The GP formulation was based on metakaolin with additional reinforcement and functional fillers. The mix consisted of:

1. Metakaolin (MT): Primary aluminosilicate precursor, containing ∼52.9% SiO₂ and 41.9% Al₂O₃ with minor oxides (Fe₂O₃ 1.1%, TiO₂ 1.8%, MgO 0.18%). Particle size distribution: D₅₀ ≈ 3 μm, D₉₀ ≈ 10 µm.

The combined action of NaOH and anhydrite promotes the formation of a dense N–A–S–H gel, with possible C–A–S–H phases, refining pore structure, lowering permeability, and improving underwater printability and durability.

The mixture has been prepared by considering the following steps: (i) all mixing was performed at room temperature to avoid premature setting due to heat from the reaction, (ii) using mechanical mixing equipment for better consistency, especially when adding carbon fibres and fillers, and (iii) the GP paste may start setting soon after preparation, worked quickly after the final mixing step. Table 1 outlines the composition of the GP mix, highlighting the specific weight fractions of each component for preparing the material.

The 3D-Paste Extrusion Development Toolhead (3D-PEDT) piston extruder was used to test small material samples. Due to their simple design and functional similarity, a commercial printer for material extrusion with thermal reaction bonding (MEX-TRB/P) was successfully converted into a MEX-CRB printer. As demonstrated by Denker et al. (2024), a standard 3D desktop printer can be adapted with a paste extrusion system using a commercially available plastic material extruder (MEX-TRB/P), a method applied in this study, as illustrated in Figure 1. Unlike conventional 3D printing methods, this process does not involve any heat source or curing procedure and is carried out at room temperature. Instead, the GP material hardens through a solution-based reaction.

A bar with dimensions 20 × 98 × 20 mm³ was printed and subsequently, after 14 days of curing, cut into four equal parts, each forming a 20 × 20 × 20 mm³ cubic sample for compression testing. The printing process parameters are summarized in Table 2.

For manually printed underwater, the moulding frame was submerged in tap water instead of being placed in the air. Using a specialized syringe tube, the material was carefully extruded into the mould to ensure uniform extruding underwater, see Figure 2a. The bars were cut into cubic samples while still in a wet condition, and they remained submerged in water throughout the entire testing period. In this test case, only the exposure of GP materials to tap water will be examined, as they were manufactured directly underwater.

For mould casting in air, bar samples were first fabricated using a moulding process in air and then cut into cubic samples. A moulding frame was prepared to create bars with dimensions of 20 × 160 × 20 mm³. The material was poured into the moulds, and after 14 days of curing, the samples were cut into cubic forms and prepared for long-term experiments in various environments. Figure 2b shows the GP materials after moulding into bar-shaped forms.

The compressive strength test was carried out in accordance with ISO 1920–4:2020–01, which specifies three samples in minimum for determining the compressive strength. The cubic samples were tested on a universal testing machine model MTS 880 with a 100 kN measuring cell. A load speed of 6 mm/min was used and four specimens were tested per environment condition and time interval.

To investigate the behaviour of GP in mould-cast and 3D-printed samples under different environmental conditions, samples were placed in four different environments air, deionized-, lake- and salt water with the salinity of the Baltic Sea for varying time intervals (baseline, 1 month, and 3 months). The compressive force was applied perpendicular to the layer buildup direction in 3D-printed samples and to the free surface of the mould-cast specimens. A summary of experimental conditions, testing, and specimen details is illustrated in Table 3.

The baseline, day zero, condition applies only to air-exposed samples manufactured by mould-casting and MEX-CRB to determine their initial compressive strength at the start of the long-term experiment. Subsequently, compressive tests were conducted on both sample types after one and 3 months in different environments. Figure 3 illustrates the compressive strength of these samples. The results reveal that mould-cast samples in air exhibit the highest compressive strength, reaching approximately 50 MPa, whereas 3D-printed samples in air demonstrate significantly lower strength, around 20 MPa, see Figure 3.

