described in Section 2.4 The slicing process for five-axis printing is described in detail in Section 2.5.

Printing on the mold necessitates complex path-planning strategies, for which appropriate software was developed and utilized. The multi-dimensional path-planning process is elaborated upon in Section 2.6. During CFRTPC printing, the mold was rotated using a 2-degrees-of-freedom print bed, ensuring that the nozzle remained perpendicular to the mold’s surface. Upon completion of the CFRTPCs print, the mold with the integrated CFRTPC structure was removed from the print bed, and finally, the CFRTPC samples were detached from the flexible molds, or the molds were dissolved in water when employing water-soluble material. Further details on the process parameters for the printed molds and CFRTPCs parts are presented in

Table 1. Process parameters for molds and CFRTPCs printing.

Material	Layer Height [mm]	Print Speed [mm/s]	Nozzle Temp. [°C]	Bed Temp. [°C]
PVA	0.5	1200	200	60
Istroflex	0.5	600	245	70
CFRTPC	0.125	15	250	-

.

Additionally, the five-axis CFRTPC printing process was also performed using multiple of the same flexible mold. To accomplish this, the mold featured a flange at the bottom with a central hole for alignment. The aluminum plate had three threaded holes that corresponded to the holes in the flange, allowing the mold to be securely bolted in place. Subsequently, a centering pin was printed directly onto the aluminum plate. This pin was used to precisely position the mold, which was then secured with bolts. The CFRTPC printing procedure mirrored the one described earlier.

2.4. Mechanical Tests

The effect of the proposed five-axis method on the mechanical properties of the part was evaluated by a three-point bending test. Three types of samples were designed, printed with CFRTPCs, and compared with each other. The sample naming is based on the mold material, fiber orientation and sample number. The printed types were as follows: standard 2.5D printed (none-0) and two types of five-axis printed with the proposed method, printed on the water-soluble mold (solu-45) and printed on the flexible mold (flex-45).

All samples have the shape of a tube with an inner diameter of 39.1 mm, a wall thickness of 0.9 mm, equal to the fiber width, and a height of 100 mm. The none-0 samples were printed vertically, causing the CFRTPCs to follow a circular path and stack on top of each other, forming a single-line wall (Figure 2A). In all five-axis printed samples, fibers were oriented at a 45° angle, and each layer had the fibers oriented perpendicular to the previous one (Figure 2B). These samples comprised 8 layers with the build direction normal to the mold’s surface. Within a single layer, the fibers were equally spaced (spacing angle) and connected to the neighboring fiber either on the top or bottom of the tube (Figure 3). The spacing angle refers to the angle by which the mold was rotated before moving to the opposite end during the manufacturing process. The spacing angle was determined experimentally by comparing the results from the three-point bending test of the none-0 sample with the solu-45 sample. The goal was to ensure that both types exhibited a similar maximum force before failure.

All molds were printed as tubes with an outer diameter of 39.1 mm, a height of 140 mm, and a thickness of 1 mm (equivalent to a single line width). After printing the solu-45 samples and dissolving the material in water, they were weighed using a scale (EMB 600-2, Kern, Balingen-Frommern, Germany) and left to dry at room temperature. Periodic weight checks were conducted until stabilization, defined as negligible weight changes over time. All samples were dried under identical ambient conditions to ensure consistency.

The experimental investigations were conducted using the three-point bending test method with a span of 80 mm and a support radius of 5.0 mm. The loading edge, also with a radius of 5.0 mm, was positioned at the center of the test specimen. Testing was performed on a universal testing machine (Z020, Zwick, Ulm, Germany) at a set speed of 1 mm/min and an initial preforce of 5 N. During the procedure, both the applied force and the displacement of the crosshead were continuously recorded. The test was terminated based on one of two criteria: when the force dropped to 80% of the maximum recorded force or when the specimen reached a maximum extension of 35 mm. The test concluded as soon as either criterion was met. For three-point bending tests, each type of sample was printed five times and labeled with the suffix “-sX”, where X represents a sample number, ranging from 1 to 5.

