Toward Sustainable Clay-based Additive Manufacturing: Fresh-State and Hardened Mechanical Characterization of 3D-Printable Benjellik Clay Paste

2.2.1. 3D Printable Paste Preparation

The clayey paste for 3D printing is synthesized through a systematic protocol to ensure homogeneity, rheological stability, and printability. Raw marl blocks were extracted, crushed, and dried (105°C, 24 h) to eliminate moisture and facilitate grinding. The material is pulverized using a Retsch Mortar Grinder RM 200 and sieved through a 212 µm mesh to ensure fine particles. The powder is sampled and analyzed via MICROTRAC Camsizer X2 laser granulometry for precise size distribution as detailed in 2.3.1. A paste is formulated with 72.73% marl powder and 27.27% water (water/binder ratio W/B: 37.5%), optimized for extrudability and structural cohesion. After 5 min of mixing, the paste rested for 24 h under sealed conditions to equilibrate hydration.

This protocol, depicted in the first block diagram of the global workflow in 2.4, ensures controlled rheological properties critical for successful extrusion-based 3D printing, as discussed and stated in [15, 33, 44].

2.2.2. 3D Printing and Molding

This investigation adopted a dual approach to fabricate 40 mm cubes as recommended for cementitious materials. First, 3D clay printing for experimental samples and conventional molding in side-opening molds for reference specimens were adopted. For 3D printing, two infill strategies were implemented using optimized parameters in Table 2. A single solid contour wall is applied to all printed samples to ensure dimensional accuracy and surface integrity. The layer height-to-nozzle ratio is fixed at a value of 0.5 according to literature recommendations [44, 45] and preliminary fast-tuning trials.

The studied infill patterns are:

• Grid-like Infill (GL): Alternating layers at 0° and 90°, forming a grid structure (Fig. 1a).

  

• Cross-hatch Infill (CH): Layers oriented at ±45°, creating a cross-hatched pattern (Fig. 1b).

Additionally, molded specimens were fabricated as a comparative reference (Fig. 1c).

G-codes were generated using Simplify 3D software, including a 20 mm offset contour skirt to stabilize extrusion initiation. The 3D PotterBot 10 Micro printer (Fig. 2), equipped with a piston extruder, ensured precision and reproducibility in material deposition. The printer’s characteristics are given in Table 3.

To ensure isotropic reference samples for comparative analysis, the clay paste is molded into side-opening steel molds using a meticulous protocol. The paste is first applied to the mold’s base and corners to minimize air entrapment, followed by incremental layer-by-layer filling of the center to further reduce voids. Each layer is compacted by gently tapping and vibrating the mold exterior, facilitating air release and enhancing homogeneity. Finally, the surface is leveled using a flat-edged ruler to achieve uniform thickness and a smooth finish.

Molded specimens serve as a process-free quasi-isotropic mechanical reference. Apparent density was computed from mass, and volumetric shrinkage measurements for all categories, and their evolution with mechanical properties, are analyzed in 4.3. Direct microstructural characterization (SEM, X-ray tomography) was not feasible in this study and is recommended for future work.

2.2.3. Design of Experiments

This study adopts a full factorial experimental design to evaluate the time-dependent effects of manufacturing procedures and air-drying time (0, 3, 7, 14, 21 days) on the shrinkage behavior and mechanical properties of 3D-printed and molded clay cubes (Fig. 1c).

The design of experiment (DoE) includes three sample categories: Grid-Like (GL), Cross-Hatch (CH), and Molded (M) reference cubes, each tested at five drying time intervals as depicted in Table 4. The freshly manufactured samples (0 days) establish initial properties, while subsequent intervals track progressive drying-induced stiffness development and strength evolution.

The integration of molded samples (M-1 to M-5) serves as a quasi-isotropic control to disentangle the a priori anisotropic effects in layered 3D printing. By testing all combinations of infill patterns and drying durations, this DoE enables robust analysis of time-dependent material behavior, critical for understanding the interplay between infill strategies, drying kinetics, and mechanical performance in 3D-printed bulky clay components.

