Investigation of binder jet 3D printing parameters of WC-Co using design of experiments

Introduction

Typically used in tooling, mining, machining, and wear parts, tungsten carbide-cobalt (WC-Co) is a ceramic-metal composite known for its combination of high hardness, provided by a 3D skeleton of WC grains, and high toughness, provided by the Co metallic binder phase.^1–5 Commonly, WC-Co parts are produced through powder metallurgy (PM) processes such as die pressing or powder injection moulding (PIM). These techniques result in dense green parts consisting of powder held together by a plasticiser.^6–10 The green part is then machined to include additional geometries before densification via high-temperature liquid phase sintering and Hot Isostatic Pressing (HIP).^11,12 Although PM has been used for almost a century, more complicated part designs with internal geometries are extremely difficult with the current capabilities. In addition, experimental part designs are expensive and time consuming to produce, especially if only a small quantity is desired, due to the necessary custom die fabrication. The adoption of Additive Manufacturing (AM) offers new opportunities for the cemented carbide (hardmetals) industry. Sinter-based AM, such as Binder Jet 3D printing (BJT), is of particular interest for WC-Co production due to the ease of integration into the current manufacturing process, when using the same powder feedstock and sintering conditions. Similar to PM methods, after BJT, parts can be sinter-HIPped or sintered and infiltrated to increase final density.

In the case of WC-Co, both post-processing routes, sinter-HIP and post-sintering infiltration, have been attempted after BJT. However, on the one side, Co-infiltration of green and sintered parts required high Co amounts (∼32 vol.%) for wicking and shape retention, constraining this path to fabricating parts for high-toughness applications.^13,14 On the other hand, sinter-HIP has been more extensively studied for BJT of WC-Co, including studies on the effect of heat-treatment temperature^15,16 and pressure. ¹⁷ Enneti et al. showed the benefits of sinter-HIP compared to sintering alone for WC-12 wt.% Co, as density increased from 13.5 to nearly ideal 14.2 g/cm³. ¹⁶ Additionally, the printed and sinter-HIPped WC-Co parts by Enneti et al. showed a superior wear resistance because of the presence of a bimodal WC grain size distribution, ¹⁸ but heterogeneous hardness between islands of larger grains within the small-grained matrix. ¹⁹ Similar microstructures displaying dual grain size combinations were also reported by Wolfe et al. for fine and coarse WC-Co grades but were not evident for the extra-coarse grade.^17,20 However, the presence of large-grain islands in the microstructre was attributed to the initial powder feedstock heat-treatment rather than the printing process. In these studies, led by Global Tungsten and Powders Corp., the powder feedstock was prepared via agglomeration and spray drying of the desired WC grains with Co powder, followed by high temperature plasma spheroidisation ¹⁷ or pre-sintering ²¹ of the agglomerates, which causes variation in the WC grain size.^17,22 Further stabilisation of the granules has been explored to increase powder flowability for 3D printing, through wet mixing of the pre-sintered feedstock with a thermoplastic binder (solvent on granules, SOG). ²³

While these approaches have resulted in increased printed green density values (up to 45% ²² ) by powder feedstock treatments, they represent an additional processing step that deters the advantages of AM. In terms of traditional WC-Co powder with untreated agglomerate particles, initial research has also been conducted to explore BJT feasability. However, because the un-sintered feedstock is porous and unstable and has low packing fraction and low flowability, sieving^24,25 and crushing ²⁶ have been attempted to increase powder packing fraction and powder–binder interaction area and thus, green density.

Because improving green density generally results in higher sintered density and contributes to lessening part damage during transfer and handling, optimisation of BJT process parameters is an alternative route to powder feedstock modifications, although their relative impact has not been assessed. This approach is of particular interest when the printed parts are aimed at being introduced into the traditional manufacturing process, where it is desirable to maintain the same powder characteristics and sintering conditions that have been chosen for PM processing. BJT print parameter optimisation is often carried out by changing one variable at a time (OVAT) to adjust for powder complexity (intricate morphology, agglomerate instability, low flowability, low spreadability, humidity) and powder–binder interactions (particle ejection, delamination, and bleeding controlled by the binder saturation and drying time).^24,27,28 However, some systematic BJT Design of Experiments (DoE) attempts have been conducted on other materials, particularly on spherical steel powders^29–33 with the general objective of optimising green density, surface roughness, and dimensional accuracy (Table 1).

