Gypsum-bonded molds occupy a specific niche in investment casting. A wax or printed pattern is invested in a slurry of calcium sulfate hemihydrate, refractory fillers and water; the slurry sets, the pattern is melted out below the dihydrate transition temperature, and the mold is calcined before pouring. The process reproduces fine surface detail, needs no high-pressure tooling, and tolerates thin sections that are difficult to produce by sand casting. The price of that capability is paid in the slurry itself: a single formulation must simultaneously flow into every pattern recess, remain workable long enough to be poured and de-aired, develop enough green strength to be handled, survive calcination without cracking, hold dimensional tolerance through a shrinkage of several percent, and resist the thermal shock of liquid metal.
These requirements conflict, and the conflict is traceable to the binder. Water-to-binder ratio governs fluidity and, inversely, porosity and strength; pure set gypsum shrinks roughly 3.75% when fired at 700 °C, far outside the 0.5–1% linear shrinkage window normally accepted for dimensionally critical castings; and the low thermal conductivity of gypsum generates steep internal gradients during heating, which is why calcination schedules are usually held to 1–2 °C/min. Formulation work therefore consists of three coupled adjustments: refractory fillers that offset binder shrinkage and raise thermal stability, chemical additives that decouple fluidity from water content, and fibers that bridge cracks. The data discussed below come from a systematic laboratory program on α-hemihydrate-based casting powder in which these three levers were varied independently.
Mold families and the choice of binder fraction
Gypsum investment molds are conventionally divided by composition and by stripping method. Non-foaming molds contain 45–55 wt% hemihydrate and reach compressive strengths of about 2–3 MPa but have poor permeability, so they suit low-pressure casting and simple shapes. Investment (lost-wax) molds contain 25–37 wt% hemihydrate, are dewaxed at low temperature and then calcined; they can be poured by gravity, centrifugal or vacuum methods. Foamed molds use a surfactant or a chemical blowing agent such as H2O2 or NaHCO3 to expand the slurry by more than 50% by volume, gaining insulation at the cost of mechanical properties.
The work described here targets the lost-wax variant at a hemihydrate fraction of 35 wt%, a mid-range value chosen to leave room for a high filler loading. The set reaction is the standard dissolution–precipitation sequence:
$$\mathrm{CaSO_4\cdot \tfrac{1}{2}H_2O + \tfrac{3}{2}H_2O \rightarrow CaSO_4\cdot 2H_2O}$$
Dihydrate has the lower solubility, so once the hemihydrate solution becomes supersaturated, dihydrate nuclei form, grow and interlock into the network that carries load. Anything that changes the supersaturation — water content, soluble ions, water-absorbing solids, retarders that sequester Ca2+ — changes both setting time and final strength. The reverse dehydration during calcination proceeds in stages, with dihydrate converting to hemihydrate near 120 °C, to soluble anhydrite near 230 °C and finally to insoluble anhydrite between 400 and 800 °C.
Baseline measurements on the α-hemihydrate alone set the working window. At a water-to-gypsum ratio of 0.4 the slurry would not flow at all; at 0.5 it spread to 120 mm with initial and final set at 6 and 10 min; at 0.6 it spread to 195 mm with set at 10 and 18 min. A ratio of 0.5 was adopted for the filler study, at which the neat binder gave 4.71 MPa wet flexural and 10.55 MPa wet compressive strength, 1.76 MPa dry flexural and 6.37 MPa dry compressive strength, and 2.0% linear shrinkage — roughly four times the acceptable limit.
| Flow diameter (mm) | Initial set (min) | Final set (min) | Wet flexural (MPa) | Wet compressive (MPa) | Dry flexural (MPa) | Dry compressive (MPa) | Linear shrinkage (%) |
|---|---|---|---|---|---|---|---|
| 120 | 6 | 10 | 4.71 | 10.55 | 1.76 | 6.37 | 2.0 |
Refractory fillers: trading shrinkage for fluidity
The filler system combined quartz (325 mesh, 99.2% SiO2), mullite (325 mesh, Al2O3 plus 3Al2O3·2SiO2), coal gangue (200 mesh, 55% SiO2 and 32% Al2O3) and diatomite (325 mesh, ≥85% SiO2, specific surface area ≥60 m2/g, bulk density 0.36–0.44 g/cm3). Quartz was fixed at 8 wt% because its expansion between 200 and 300 °C counteracts binder contraction in that interval, but its transformation near 570 °C, accompanied by about 0.82% volume expansion, seeds microcracks and is best kept to a modest fraction. Mullite, gangue and diatomite made up the balance in a nine-formulation matrix.