Furthermore, the influence of the printing method on the surface quality of the printed part was assessed using a digital microscope (Digitalmikroskop VHX-7000, Keyence, Osaka, Japan). The objective is to evaluate how the deposition of fibers affects surface quality and to determine whether the printed parts achieve surface characteristics suitable for skin-contact applications. For this examination, tube samples akin to those employed in the three-point bending tests were printed, although with a height of 40 mm (Figure 2C,D). Additionally, the tube samples were printed with fibers placed in 0°, similar to the one printed in 2.5D, yet printed on the mold (solu-0 and flex-0) for direct comparison (Figure 2E). Finally, a 2.5D printed sample with a square cross-section and fibers in the 0° direction was additionally printed (none-SQ) (Figure 2F). Subsequently, all the samples were cut in quarters along their height in order to investigate their inner surface.

In the surface quality test, the inner and outer surface area of 6.709 mm by 6.406 mm for each sample with fibers in the 0° direction was examined. The roughness measurement was performed along three lines directed perpendicular to the fibers and spaced 1 mm apart. In the samples with fibers positioned at 45°, only the inner side along the fibers was examined. All values were calculated with the Keyence software. The average roughness value $R_a$ was calculated as the average of all deviations of the roughness profile from the average line along the reference distance, while the average roughness depth $R_z$ was determined as the difference between the maximum and minimum values along the reference distance. An average value was taken for all the parameters. The shape correction function for a sphere and cylinder of the software was used to compensate for the curved surface of the cylindrical specimens.

2.5. Five-Axis Slicing

This section demonstrates the conformal slicing approach of the five-axis HAGE3D 1750L CNC machine using the visual scripting language Grasshopper; it works within Rhino 3D CAD software (Rhinoceros Ver. 6.0; Robert McNeel & Associates, Seattle, WA 98103, USA). The machine features three prismatic joints along the X, Y, and Z axes, a printing bed that rotates around the Z-axis (C-rotation), and a tilting table that rotates around the Y-axis (B-rotation). It has a rotary axis offset in the Z-direction ($p_z$). The process of slicing the model and generating the G-Code is explained using the flowchart, as shown in Figure 4.

Initially, the model is imported into the Rhino software as a line-based model, and the user can choose its orientation in the desired direction. In Grasshopper, the model is sliced using the divide curve functional block based on the required layer height. Later, each layer is divided into several points, enabling the printer to detect and follow a continuous spiral path to prevent seam formation. Here, seam refers to the visual mark that appears when the printer head starts and stops the extruding material at the same point on each layer. It can affect the visual quality and structural strength of the printed models. In Grasshopper, the spiral print is created using the flip matrix and vector 2 pt functional blocks. These blocks help swap the rows columns and generate a vector between two points. The range functional block adjusts the domain values from zero to one based on the number of points. The spiralized concept allows the print head to move along the spiral path as it advances rather than shifting in one direction after each layer. The slicing process maintains nozzle orientation perpendicular to the mold surface to ensure uniform fiber adhesion. Additionally, continuous printing is achieved through spiralizing, which minimizes seam visibility and enhances the quality of the final part. Due to the multiple solutions available from the inverse kinematics joint angles, the algorithm is refined to ensure smooth transitions by selecting continuous joint movements, effectively avoiding abrupt jumps. This approach ensures precise fiber placement, ensuring structural integrity and continuous printing.

After spiralizing, the kinematic equations of the machine are computed to find the joint parameters of the five-axis printer to reach the target points. The homogenous transfer matrix is calculated to describe the relation between the printer base and extruder tip, which was detailed in the previous publication [36], as shown in Equation (1). In the subsequent equations, $s_i=sin\left({\theta }_i\right)$ and $c_i=cos\left({\theta }_i\right)$ hold for $i=4,5$.