When exposed to water environments, such as deionized, lake or salt water, the compressive strength of all samples generally decreases. However, mould-cast samples consistently retain higher strength and force values compared to 3D-printed samples across all conditions. The most noticeable strength degradation occurs in water-immersed conditions, where both sample types show reduced values over time. A consistent trend emerges, demonstrating that mould-cast samples outperform 3D-printed samples in all environments. The largest strength difference is observed in air, but this gap narrows slightly in water-exposed conditions. For instance, after 3 months in salt water, mould-cast samples still exhibit strengths around 30 MPa, whereas 3D-printed samples remain slightly below 20 MPa.

These findings indicate that mould-casting offers higher mechanical performance compared to 3D printing, particularly in dry conditions. However, exposure to water further weakens both sample types, though mould-cast samples consistently retain relatively higher strength and require greater force for failure. The box chart overlays and error bars highlight variability within the dataset, suggesting possible inconsistencies arising from both material-related factors and experimental conditions, such as voids from mould casting or weak interlayer bonding in 3D-printed specimens, that may influence the final compressive strength and force measurements.

Dimensional and physical changes were assessed by comparing baseline and pre-test values after water exposure. Figure 4 presents the mean percentage differences in mass (top left), density (top right), width (bottom left), and height (bottom right) for mould-cast and 3D-printed samples after exposure to deionized, lake, and salt water for one and 3 months.

Mass changes (Figure 4a) reveal that all specimens absorbed water, with moulded samples showing moderate increases (∼5–6%) and 3D-printed ones higher gains (∼8–11%). In salt water, the samples show the highest mass absorption because of both water and salt. Density variations (Figure 4b) were between 10% and 15% for all groups. Both mould-cast and 3D-printed samples showed a similar trend, with only minor changes during exposure, indicating that water environment had limited influence on density. Dimensional changes were minor compared to mass and density. Width (Figure 4c) remained nearly constant across all environments; in 3D-printed and moulded specimens, almost all values were slightly negative and close to zero. Height changes (Figure 4d), however, showed a clear contrast: mould-cast specimens contracted markedly, approaching several percent shrinkage (∼3–8%), whereas 3D-printed samples fluctuated between minor shrinkage and swelling within about ±1%–2%, remaining overall closer to dimensional stability. In summary, mould-cast samples demonstrated lower water uptake but greater shrinkage, while 3D-printed samples absorbed more water yet retained better geometric stability.

Moreover, the compressive strength of the hand-printed samples underwater has been examined over different time intervals (zero, two, four, six, twelve and 24 weeks). Figure 5 presents the result of the compressive tests over different time intervals. The samples were removed from the water 1 hour before the test, measured, and stored in an airtight bag.

Observing the trends, the strength increases from the Baseline, day zero from ∼8 MPa to the second week to ∼13 MPa, suggesting material hardening or improved curing over time also underwater. The highest compressive strength occurs on the fourth week with ∼14–15 MPa, but with high variability, as indicated by the larger boxplot and extended error bars. By the sixth week and 12th week, the strength stabilizes by ∼13 MPa, with reduced variation, as seen in the smaller box and shorter error bars. By week 24, a slight decrease in strength is observed, with values dropping to ∼11.5 MPa, indicating a potential plateau or mild degradation in mechanical performance over extended immersion. These results confirm that these samples experience ongoing strength development underwater, with curing effects enhancing performance up to a point. The fourth week marks the peak strength phase, while subsequent weeks show stabilization or modest decline, suggesting a limit to underwater curing benefits over time.

The compressive strength of our 3D-printed samples (approximately 20 MPa in air, lower in water) falls within the lower range reported for printable GP systems (Elsayed et al., 2022), and remains well below values achieved in optimized mixes incorporating copper slag or improved curing protocols, which exceed 40 MPa (Kozub et al., 2024). This supports the common observation that enhancing printability, through reduced solid content or adjusted rheology, can compromise mechanical performance, particularly under challenging conditions such as water immersion (Wang et al., 2021).

Figure 6 shows two microscope images at ×20 magnification of the cast, additively manufactured, and manually syringe-printed cubic samples in water before and after the compression test. The 3 mm long recycled carbon fibres in the printed samples are unidirectionally aligned in the printing direction, whereas the carbon fibres in the cast samples are non-directional. Regardless of the manufacturing process, almost all samples have pores on the analysed cut surfaces.