| No. | α-hemihydrate | Quartz | Mullite | Coal gangue | Diatomite |
|---|---|---|---|---|---|
| 1 | 35 | 8 | 47 | 10 | 0 |
| 2 | 35 | 8 | 47 | 5 | 5 |
| 3 | 35 | 8 | 47 | 0 | 10 |
| 4 | 35 | 8 | 37 | 20 | 0 |
| 5 | 35 | 8 | 37 | 15 | 5 |
| 6 | 35 | 8 | 37 | 10 | 10 |
| 7 | 35 | 8 | 27 | 30 | 0 |
| 8 | 35 | 8 | 27 | 25 | 5 |
| 9 | 35 | 8 | 27 | 20 | 10 |
Both gangue and diatomite are porous. Scanning electron microscopy shows diatomite as disc-shaped particles perforated by circular pores of varying diameter, and gangue as a stack of lamellae with loose, void-rich gaps between them. Both absorb water from the slurry, and diatomite, with its far larger specific surface, absorbs more strongly. The consequence is a direct loss of flow: at 47% mullite, replacing mullite with 5% and 10% diatomite reduced the spread diameter by 25.0% and 50.0%; at 37% mullite the losses were 26.3% and 51.6%, and at 27% mullite 30.2% and 56.1%. Adding gangue at fixed diatomite content cost at least 14.6% of fluidity. A 15% diatomite loading at a water-to-gypsum ratio of 0.5 produced a slurry that could not be stirred at all, which is why the diatomite fraction was capped at 10%.
Setting accelerates in parallel, for the same reason. Absorbed free water raises the supersaturation of the remaining solution, so dihydrate nucleates and grows faster. Each additional 5% of diatomite shortened the initial and final set by roughly 1 min, and increasing gangue shortened both times as well. Formulation 9, the most water-hungry of the set, still gave a workable 7 min initial and 17 min final set with a 90 mm spread, whereas formulation 1 (10% gangue, no diatomite) spread to 240 mm with set times of 13 and 26 min.
Fired strength moves in the opposite direction to fluidity, and here the porous fillers are beneficial. Absorbing free water lowers the effective water-to-binder ratio locally, so more hemihydrate participates in hydration and the dihydrate network grows denser. At 10% diatomite, fired compressive strength rose by 94.6%, 28.8% and 57.0% relative to the diatomite-free molds at 47%, 37% and 27% mullite respectively; fired flexural strength rose by 33.9%, 22.6% and 52.7% over the same series. Without diatomite, raising gangue from 10% to 20% and 30% still increased fired compressive strength by 30.4% and 21.1%. Micrographs of fracture surfaces show the reason: calcined diatomite is tightly embedded in a gypsum matrix that wraps the filler particles, and in the diatomite-bearing formulations the differently shaped grains pack into a compact, continuous skeleton. Without diatomite, the residual free water leaves larger voids after calcination and the intergranular contacts are visibly loose, reducing the load-bearing cross-section.
| Property | Formulation 1 (20% mullite-equivalent filler, 10% gangue, 0% diatomite) | Formulation 9 (20% gangue, 10% diatomite) |
|---|---|---|
| Flow diameter (mm) | 240 | 90 |
| Initial / final set (min) | 13 / 26 | 7 / 17 |
| Wet flexural strength (MPa) | 1.17 | 2.02 |
| Wet compressive strength (MPa) | 1.78 | 2.30 |
| Fired flexural strength (MPa) | 0.295 | 0.458 |
| Fired compressive strength (MPa) | 0.56 | 1.07 |
| Linear shrinkage (%) | 0.669 | 0.320 |
Shrinkage follows the same trend as strength. Diatomite at 5% and 10% lowered fired linear shrinkage by 17.6% and 28.3% at 47% mullite, by 22.0% and 31.6% at 37% mullite, and by 17.5% and 54.9% at 27% mullite. Gangue without diatomite was less effective but still stable: relative to 10% gangue, additions of 20% and 30% raised shrinkage by only 3.1% and 6.1%, indicating that gangue undergoes smaller dimensional change on firing than mullite. Across all nine formulations shrinkage stayed at or below the upper bound of the 0.5–1% acceptance window, with a minimum of 0.320%. Thermogravimetry supports the interpretation: the α-hemihydrate lost 5.47% of its mass against a theoretical 6.20% for pure hemihydrate, consistent with a small dihydrate and soluble-anhydrite fraction detected by XRD; quartz and diatomite were mass-stable to 800 °C; mullite lost 4.3% between 640 and 740 °C; and gangue lost 10.1% by 800 °C as carbon and sulfur burned off above 360 °C. The blended formulation lost 15.6% against a calculated 11.6%, most of it (12.4%) below 200 °C, so the bulk of the volatile load is removed early in the cycle.