$${}_5{}^0\mathbf{T}=\left[\begin{array}{cccc}{c}_4{c}_5& -s_4& c_4{s}_5& c_4{c}_5{d}_1-s_4{d}_2+c_4{s}_5{d}_3-c_4{s}_5{p}_z\\ s_4{c}_5& c_4& s_4{s}_5& s_4{c}_5{d}_1+c_4{d}_2+s_4{s}_5{d}_3-s_4{s}_5{p}_z\\ -s_5& 0& c_5& -s_5{d}_1+c_5{d}_3-c_5{k}_z+p_z\\ 0& 0& 0& 1\end{array}\right]$$

(1)

The first three prismatic joint values of the printer are calculated using the inverse kinematics of the machine, as shown in Equations (2)–(4). These are employed to find the desired position of the extruder tip. The normal vector $\mathbf{N}$ at the target points of the spiral path is used to find the orientation of the extruder tip, as shown in Equations (5) and (6).

$$\begin{array}{cc}\hfill d_1& =c_4{c}_{5}x+s_4{c}_{5}y-s_{5}z+s_5{p}_z\hfill \end{array}$$

(2)

$$\begin{array}{cc}\hfill d_2& =-s_{4}x+c_{4}y\hfill \end{array}$$

(3)

$$\begin{array}{cc}\hfill d_3& =c_4{s}_{5}x+s_4{s}_{5}y+c_{5}z-c_5{p}_z+p_z\hfill \end{array}$$

(4)

$$\begin{array}{cc}\hfill {\theta }_4& =\mathrm{atan}2(n_y,n_x)\hfill \end{array}$$

(5)

$$\begin{array}{cc}\hfill {\theta }_5& =\mathrm{asin}(-n_z)\hfill \end{array}$$

(6)

Due to the normal vector alignment in the Grasshopper, the difference in consecutive revolute joint angles sometimes gives 180°. This is not desirable for continuous printing; the angle difference should always be small to avoid the jerks. The Python script is created to adjust and rearrange the values from 0° to 360°. The flow rate is dynamically adjusted based on path curvature and layer thickness, maintaining consistent extrusion to prevent defects and ensure strong interlayer bonding. The flow rate affects fiber tension, influencing the risk of misalignment and delamination. The flow rate of the extruder is calculated using the length block in the software because layer height does not vary across the spiral path in the designed models for continuous carbon fiber printing. The slicer automatically generates the G-Code based on the inverse kinematics and flow rate.

2.6. Five-Axis Toolpath Planning

The five-axis toolpath planning starts by importing the model into Grasshopper; the models are designed using Autodesk Inventor CAD software and saved as STEP files. The slicer facilitates different parameter settings to generate the toolpath and G-code automatically. Users can select various options in toolpath parameters, such as layer height, number of layers to create the offsets, layer thickness, nozzle diameter, and model orientation. Similarly, G-code slicing parameters are also provided in the software, such as the print speed of the motors, flow/extrusion rate, filament diameter, extruder, and bed temperature along the chosen values, as shown in Figure 5. Once the user selects these parameters, the slicer generates the toolpath (on-screen Rhino CAD) and G-code (on-screen Grasshopper) based on the inverse kinematic equations, as detailed in Section 2.5. The G-Code is saved in the MPF format and connected to the HAGE3D 1750L machine using a USB stick for five-axis printing of the models.

3. Results and Discussion

3.1. Printing Process

The fabrication process of adhesion samples for material selection in mold design yielded 18 out of 22 materials, which successfully adhered and were deemed suitable for subsequent evaluation. However, it was observed that CFRTPCs could not be printed onto four materials, including polypropylene and three flexible materials, owing to inadequate bonding strength. Among the tested materials, eight exhibited moderate adhesion, while ten demonstrated strong adhesion. All short-fiber materials had strong adhesion since they were using same matrix material as CFRTPCs. Notably, printing on ABS material led to some fiber warping. The results are detailed in Appendix A.

Following the assessment, two materials were identified for further consideration: the water-soluble PVA, displaying the highest adhesion strength, and Nanovia material, which was one of two flexible materials capable of separation. Additionally, ABS, PLA and PETG were evaluated as a potential mold material, since CFRTPCs were barely adhering to them; however, detachment of the later-produced five-axis sample was not possible.