Three specimens were fabricated and tested per experimental condition, yielding 45 specimens in total (5 drying intervals × 3 categories × 3 replicates). The mechanical properties presented in the Results give the mean values of the three replicates per condition. For the two-way ANOVA analysis of hardened properties, fresh-state specimens (Day 0) were excluded, resulting in 12 mean observations entering the analysis (3 infill strategies × 4 drying times). The use of mean values as ANOVA entries is consistent with the exploratory nature of this characterization study, and inter-replicate variability is quantified in 4.3.3 below.

The following parts present a comprehensive geometrical analysis of the cube specimens used in uniaxial compressive testing (UCT). 2.2.4 introduces a surface area correction method designed to account for barrelling effects in fresh specimens, ensuring a more accurate representation of the likely true-stress evaluation. 2.2.5 focuses on shrinkage analysis over time, which aims to refine the cross-sectional measurements of the cube samples, thereby enhancing the precision of stress calculations for air-dried samples.

2.2.4. Surface Area Correction for Fresh UCT

In this study, a semi-automatic MATLAB-based procedure is employed to correct the surface area of 3D-printed clay samples during fresh UCT (Fig. 3a). The process involves video recording of the specimen during testing and frame processing to estimate the deformed cross-sectional area illustrated in Fig. (3b), as described in the following procedure:

1. Frame calibration to convert pixel measurements to millimetres;
2. Manual capture of vertical and horizontal edges of the specimen during the test at predefined frame intervals for vertical and horizontal strain computations ϵ_v and ϵ_h as indicated in Eqs. (1) and (2);
3. Computation of the barrelling factor (BF) as indicated in Eq. (3);
4. Computation of the corrected surface S₁ area by adding to the initial cross-section area S₀ 4 disk segment areas of chord L ₀ and sagitta as depicted in Fig. (3);

Where:

• L₀ is the initial side of the cubic specimen, equal to 40 mm;
• h is the current height of the specimen under compression;
• L is the current width of the specimen under compression;

Note that h ≤ L₀ and L ≥ L₀ during testing. The barrelling factor is interpreted as a generalisation of Poisson’s ratio in plastic deformation.

2.2.5. Shrinkage Analysis

To quantify dimensional changes during drying, all specimens underwent daily shrinkage measurements. For each sample, four height measurements (h₁, h₂, h₃, h₄), two width measurements (w₁,w₂), and two diagonal measurements (d₁,d₂) were obtained at mid-height (corresponding to the 10th layer for 3D-printed cubes) as depicted in Fig. (4). These dimensions are averaged to compute the shrinkage ratio relative to the initial (Day 0) values, using Eq. (4).

Where D₀ is the initial mean dimension at day 0, and D_t is the mean dimension at time? t.

During drying, temperature and relative humidity are recorded with a thermometer and a hygrometer; dimension measurements are taken via a digital caliper with a precision of 0.01 mm. Measurements were reported by the same operator to minimize variability.

The shrinkage measurements provide critical data on the dimensional changes of samples during the drying process; these measurements are essential for correcting the sample cross-sectional area, ensuring accurate stress calculations during uniaxial compressive tests. Mass shrinkage measurements serve as a macroscopic proxy for moisture state evolution, providing an indirect time-resolved indicator of drying progress rather than a direct quantification of moisture transport through the specimen.

2.3.1. Particle Size and Shape Analysis

The particle size and shape distribution of marl powder, sieved to a 212 µm cut-off and sampled, are characterized using Dynamic Image Analysis (DIA) with a MICROTRAC CAMSIZER X2 system. A dual-camera configuration is employed to simultaneously capture high-resolution images of fine and coarse fractions (0.8 µm–8 mm range). The dry dispersion module (X-Jet) is selected to ensure complete deagglomeration while minimizing particle fracture.

Particle size and shape parameters (aspect ratio, symmetry, sphericity) are calculated using algorithms to quantify the morphology of clay powder.

2.3.2. Uniaxial Compressive Tests

Uniaxial compressive tests were performed using an MTS Criterion C45 universal testing machine; displacement-controlled tests were conducted at a rate of 0.1 mm/s on 40 mm-height specimens, resulting in a strain rate of 0.0025 s⁻¹, ensuring quasi-static conditions well below the recommended maximum of 0.1 s⁻¹ [46, 47]. In the absence of established testing standards for clay-based printable pastes, the 40 mm cube geometry was adopted following cementitious material guidelines [26, 27], a practice commonly used in the emerging 3D clay printing literature. A 10 kN load cell was used to record current strength measurements, while a fixed camera captured sample deformation for subsequent video processing in 2.2.4. To minimize slippage, samples were held between two rough aluminum platens, ensuring non-slipping boundary conditions to reach a quasi-pure vertical deformation.