The size of the pores also varied from microscopic to macroscopic with up to 4 mm. The pores and defects of the moulded samples are mostly roundish and occasionally oval with vertical orientation, Figure 6a, whereas the pores of the samples processed with MEX-CRB and the syringe samples are mostly elongated in the direction of printing, round pores are hardly pronounced, Figure 6b. Elongated, oval pores have a larger surface area than round pores, which also provides a larger reaction surface for the GP with water. This may be the reason for the reduced compressive strength of the printed GP.

Regardless of the manufacturing process, all samples exhibited brittle fracture behaviour. Only the samples printed manually with a syringe in water initially deformed into a trapezoidal shape after 2 weeks of storage in water before the brittle fracture occurred. In addition, the residual moisture was pressed out of the specimen. The curing process of the GP was not yet complete at this point. The compressive strength tests show that the final strength of the GP is reached after 28 days, as with conventional concretes. At the end of the 28 days, brittle fracture occurred in all samples without further deformation.

The brittle fracture of the cast samples always occurred parallel to the direction of force application. Occasionally, small, thin material flaking occurred on the sides of the samples. Figure 7 shows the brittle fracture on the left, in the upper area pores and the non-directional carbon fibres can be seen, which are visible through the material spalling.

In the samples produced with MEX-CRB, in addition to the brittle fractures running parallel to the force application, Figure 8a, there are also cracks between the layers, Figure 8b. The areas between the deposited pressure paths behave like notches; in contrast to casting, the material properties of the GP are not homogeneous in this area, which favours fracture. Due to the additive manufacturing process, there are more pore accumulations between the deposited webs in this area, which further favours the fracture effect.

The effect of cracks in the print paths is less pronounced with samples printed manually with a syringe in water than with MEX-CRB, as the material immediately comes into contact with water when it emerges from the syringe. The toughness of the GP decreases and cavities and unevenness in the deposited print paths are levelled out, reducing the accumulation of pores.

CF improve fracture behaviour through crack bridging and pull-out, delaying crack propagation and absorbing energy. In mould-cast samples, randomly oriented fibres provide multidirectional reinforcement and higher strength, whereas in printed samples, fibre alignment and interlayer weaknesses limit stress transfer and reduce load capacity (Figure 8a).

Figure 9 shows the representative force–displacement responses of mould-cast and printed samples stored in different environments, corresponding to the compressive strength results presented in Figure 3. For clarity, only one specimen per storage condition and manufacturing route is displayed. The curves indicate higher peak forces and more stable post-peak behaviour for mould-cast samples, while printed samples show lower load capacity and steeper softening. The jagged shape of many curves, characterized by sudden load drops followed by recovery, reflects fibre debonding, pull-out, matrix heterogeneity, and interlayer weaknesses for printed samples.

It is worth noticing that the behaviour of CF in the GP matrix is coherent with previous research; they enhance the mechanical properties of the material (Růžek et al., 2023), (Hassan et al., 2025). However, the comparative works on the usage of the AM technology usually show better properties obtained in the case of the casted materials (Korniejenko et al., 2022). It should be noted that in the case of 3D printed elements, this kind of reinforcement can play an even more important function. In the case of using AM, there are limited possibilities to implement other forms of reinforcement into the composition. Traditional bars are completely not coherent with this technology and the possibility of implementation of continuous reinforcement is also limited (Lv et al., 2022). In this case, the short fibres are the only reasonable option. This kind of reinforcement has to be carefully selected according 3D printing parameter process to obtain proper cohesion between fibre and matrix material. With the above limitations, it should also be taken into account that the 3D printing process itself will affect the mechanical properties of the material; in particular, it will be characterised by significant changes in properties depending on the direction of research (Lv et al., 2022), (Maralapalle et al., 2025). An additional limitation is that the usage of recycled fibres is a kind of compromise between improving the material properties and reducing the environmental impact of the material. Current recycling technologies significantly influence the final properties of the CF, lowering their properties compared to virgin ones (Růžek et al., 2023), (Yu et al., 2025).