All samples exhibited no printing complications and were successful. The printing time for the 2.5D samples was 2 h, while the five-axis samples took 2.5 h. The soluble molds required 30 min to print, and the flexible molds took 2 h and 10 min.

Extraction of structures from flexible molds was facile, achieved by bending them away from the structure using fingers. Similarly, structures from soluble molds, after a 45 min bath, could be removed in a similar manner due to the material losing adhesion to the CFRTPCs and becoming flexible. Alternatively, after a 3-h soak, the molds dissolved completely.

Post-extraction signs of wear were evident on the flexible molds, particularly near the edges of the CFRTPC structure. It was identified as a result of the high temperature of the nozzle and the relatively small distance from the mold during CFRTPC printing. It led to slight melting of the material. This effect is less visible when accelerating the print. Moreover, some dimensional deformation of the mold due to removing of CFRTPCs was also observed; however, inserting a core restored them to their original shape.

At this point, it was decided to print an additional sample using the same flexible mold (re-flex-45) to determine if the mold could be reused. This sample also underwent a three-point bending test, and its surface quality was examined. During the second iteration using the flexible mold, certain regions exhibited partial fiber detachment from the mold. Despite these minor concerns, the sample was successfully printed.

During the preliminary observation, it was noted that the 2.5D structures exhibited a noticeably rough surface and visible seam lines. Some of the five-axis printed samples showed slightly detached rims where the fibers change direction. Nevertheless, samples printed on soluble molds exhibited high-quality surfaces without any seam lines. Similarly, samples from flexible molds displayed smooth surfaces, although with minor imperfections such as detached or missing lines on the inner surface.

3.2. Three-Point Bending

The three-point bending tests conducted on the three categories of samples (none-0, solu-45, and flex-45) revealed significant differences in their mechanical behavior (Figure 6). Before conducting these tests, several samples were printed on a water-soluble mold with varying spacing angles to identify one that could withstand maximum force similar to that of the none-0_s1 sample. The spacing angle was determined to be 4.74°.

The standard 2.5D printed none-0 samples sustained an average maximal force of 438.45 N with a standard deviation (SD) of 25.69 N. These samples tended to break instantly into two pieces after reaching this force, typically between a pair of layers at the weakest fiber connection, rather than at the point of applied force. An exception was sample none-0_s5, which did not break immediately; instead, it deformed at the point of applied force and broke only in this region. On average, none-0 samples deformed by 5.93 mm (SD: 1.83 mm) before breaking. Notably, sample none-0_s5 deformed up to 9.00 mm without breaking into two pieces, though it reached maximal force at 6.14 mm, similar to the other none-0 samples. The weight of none-0 samples was highly consistent, averaging 14.74 g with an SD of 0.03 g, indicating uniform manufacturing quality.

In contrast, the five-axis printed samples on a water-soluble mold demonstrated superior mechanical performance in both strength and flexibility. These samples sustained an average maximal force of 466.12 N (SD: 19.10 N), slightly higher than the none-0 samples. All the five-axis printed samples (including those printed on the flexible mold) reached the test condition limit of 35 mm deformation, while not breaking into pieces. Instead, after being compressed, they partially returned to their original shape once the force was removed, with only the area of applied force remaining deformed. Upon closer examination, small delaminations were also visible in this area. Both solu-45_s1 and solu-45_s3 broke in the area of applied force; however, they remained in one piece, returning to a shape similar to the original. Solu-45_s2 exhibited only small delaminations and fully returned to its original shape. The solu-45 samples showed significantly higher deformation under maximal force, averaging 19.22 mm (SD: 2.44 mm), which is 3.7 times higher than that of none-0 samples. The average weight of solu-45 samples was 12.40 g (SD: 0.14 g), with one exception (solu-45_s5) weighing 12.64 g. The weights were approximately 15% lower compared to none-0 samples.

The samples printed on a flexible mold were generally less strong than the none-0 samples, sustaining an average maximal force of 367.61 N (SD: 34.99 N). Deformation under maximal force for flex-45 samples averaged 17.19 mm (SD: 2.00 mm). Although slightly lower than the solu-45 category, these values still indicate significant flexibility. The weights of the flex-45 samples were fairly consistent and similar to solu-45 samples, averaging 12.59 g (SD: 0.07 g).