3.1.1. Particle Size and Shape Effect on Moisturization

Particle size and shape critically influence the moisturization capacity of marl powder; finer particles (<2 µm) possess higher surface-to-volume ratios, enhancing water adsorption [52, 53]. However, excessive fines may promote agglomeration, potentially hindering uniform hydration. The predominance of rounded particles (mean roundness: 0.82) suggests smoother surfaces, which reduce interparticle friction and facilitate water penetration. Conversely, lower aspect ratios (mean: 0.755) imply less elongation, promoting tighter packing and reducing pore space for water retention. Based on the aforementioned analysis, it is worthwhile to overcome such undesirable effects. Hence, the marl paste was rigorously mixed and allowed to rest for 24 hours in a plastic film for complete hydration.

3.1.2. Particle Size and Shape Affect the Mechanical Properties of Clayey Paste

The particle size distribution and shape parameters were previously observed to affect the paste’s mechanical behavior; a leptokurtic distribution with positive skewness indicates a high proportion of fine particles, which can enhance packing density and cohesion but may compromise strength due to reduced interlocking [54]. Rounded particles (high symmetry and roundness) diminish mechanical interlocking, potentially lowering shear strength, while the residual angularity may mitigate this effect by creating localized stress concentrations [55]. Subsequently, the balance between these factors likely governs the paste’s rheological and compressive properties, as observed in analogous clay-cement systems [56].

This discussion suggests that the shear yield stress of the paste will likely be low, given that the powder exhibits a high proportion of fine particles with a spherical and relatively symmetrical morphology. This effect is further compounded by the water content, which predominantly influences the rheological properties of the paste.

3.4.1. Fresh Compressive Tests

The fresh state test, also called “green state,” critically assesses the immediate post-fabricated 3D-printed clay’s mechanical properties, prior to drying or curing; this allowed evaluation of early-stage structural integrity and mechanical strength.

Unconfined Compression Tests (UCT) were performed under non-slipping boundary conditions to enforce pure vertical deformation, with no lateral slippage. Specimens with Grid-Like (GL), Diagonal Cross-Hatch (CH) infill patterns, and molded ones were tested.

Stress-strain profiles revealed nonlinear behavior, reflecting clay’s hydrous microstructure, and three distinct phases as depicted in Fig. (9):

1. A very tight elastic phase where yielding occurs at most 2% strain;
2. A plastic phase initiating strain hardening as depicted by the post-elastic concave area of the curve;
3. A consolidation phase (convex region) characterized by an increase in force with minimal displacement, suggesting progressive particle compaction. This interpretation is consistent with analogous behavior reported in similar clay-based pastes [61, 62], though direct microstructural validation was beyond the scope of this study.

In Figure 10, frames from the video recording, as explained in 2.2.4, are provided at each test phase. The material initially deforms vertically under compressive load while expanding laterally due to the positive Poisson’s effect, leading to early-stage cracking. Upon reaching ~20% vertical strain, extensive surface crack propagation occurs. Crack analysis reveals inclined shear cracks, around 45° to the loading axis from the shear stress, and tensile cracks perpendicular to the loading axis, consistent with Poisson-induced lateral expansion, as visually confirmed in the video-recorded failure sequences (Fig. 10).

Post-test analysis of the smashed specimen presented in Fig. (10a) reveals no cracking in the bulk material, and no visible material slipping at the compressed area. The lateral deformation pattern observed in the test is in close agreement with the mathematical correction of the surface area obtained via video processing from the procedure in 2.2.4 (see Fig. 10b). This correction was employed to compute the true stress-strain behavior from raw load-displacement curves provided in Fig. (15-A2) of Appendix 2.

The video-based strain analysis revealed a linear correlation between vertical and horizontal deformation, with the fitted model's slope representing the barrelling factor as reported in Table 7. This linear relationship implies that Poisson's ratio applicability extends into the plastic deformation regime in this case. Table 7 additionally summarizes Young's moduli at the elastic phase, along with yield stresses and strains. Notably, infill patterns showed negligible influence on mechanical properties.