OPEN IN VIEWER

Table 1.

BJT literature design of experiments, with initial powder feedstock, parameters studied (optimum parameters bolded), and the effect of the main parameters on selected responses (abbreviations in Appendix A).

WC-Co BJT OVAT experiments have been previously performed to optimise green density and strength of pre-sintered agglomerates and plasma spheroidised powders through binder saturation and layer thickness iterations^27,28 and through binder saturation and drying time setting combinations for un-sintered broken agglomerates. ²⁴ Results indicate that higher binder saturation levels (70%) and lower layer thickness (50 μm) improve the green strength, ²⁸ but deposited binder should be optimised in conjunction with drying time ²⁴ to improve green density. ²⁴

Because little agreement of the optimum printing parameters exists between studies, and the parameters are highly dependent on powder size and morphology, new powders require new parameter optimisations. This study focused on systematically evaluating the effect of BJT spreading parameters (layer thickness, roller rotational speed, and feed-to-build ratio) and powder-binder parameters (binder saturation, BS, and drying time, DT), through a two-step full factorial DoE with the overall goal of increasing green and sintered density of un-sintered WC-Co powder feedstock. Because of the complexity of agglomeration, the powders were first sieved and characterised, and then, the print parameter range was selected based on previous OVAT experiments. ²⁴ The first DoE step targeted packing fraction and the second DoE density optimisation (while utilising the first DoE's results). The sinter-HIPped properties were also characterised, including density, grain size, hardness, indentation fracture toughness, and magnetic properties, to contextualize BJT parts with traditional industry standards for a given WC-Co composition.

Materials and methods

Spray-dried powder (un-sintered, low-temperature de-waxed, spherical granules) from General Carbide Corporation with 12.1 wt.% Co was printed using an ExOne X-1 Lab binder jet printer with solvent-based binder after 38 μm sieving. The first randomised full factorial DoE (two levels, three factors) was used to determine the spreading parameters (layer thickness, roller traverse speed, and feed ratio) for optimal powder packing measured via blank prints. The chosen parameters were used in a second randomised full factorial DoE (three levels, two factors: binder saturation and dry time, Table 2) to determine the printing parameters for achieving optimal green and sintered densities.

OPEN IN VIEWER

Table 2.

Randomised printing parameters runs for the first and second DoE.

First DoE (DoE1)			Second DoE (DoE2)
Layer thickness [μm]	Roller traverse speed [mm/s]	Feed ratio [–]	Binder saturation [%]	Dry time [s]
100	5	1.5	120	45
50	15	1.5	100	45
50	15	2	120	60
50	5	1.5	80	60
100	5	2	80	30
100	15	2	120	30
50	5	2	100	60
100	15	1.5	80	45
			100	30

For each run of the second DoE, 12 coupons (10 × 8 × 5 mm³) were printed and cured (8 h at 180 °C). After de-powdering, green density was measured using callipers and a balance (Ohaus AX324 ± 0.1 mg). Multiline surface profile (P_a) and roughness (R_a) were measured with a Keyence VR-3200 optical profilometer. Tilt correction and a cutoff wavelength of 8 mm (31 lines, 4 mm line-length, perpendicular to jet lines, 2 samples per condition) were used. ³⁴ Sinter-HIP was performed at General Carbide at 1440 °C. Sintered densities were measured via the Archimedes method. Three parts from each run were cold mounted for xy, xz, and yz plane examination. Vickers hardness tests were performed on these three polished planes. Three indentation fracture toughness measurements were acquired using a Rockwell Wilson 754 with a custom Vickers indenter (30 or 100 kgf load, following ISO 28079). Indentation edge cracks were observed using a Zeiss SmartZoom5 optical microscope, and crack lengths were measured using ImageJ. ³⁵ Thealmqvist indentation fracture toughness was calculated using Shetty's model (A = 0.0889).^36,37

Secondary electron micrographs were taken using a Zeiss Sigma 500 VP scanning electron microscope (SEM). General Carbide performed LECO analysis for carbon content evaluation. Tungsten and cobalt contents were measured with X-ray fluorescence (XRF) by Clark Testing. Magnetic hysteresis loops were measured using a LakeShore 8600 vibrating sample magnetometer (VSM).