When considering the load-to-weight ratio, solu-45 samples were the most robust, with 37.64 N/g (SD: 1.81 N/g), which is 27% higher than the 29.75 N/g (SD: 1.68 N/g) of none-0 samples. Despite sustaining the lowest force, flex-45 samples had a similar load-to-weight ratio to none-0 samples, at 29.20 N/g (SD: 2.75 N/g).

An additional sample reprinted on the flex-45_s4 mold indicated lower mechanical properties compared to its predecessor, with a similar weight of 12.62 g. This sample sustained a force of 306.37 N, around 17% less than the previous samples, and deformed by 12.54 mm under this force, which is 27% lower. The load-to-weight ratio was 24.28 N/g, also 17% lower. Nonetheless, this sample, as well as all other five-axis printed samples, showed almost six times higher deformation, reaching the stopping condition without breaking and returning to the shape close to original with visible deformation in the area of applied force.

3.3. Surface Quality

The surface quality analysis of 3D-printed samples reveals notable differences between the inner and outer surfaces, as well as between conventionally printed samples and those printed using the method proposed in this paper (

Table 2. Mean surface quality measurements for tested samples. For samples with a fiber orientation of 0°, measurements were taken on both the inner and outer surfaces. For samples with a 45° fiber orientation, measurements were taken along the fiber direction.

	None-0		None-0_sq		Solu-0		Flex-0		Solu-45	Flex-45	Re-Flex-45
	Inner	Outer	Inner	Outer	Inner	Outer	Inner	Outer	Inner	Inner	Inner
$R_a$ [µm]	37.63	14.50	26.90	25.30	12.53	2.20	11.87	1.90	3.50	6.80	3.83
$R_z$ [µm]	231.67	95.50	164.90	153.53	58.57	12.77	70.93	12.37	26.23	33.87	23.57

). All cylindrical samples exhibit rougher inner surfaces, while the outer surfaces are smoother. The none-0_sq sample, printed in a square shape, shows consistent surface roughness across both sides. Additionally, the surface parameters for the none-0_sq sample have values roughly in the middle compared to the none-0 sample since the fibers were laid in a straight line. Both of these samples exhibit high $R_z$ values, indicating more surface irregularities compared to the five-axis printing method.

The solu-0 and flex-0 samples demonstrate significantly superior surface quality compared to the none-0 sample, with all surface parameters improved by three to eight times. The inner side of the solu-0 sample shows a clear influence of the mold on the surface, characterized by systematic peaks with a distance of 0.5 mm, corresponding to the layer height of the mold. On the outer layer, the peaks occur every 0.9 mm, equal to the CFRTPCs deposition width. In contrast, the inner side of the flex-0 sample does not display an obvious pattern, but grooves occur at intervals of 0.9 mm. This is likely due to the mechanical removal of the sample from the mold, where weak peaks broke, flattening the surface. This is evident when comparing the Rsk parameters of both samples: 0.2 for solu-0 and −0.6 for flex-0, indicating fewer peaks in the latter, yet they still have similar $R_a$ and $R_z$ parameters. On the outer surface, the pattern is even less visible and the sample surface is much more uniform.

The samples printed with fiber orientation at 45° present a more complex surface analysis due to the occurrence of gaps in the print, caused by the lattice design. When comparing the roughness along the fiber, thus omitting the gaps, these samples exhibit a smoother surface compared to samples with a fiber orientation of 0°, with $R_a$ values between 3 µm and 7 µm compared to around 12 µm. However, measuring along the fibers also omits the layer change, resulting in better measurement outcomes. In this scenario, solu-45 shows the best surface quality, surprisingly followed by re-flex-45, not flex-45. Closer analysis revealed that flex-45 has a less compact structure with larger gaps while the solu-45 sample has the smallest gaps. In all samples, the lattice is not equally spaced, with two fibers close together and a larger gap between them. Significant results are illustrated in Figure 7.