The apparent density of fresh specimens, reported in Table 7, ranges from 1680.00 kg/m³ (Molded) to 1711.41 kg/m³ (Cross-Hatch), reflecting minor fabrication-induced differences in compaction state. The slightly higher density of printed specimens (GL and CH) compared to molded ones can be attributed to the mechanical compaction exerted by the extrusion process, which promotes tighter particle packing during deposition. Despite these density differences, no systematic correlation with Young's modulus or yield stress is observed across the three categories in the fresh state, suggesting that at equivalent moisture content, the mechanical response is primarily governed by the paste's rheological state rather than by density alone. This is consistent with the dominance of hydrous interparticle interactions in fresh clay pastes reported in the literature [52, 55].

3.4.2. Dry Compressive Tests

Uniaxial compressive tests were conducted on clay cube samples, including 3D-printed specimens with grid-like (GL) and cross-hatch infill patterns, as well as molded specimens, following the dry testing procedure outlined in 2.3.2. Shrinkage data (Tables 14-A1 and 15-A1 of Appendix 1) of width and height were used to correct cross-sectional area and specimen height for accurate true stress and strain computations, a methodology aligned with practices in evaluating anisotropic materials [63]. The resulting stress-strain curves of Fig. (11) and corresponding load-displacement curves of Appendix 2, Fig. (16-A2), reveal key trends in mechanical behavior during compression tests.

With increasing drying time (open-air drying), the clay exhibits a transition toward more linear stress-strain responses, accompanied by a pronounced stiffening effect - evidenced by the progressive upward shift of the curves. This behavior mirrors observations in layered rocks, where stress combinations and material anisotropy influence failure modes [64]. Additionally, the failure mode evolves from ductile to brittle fracture, as indicated by the broadening of the peak at 3 days compared to the sharp, well-defined peak at 21 days.

Similar transitions in fracture patterns under varying curing or drying conditions have been documented in studies on nano-clay stabilized soils and historical clay bricks [65]. Minor load drops in the curves suggest localized crack propagation during testing, a phenomenon consistent with crack initiation and growth mechanisms observed in layered geomaterials [64, 66]. Fracture patterns, consistent with compressive failure, are detailed in 3.5. Notably, the infill strategy had a negligible discernible influence on the stress-strain response.

The derived mechanical properties, including Young’s modulus, peak stress, and peak strain, are summarized in Table 8. These results align with experimental frameworks used to evaluate compressive strength in masonry prisms, where parameters such as brick quality and curing duration critically determine structural performance [63].

Additionally, the relative differences between the parameters obtained for the printed samples and those of the molded reference are calculated. This analysis reveals notable variability in the mechanical properties of printed versus molded specimens. These samples exhibit significantly higher stiffness and strength compared to the printed ones, with differences reaching up to 25.55%. Similarly, the peak stress shows a 9.71% deviation, while the peak strain varies by as much as 20.06%. Furthermore, the molded specimens demonstrate substantially lower deformation than their printed counterparts.

3.4.3. Infill Strategy and Time Effect on Mechanical Properties

To further examine the effect of drying time and scanning strategy on the mechanical properties of the samples, an analysis of variance ANOVA was conducted. In this study, a two-way ANOVA was employed to assess the individual and combined effects of infill strategy (MOLDED, CH, GL) and drying time (3, 7, 14, 21 days) on three key mechanical properties:

• Young’s modulus representing the material’s stiffness;
• Peak stresses representing compressive strength of the material;
• Peak strains representing the ductility of the material;

The apparent density of dried specimens, reported in Table 8, evolves with drying time, reflecting progressive mass loss and volumetric consolidation. At early drying stages (3 days), higher density correlates with higher stiffness and strength, suggesting that compaction state contributes to early-stage mechanical performance. However, beyond 7 days, this correlation weakens - specimens with comparable densities exhibit notably different mechanical properties across fabrication categories - indicating that moisture loss and associated microstructural clay particle reorganization become the dominant governing factors, consistent with the ANOVA finding that drying time is the most significant factor (p < 0.001) for all mechanical properties.