Results and discussion

Powder characterisation

The feedstock powder before (as-received) and after sieving is shown in Figure 1. The original spherical and porous granular powder (soft agglomerates) was broken during sieving, resulting in fine powder, as characterised in Rodriguez De Vecchis et al. ²⁵ The fragmentation of the granules into their individual, irregular, primary particles or groups of particles decreased the packing fraction of the powder feedstock, but contributed to increase printability, as the porous granules had acted as binder soaking sites.

OPEN IN VIEWER

Figure 1.

SEM images of spray-dried powder before and after sieving.

Design of experiments and microstructure

The DoE1 results for optimising packing fraction are summarised in the parameter effect plot in Figure 2, indicating that slower roller traverse speed (5 mm/s), larger layer thickness (100 μm), and larger feed-to-build ratio (2) contribute to increase the apparent density of the build bed. Similar conclusions have been reached previously for parts printed with spherical powders. A higher feed-to-build ratio ³⁰ ensures complete, more consistent layer coverage of the build bed increasing packing. In contrast the green strength has reportedly decreased, ²⁹ likely due to reduced interlayer bonding when providing too much powder. Slower roller traverse speed slightly increased green density and decreased green surface roughness ³⁸ as slower speeds resulted in better coverage and consistency across the build bed for multiple printed parts. Lastly, layer thickness is generally inversely proportional to green density. ³⁰ In this case, though, as in Shanmuganatan et al., ³³ the larger value seemed to promote better packing. This is potentially caused by the low flowability of powders, which are more consistently spread when thicker feeding is provided, giving the opportunity for more compaction from the roller and to accomodate the heterogeneous feedstock. In this case, the layer thickness parameter provides a similar influence on green density as the powder-to-feed ratio, with a stronger impact from the feeding than the building side of the powder build-bed. However, for this study, none of the parameters appear to be highly significant as their individual parameter effects were smaller than the t-statistic value, suggesting the effect of the unstable sieved agglomerated feedstock is more impactful on the packing behaviour than any spreadingparameter. Together, the three parameters and their interaction account for nearly 13.5% of the packing fraction variability (adjusted R²).

OPEN IN VIEWER

Figure 2.

Spreadability DoE1 for optimising packing fraction based on (a) feed-to-build ratio, (b) roller traverse speed, and (c) layer thickness.

For green density optimisation (DoE2), individually, higher saturation and intermediate drying time appear to be preferred, but little variation was observed on the overall green density range (22.6–24.1%), with standard deviations near 1%, across parts of the same print. The interaction of the two variables, binder saturation and drying time, had the strongest effect on the density results, followed by drying time alone, according to the P-values obtained from the ANOVA on 10 samples per condition, for a significance level of 0.05. Quantitatively, the P-values resulted in magnitudes of 0.023 for drying time, 0.948 for binder saturation and 0.004 × 10⁻³ for their interaction. The variables together account for nearly 30% of the green density variation, as quantified by the adjusted R². Figure 3 shows a summary of the parameter impact and the optimised parameter window for density and surface quality measurements. Therefore, it is statistically concluded that drying time and saturation must be optimised together in order to maximise the green density magnitude, as lower saturation values require shorter drying time to avoid surface over-drying and delamination, and higher saturations require longer drying times to avoid layer smearing and binder bleeding and deformation.

OPEN IN VIEWER

Figure 3.

Powder-binder DoE (2) for (a–b) green density, (c–d) green Pa, (e­–f) green Ra, and (g–h) sintered density responses. Layer boundary contour maps were generated in Origin with 100 total points increase factor.