In order to investigate the long-term interaction of GP composites with recycled carbon fibres under realistic and chemically diverse aquatic conditions, samples produced via both casting and AM were immersed in three different water types: deionized water, synthetic seawater (simulating Baltic Sea salinity), and untreated lake water. The water quality development over a period of 3 months was assessed in terms of ionic conductivity and visible changes in water and sample surfaces. The presence of metals in the water was analysed using ion chromatography with the 850 Professional IC measuring device from Metrohm.

The selection of these three water types serves to simulate a range of realistic environmental and laboratory conditions. Deionized water, while not distilled, was obtained from a workshop filtration system and is intended to provide a controlled, low-ion background to observe the baseline leaching behaviour of ions from the GP materials. Synthetic seawater was created by dissolving 200 g of marine-grade salt in 10 L of deionized water and was used to replicate the alkaline and saline conditions similar to those of the Baltic Sea. This setup aims to determine whether the presence of salts and elevated ionic strength changes the leaching behaviour or affects the structural stability of the composites. Untreated lake water, containing natural microorganisms and organic matter, introduces a biologically active environment, enabling investigation into potential biofilm formation, biological degradation or surface colonization effects. All samples - both printed and cast - were placed in water containers after production. By combining both manufacturing techniques with the three distinct water types, a total of six experimental containers were prepared. Each container held a volume of 10 L and was stored at room temperature in a location protected from direct sunlight to avoid temperature fluctuations or biological bias. This experimental design enables a comprehensive investigation of the durability of the GP composites under various long-term exposure conditions.

The analysis of the element concentrations in deionized water, Figure 10, over time reveals a continuous increase for both printed and moulded samples, indicating that substances are consistently leaching from the GP composites into the surrounding water. This suggests that the materials are not chemically inert under these conditions and actively release elements into the solution.

At the beginning of the experiment, all measured elements - except for calcium and magnesium - were very low and partly below the detection limit. The measured elements, calcium, magnesium, lithium, and aluminium show the most significant increases in concentration. These elements likely originate from the GP matrix or residual phases associated with the recycled carbon fibres. Their increased presence points to either partial dissolution of specific mineral phases or ongoing ion exchange processes at the material-water interface (e.g., metakaolin and anhydrites).

Both types of samples - printed and moulded - show similar leaching behaviour over the observed period, indicating comparable material-water interactions. However, it is noteworthy that the aluminium concentration in the cast specimen shows a distinctly stronger increase after 3 months compared to the printed variant. The concentrations for calcium, magnesium, and lithium increase less in the container with the moulded samples.

Importantly, the fact that concentrations have not plateaued by the end of the three-month observation period implies that chemical equilibrium has not yet been reached in either system. This indicates the potential for continued leaching over longer timeframes. Especially in large bodies of water, there will always be dilution in contrast to the buckets used.

In the lake water, Figure 11a, calcium and magnesium concentrations were initially high, reflecting the natural composition of the freshwater environment. Unlike in deionized water, these concentrations decreased over time, indicating potential precipitation of these ions or uptake by biological activity such as algae or biofilms forming on the sample surfaces.

The behaviour of the other measured elements was comparable to that observed in the deionized water experiments. This suggests that, aside from the calcium and magnesium concentrations, the leaching behaviour of the GP materials is consistent across both water types.

As seen previously, the aluminium concentration in the moulded samples was again higher after 3 months than in the printed ones. This aligns with the observations in deionized water and supports the assumption that the casting process might favour the release of aluminium due to differences in the material structure.

In the saline water environment, Figure 12, only a limited number of elements could be reliably quantified. Due to the salt content, the samples had to be diluted before being analysed as the concentration exceeded the measuring range of 0.5–200 mg/L. This resulted in some metal concentrations falling below the detection limits of the measuring devices. As a result, fewer elements were measurable compared to the deionized and lake water conditions.

Notably, the concentrations of several detected elements decreased over time. This behaviour is consistent with the visual observation of white precipitates forming on the surfaces of the samples, indicating the removal of certain ions from the water column via precipitation. Furthermore, all other elements remained at significantly lower concentrations.