ANOVA decomposes the total variability of the data into contributions from each factor (infill strategy, drying time) and their interaction, followed by a Fisher test to determine the statistical significance (p-value), which should be less than 5% as a rule of thumb. The ANOVA was performed at a significance level of α = 0.05. Twelve mean observations entered the analysis (3 infill strategies × 4 drying times), each representing the mean of 3 replicates per condition. Inter-replicate variability was assessed through the coefficient of variation (CV) across the three specimens per condition. CV values remained below 15% for all mechanical properties across all conditions, consistent with acceptable repeatability thresholds reported for earth-based and clay materials in the literature [67, 68].

The ANOVA results, presented in Tables 9-11, reveal that drying time is the most prominent factor influencing all three mechanical properties, exhibiting statistically significant effects (p < 0.001) on Young’s modulus, peak stresses, and peak strains. Specifically, longer drying periods substantially enhance stiffness and strength while reducing ductility, indicating a clear trade-off between these properties. In contrast, the infill strategy (MOLDED, CH, GL) demonstrates no statistically significant impact (p > 0.05 for all properties), suggesting that variations in infill patterns do not meaningfully alter mechanical performance under the tested conditions. In contrast, this result aligns well with Murica’s findings on 3D concrete printing [69]. Meanwhile, we should note that the infill pattern effect behaves differently with the materials, as it was demonstrated to be significant in polymer 3D printing, as reported by Ouazzani et al. [70].

To complement the significance testing above, partial eta-squared (η²p) effect sizes were computed from the sums of squares reported in Tables 9-11 (η²p = SSeffect / (SSeffect + SSerror)). Drying time showed very large effect sizes for all three properties (η²p = 0.980, 0.995 and 0.976 for Young's modulus, peak stress and peak strain, respectively), consistent with its dominant and statistically significant role. Infill strategy, despite not reaching statistical significance, showed moderate-to-large effect sizes for peak stress (η²p = 0.492) and peak strain (η²p = 0.522), and a smaller effect size for Young's modulus (η²p = 0.233); this contrast between a non-significant p-value and a non-negligible effect size reflects the limited statistical power associated with the low error degrees of freedom (df = 6) rather than a definitive absence of effect, and the finding should be interpreted with this caveat in mind. Regarding the interaction between infill strategy and drying time, it was not evaluated: the ANOVA was performed on the twelve condition-level means (one mean per infill strategy × drying time combination) rather than on the 45 individual replicate measurements, leaving no residual degrees of freedom available to estimate an interaction term within this design. Testing the interaction and computing confidence intervals around condition means would require re-running the analysis on the raw replicate-level dataset.

In the following, the mean effect charts are analysed to investigate the mechanical properties evolution during drying.

3.4.4. Mechanical Properties Evolution during Drying

The ANOVA results confirmed that drying time governs the evolution of mechanical properties in clay samples, while infill strategy showed negligible influence (p > 0.05 for all properties). To quantify this time dependency, exponential models were fitted to the mean values of Young’s modulus , peak stresses , and peak strains across all infill strategies. The models are presented in Eqs. (6-8).

Young's moduli and peak stresses, depicted in Fig. (12a and b), follow distinct exponential growth trends, reaching asymptotic values of 130.6 MPa and 3.08 MPa, respectively, with rate constants τ) of 0.06 and 0.14 days⁻¹. These models demonstrated excellent agreement with experimental data, evinced by R² values exceeding 95% as presented in Table 12.

Peak strain exhibited a different behavior, following a shifted exponential decay model (Fig. 12c) with a rapid initial decrease (τ_ϵ = 0.32 days⁻¹) before stabilizing at 3.35% residual strain. This inverse relationship between ductility and stiffness reflects the material's progressive hardening during drying. Additionally, the extrapolation of peak strains to initial time ( ) aligns with the range of strain at the end of the plastic phase obtained from fresh UCTs in Fig. (9).

While these models effectively describe the mid-to-late drying phases, early-age behavior (t < 1 day) reveals more complex dynamics. Chen et al. [8] observed slower initial property development; they reported convex strength-time curves in fresh clay samples during the first 240 minutes post-printing, suggesting that logistic growth models might better capture the complete drying timeline by accounting for the delayed onset of property development. This trend suggests a predominance of thixotropic effects over drying actions, as recalled by Oulkhir et al. in their systematic review [14].