Sinter-HIP density values follow a similar trend to that of green density, with a maximum value of 97.8% for the 100%–45 s combination (DoE2, Run 2). Similarly, the ANOVA (6 samples per condition) for this response suggests that the interaction between drying time and binder saturation is the significant variable input (0.039 P-value), accounting for 14% of the response variation (adjusted R²), even after densification, thus requiring them to be optimised together when performing BJT parameter studies. Differently from the results found by Mariani et al. ²⁷ the chosen range of parameters in this study had a stronger impact on the sintered density response range (94.0–97.8%) compared to their variations between 60% and 90% binder saturation sinter-HIP density changes (96.0–97.4%). ²⁷

On the other hand, surface profile and roughness measurements suggest that higher binder saturation and lower drying time combinations result in better surface finish, possibly because of surface flattening effects during excessive binder deposition. Results by Lores et al. for spherical Invar36 powders showed a similar trend as higher binder saturation resulted in increased particle bonding and therefore less detachment, surface irregularities, and cracks that might disrupt the powder bed. ³⁹

Evidence ²⁷ of internal and external porosity, that hindered densification (~ 97–100% typically acquired with traditional manufacturing methods) can be seen in the representative cross-sectional optical micrographs in Figure 4. Both porosity and cracking were seen, with Co-pools appearing (Appendix B) along printhead jet lines (vertical lines on xy cross-section) and between layers in the xz and yz sections. The larger defects connected to the surface could not be removed even after the sinter-HIP treatment.

OPEN IN VIEWER

Figure 4.

Optical micrographs of the complete sinter-HIP sample for each DoE2 print run, in each orientation.

It was observed that for large binder saturation values (120%), both xy jet lines and xz, yz interlayer porosity are present regardless of the drying time, while for low binder saturation (80%), only interlayer porosity is observed, particularly at the highest drying time (60 s), but jet line defects are minimised. However, poor shape retention resulted from uneven edges and detached corners for all drying times. Similarly, for the intermediate binder saturation (100%), a long drying time (60 s) resulted in poor shape retention and layer detachment. The intermediate drying time (100%, 45 s) resulted in jet-line porosity on the xy surface and random spherical porosity in the xz direction, but an overall smaller porosity. The deep-grooved cracking in the direction of the jet lines observed in 80%–45 s, 100%–30 s, and 100%–60 s is attributed to the low packing fraction of the WC-Co powder used.

Sintered properties

SEM microstructural examination revealed Co-pooling, more significant in size and frequency for the lower green density samples. Grain sizes were measured through SEM micrograph analysis as proposed by Tarrago et al., ⁴⁰ but using water-shedding instead of blurring to delineate boundaries. Grains were idealised as having squared cross-sections (0.1 μm² minimum area filter). All samples showed a similar grain size distribution (Figure 5(a)) with a mean value of 1.2 ± 0.8 μm for Run 2, (100%, 45 s) and similar for other runs (for example, 1.3 ± 0.7 μm for Run 8–80%, 45 s, and 1.1 ± 0.6 μm for Run 9–100%, 30 s), falling within the fine-medium grain categories. ⁵ Elemental analysis by LECO and XRF revealed, 5.35 wt.% C, a Co amount between 10.3 and 11.8 wt.% for three measured samples, and a W-to-C ratio of 1.00–1.04, indicating no excess C was introduced by the binder, and confirming no carburising step is required during heat-treatment as performed by Enneti et al. ¹⁶

OPEN IN VIEWER

Figure 5.

Sinter-HIP properties: (a) grain size distribution with insert of SEM micrographs, (b) Vickers hardness distribution per printing condition, with grey range for the General Carbide GC313 grade, ⁴¹ (c) Fracture toughness distribution per printing condition with grey range for General Carbide fine-medium grades with Co content between 10.3 and 11.8 wt.%, ⁴¹ and (d) VSM magnetisation per printing condition with grey range as suggested for the 10.3–11.8 wt.% Co range suggested by Topić et al., ⁴² and insert of the coercivity values.

For mechanical properties, hardness values (Figure 5(b)) appear within the range of the starting powder feedstock, the Co content, and the grain size. ⁴¹ However, indentation fracture toughness has a larger magnitude than expected (Figure 5(c)). It is thought that this readings are caused by the Co-pools in the microstructure, given that no significant detrimental phases were detected (Appendix C). The large voids inherited from the printing process, the low green density, and the handling process are recovered during sinter-HIPping through Co filling, which results in an acceptable density but excessive fracture toughness.