3.4. Discussion

The adoption of the proposed method for parts manufacturing offers several advantages over conventional 3D printing of CFRTPCs. Samples printed using a water-soluble mold demonstrated a 27% improvement in mechanical properties when comparing the load-to-weight ratio. It is important to note, however, that this improvement was largely due to the unique design of the sample, which is not achievable with other methods, rather than the material processing itself.

A notable difference between the two approaches lies in the failure behavior. Samples produced with conventional 3D printing tended to break suddenly upon reaching their maximum load, with no visible signs indicating that the part was about to fail and leaving the part unable to bear any further loads. This breakage occurs between layers, where the bonding is weakest, allowing cracks to propagate easily and causing the part to separate into two pieces.

In contrast, none of the five-axis samples exhibited this sudden breakage. Instead, they demonstrated a more gradual failure mode, where even after exceeding their load capacity, the parts remained connected. This behavior enhances safety by making failure more visible, eliminating the risk of sudden collapse and allowing the part to continue supporting some load.

All samples produced using the five-axis printing method showed surface quality superior to conventional ones by a significant margin. This improvement is particularly beneficial for applications involving skin contact, such as orthotics and prosthetics. Unlike traditionally printed samples, the five-axis printed ones achieved a surface roughness $R_a$ below 17 µm, which is within the range of commercially used products in this field [37]. Moreover, further optimization of the printing parameters could potentially achieve even lower roughness values. Nevertheless, these optimized values are expected to remain within the same order of magnitude, as demonstrated in the work of Abas et al. [38], which highlights the significant influence of parameter settings on surface quality.

The inner surfaces of all cylindrical samples are rougher, while their outer surfaces are noticeably smoother. The square-shape sample has consistent surface roughness on both sides, indicating the significant impact of geometry on surface finish. This variation in surface quality is likely influenced by the stresses acting on the fibers during printing. The inner side fibers are compressed, resulting in a rougher surface, while the outer side fibers are extended, resulting in a smoother surface.

Closer analysis of the surface quality samples revealed that flex-45 has a less compact structure with larger gaps, likely due to being printed on a farther part of the mold, which experienced greater deflection and resulted in a higher nozzle distance and thus poorer print quality. This phenomenon is also visible when comparing both ends of three-point bending samples. The solu-45 sample has the smallest gaps, probably due to mold material calibration leading to slight over-extrusion and greater fiber compression. In all samples, the lattice is not equally spaced, with two fibers close together and a larger gap between them. This is likely due to the nozzle hole being significantly larger than the fiber, causing the fiber to shift towards the direction of mold rotation during neighboring fiber placement, resulting in uneven fiber deposition.

Another advantage of this method is the ability to create shell models with fibers oriented in multiple directions. This is particularly beneficial in applications such as lower limb prosthetic sockets, where the part must withstand multidirectional loads while remaining lightweight. While carbon fibers are already used in such applications, their placement is typically done manually. Additionally, the flexibility in design allows for enhanced customization to meet the unique biomechanical demands of each individual user, improving comfort.

When printing on cylindrical samples, the distance between the nozzle and the mold is not constant due to the curvature of the mold. Given the nozzle size, this distance varied between 0.125 mm and 0.151 mm. According to Oberlercher et al. [39], a distance exceeding 0.135 mm during printing significantly affects fiber quality. Therefore, smaller cylindrical objects printed with this method are likely to have inferior quality compared to larger ones. One way to address this issue could be to reduce the nozzle distance or use smaller nozzles, which might also lead to more regular pattern placement. However, in that case, the increased pressure on certain sections of the fiber could lead to lower dimensional accuracy, potentially compromising the precision of the final printed object [40]. The impact of these factors warrants further investigation, as they also influence the mechanical properties of the final part.

Another noteworthy finding from this work is the impact of fiber path placement curvature on surface quality. Inner surfaces tend to have lower surface quality compared to outer surfaces due to fiber compression. Moreover, edges on both ends exhibited minor defects due to fiber detachment, which was potentially caused by the high bending angle required to change the path direction [41]. This issue could potentially be mitigated by smoother transitions and lower speeds at the ends.