Finally, magnetic coercivity (H_c) and magnetic saturation (M_s) results from VSM hysteresis curves are presented in Figure 5(d). Magnetic measurements are complementarily used for cemented carbides because they provide information about the Co content and the WC grain size. Because Co is the only ferromagnetic component in the microstructure, M_s is directly proportional to its content. ⁴² Additionally, all other detrimental Co-containing phases and the dissolution of W in the binder decrease the M_s. From the linear relationship between M_s and Co-content shown by Topić et al., ⁴² the expected M_s for this study is shown in Figure 5(d) for a 10.3–11.8 wt.% Co range. Higher magnetisation values for some DoE2 runs are attributed again to the Co-pooling promoted by the low green density, as mentioned above, and consistent with the high indentation fracture toughness. Alternatively, H_c is more influenced by the WC grain size, as an increase in Co-WC boundary reduces demagnetisation. ⁴² Topic et al. developed an equation for the WC grain size based only on the H_c measurement for ultra-fine to coarse-grained microstructures, ⁴² which in this case resulted in a calculated grain size of 0.81–0.91 μm for the measured range of all sintered samples. This value falls within the submicron category for the overall structure, suggesting the few larger grains in the microstructure significantly skew the grain size distribution measured through image analysis, although the two results are still within deviation of each other.

An overall comparison of the tested properties of sintered parts is shown in Table 3, for a high density (optimised, 100%–45 s), a medium density (100%–30 s), and a low density (80%–45 s) sample, along with reference values reported by General Carbide for the respective powder grade. ⁴¹

OPEN IN VIEWER

Table 3.

Comparison of the sintered properties for three different DoE runs (high, medium, and high density) and reference general carbide values.

	DoE2 run (BS – DT)
	High-density	Medium-density	Low-density	Comparison
Property	100%–45 s	100%–30 s	80%–45 s	GC 313
Relative green density [%]	23.80	23.49	22.71	–
Relative sintered density [%]	97.75	96.22	95.25	97–100
Grain size [μm]	1.2	1.1	1.3	–
Hardness [HV]	1310	1311	1293	1165–1430
Fracture toughness [MPa*m1/2]	20.5	23.5	28.0	–
Magnetic saturation [emu/g]	19.9	22.1	18.0	–
Magnetic coercivity [Oe]	135	122	127	–
Carbon [wt.%]	5.4	5.3	5.4	5.4
Cobalt [wt.%]	10.3	11.3	11.8	12.1

Conclusions

BJT of WC-Co cermet parts has previously been attempted with relative success for heat-treated pre-sintered agglomerate powders. However, easier integration of this AM technique in traditional press-and-sinter powder metallurgical methods would be facilitated by employing un-sintered soft agglomerates. These granules and their composing primary particles have large porous fractions, and poor flowability that difficult fabrication of stable BJT green parts. In this study, a systematic approach for identifying significant printing parameters and their impact on green and sintered WC-Co parts was pursued through two DoEs, exploring first flowability parameters (roller speed, layer thickness, and feed-to-build ratio) and then powder-binder interaction settings (binder saturation and drying time) for sieved, un-sintered WC-12 wt.% Co powder feedstock.

Results suggested that spreading parameters have an overall low impact on the powder packing behavior of the sieved feedstock, signalling the stronger effect of the original powder condition (stable vs soft agglomerates), as long as sufficient powder is provided to cover the build bed. The sieved agglomerates result in an extremely low packing fraction of 21%. The low packing, printing, and handling difficulties were the motivation for further parameter exploration. For the second DoE, it was confirmed that binder saturation and drying time should be optimised together, as their linear interaction had the most statistically significant impact on the green and sintered density. These variables accounted for nearly 30% of the green density variation. Green surface profile and roughness, as indicators of green part quality, showed a different trend than density, as mismatched values of large binder saturation and short drying time resulted in more even surfaces, likely due to flattening when oversaturated with binder.

After sinter-HIPping, final part density trended similarly to the green density with a maximum of 97.8 ± 1.8% (100% binder saturation, 45 s drying time, 5 mm/s roller speed, 100 μm layer thickness, 2 feed-to-build ratio). The sintered microstructure showed a fine-medium grade, normally distributed grain size (1.2 ± 0.8 μm) and hardness values (1310.4 ± 51.6 HV) in-range for this WC-Co grade and Co-content. However, regions of Co-pooling were observed, likely formed during liquid phase sintering of the highly porous green parts, which contributed to increase the indentation fracture toughness (20.5 ± 7.5 MPa·m^1/2) and M_s (19.9 emu/g). Lastly, no excess C was detected, suggesting the C-based binder did not negatively impact the composition.