To further demonstrate the capabilities of the proposed method and the developed software, three additional demonstrators were printed (Figure 8). The first demonstrator is a scaffold-like structure with sparse lattices, similar to the previously presented tubes, effectively highlighting the lightweight nature and structural efficiency of the printed designs. The second demonstrator, a conical frustum connected to a cylinder, illustrates the advantage of incorporating an additional axis during printing, enabling more complex geometries.

The third and most sophisticated demonstrator is a wrist orthosis, showcasing the versatility of the method. This design demonstrates that the system does not require continuous rotation during printing, as proven by having a part with flat fragments and an open structure on one side. Additionally, fibers are placed in the longitudinal direction. Despite these unique capabilities, the process still remains fundamentally a 3D printing technique, allowing the part to being built layer by layer with infill where needed, maintaining the flexibility and adaptability inherent in traditional 3D printing processes.

Although the proposed method provides significant freedom in design and optimization, certain limitations remain. The designs are inherently constrained to being built directly on the mold’s surface or sequentially layered on top of each other. Additionally, even with a pre-printed mold, not all areas are accessible to the nozzle due to potential collisions, limiting the reach of the printer in certain intricate geometries. Furthermore, surfaces oriented toward the print bed are often inaccessible, as the printhead risks colliding with the print bed.

When utilizing water-soluble molds, there are no additional restrictions on their geometries. However, when using flexible molds, certain limitations must be taken into account. The mold must be able to deform sufficiently to be removed from one side of the part. While making the molds thinner can improve their flexibility, it also reduces stability, potentially leading to tilting and printing inaccuracies. For some geometries, temporarily inserting a supporting core into the mold during printing can improve its stability. Additionally, with intricate or enclosed geometries, accessing certain areas and detaching the CFRTPC structure from flexible molds can pose significant challenges.

While the project achieved significant milestones, it is important to acknowledge certain challenges and areas where success was not fully realized. Although the structures were successfully detached from the flexible molds, issues with the inner surface quality and mold deformation were observed.

Furthermore, while structures printed on water-soluble materials demonstrated high quality, the single-use nature of these materials due to their dissolvability raises concerns about sustainability and practicality for certain applications. This characteristic poses challenges in terms of reusability and may impact the economic feasibility of the technology in some cases.

The current design process involves designing line-based models, which is time-consuming and inefficient for complex or less repetitive parts. Efforts are undertaken to enhance the software by automating the creation of 3D lattice structures, similar to infill features but extended to three-dimensional domains, aiming to reduce manual effort in handling geometries. The software can be further enhanced by integrating an optimization loop combined with finite element analysis, enabling adaptive fiber placement tailored to specific conditions.

Further investigation into reprinting samples on the same mold and optimizing their mechanical properties could provide better insights into the sustainability of this method. In this approach, accurately positioning the mold in the center proved challenging. While the use of centering pins provided sufficient precision for successfully printing the CFRTPC structure, the mold tended to tilt slightly when bolted to the print bed, resulting in uneven spacing during rotation and leading to variations in the nozzle-to-mold distance, which could have contributed to the fiber adhesion issues. Using longer centering pins may help address this issue by improving alignment and stability. Additionally, creating more complex structures and utilizing dual printing could expand the potential applications of the proposed method.

4. Conclusions

This work advances the capabilities of 3D printing by integrating five-axis printing with continuous carbon fibers on the mold, addressing some limitations of conventional methods. This approach enables the creation of intricate, high-stiffness structures with enhanced design freedom and automated production. Detaching continuous fibers from the mold results in lighter yet robust designs and superior surface finishes of the produced parts. The application of technology in fields such as orthopedics, automotive, and aerospace engineering, where lightweight and durable 3D geometries are crucial, opens new possibilities for efficient and customized production. The proposed method meets the growing demand for manufacturing processes that deliver superior surface quality, enhanced functionality and greater design freedom compared to traditional 3D printing, thus opening new avenues for innovation and development in the field of additive manufacturing.