Chemical additives: decoupling fluidity from water content
If fillers set the strength–shrinkage balance, additives address the fluidity–strength conflict at fixed solids loading. Four were examined against a baseline slurry spreading to 90 mm: a polycarboxylate ether superplasticizer, a melamine-formaldehyde (BASF F10) superplasticizer, a colloidal silica sol (29–31% SiO2, pH 9.0–10.5, 12–20 nm particles) and boric acid.
The polycarboxylate was the more efficient dispersant. At 0.1% it raised the spread to 200 mm, against 150 mm for melamine at the same dosage, and 0.15% of polycarboxylate already matched the 220 mm obtained at melamine saturation. Saturation dosages were 0.4% and 0.8% respectively. The difference lies in mechanism: melamine works mainly by electrostatic repulsion, which is progressively screened as hydration products coat the particles, whereas polycarboxylate combines electrostatic repulsion with steric hindrance from grafted side chains that keep particles apart. Melamine also proved unstable in storage — at 0.3% the spread fell from 190 mm on day 0 to 155 mm on day 7, and at 0.8% from 220 to 180 mm — so its dosing would require re-qualification of each batch.
Retardation is the penalty. Polycarboxylate at 0.1% extended initial and final set to 20 and 34 min, at 0.2% to 50 and 75 min, and at 0.8% to a final set of 350 min; at the strength-optimum dosage of 0.4% the values were 140 and 220 min. Melamine retarded far less, with a maximum final-set extension of 47.1% and longest values of 12 and 25 min. The contrast with Portland cement is instructive: 0.1% polycarboxylate lengthens gypsum set by about 100% but lengthens cement set by only 5–9%, because the organic functional groups complex the relatively abundant free Ca2+ in the gypsum slurry and suppress dihydrate nucleation. For a mold shop, a 220 min final set at the optimum dosage is a serious throughput constraint, and a low water-to-gypsum ratio combined with polycarboxylate failed outright — at 0.5 water-to-gypsum the slurry could not be mixed, so the dispersant had no opportunity to act. Strength specimens were therefore cast at 0.6 water-to-gypsum with dosages adjusted by the measured water-reduction rate.
Fired strength peaked at 0.4% polycarboxylate and 0.8% melamine and declined beyond those dosages. At the same 0.4% addition, polycarboxylate-containing molds exceeded melamine-containing molds by 91.2% in flexural and 118.1% in compressive strength, reflecting its much higher water reduction. Past saturation, however, the water reduction no longer improves, while the entrained air and the retardation both coarsen the dihydrate network — so the flexural strength of over-dosed polycarboxylate molds fell below that of the corresponding melamine molds.
Colloidal silica acts differently, since it is a slurry-phase additive rather than a dispersant. Mixed into the gauging water at water-to-silica-sol ratios from 1:0 to 1:1, it raised the spread diameter to a maximum at 4:1 and then reduced it, with flow lost entirely once the sol exceeded the water fraction; ratios below 1:1 are therefore unusable. Setting was retarded most strongly over the 1:0 to 4:1 interval, with initial set extended by 214.3% and final set by 70.6%. Fired strength increased monotonically with sol content: at 4:1 the compressive strength was 51.4% and the flexural strength 44.1% above the additive-free mold, and at 1:1 the gains reached 231.8% and 153.3%. The sol fills intergranular voids, and its gelling ability binds adjacent gypsum particles into a tighter framework; sub-20 nm particles penetrate capillary cracks, which is the same sealing mechanism exploited in grouting and reservoir applications.
Boric acid had no effect on fluidity at any dosage from 0.1% to 10%, and only a marginal retarding effect at low loading (0.5% delayed initial set by 1 min and final set by 3 min). Its value is in fired strength, which rose steadily with dosage and nearly linearly above 3%: 1.7 MPa compressive and 0.6 MPa flexural at 3% (up 58.9% and 31.0%), and 3.01 MPa compressive and 1.22 MPa flexural at 10% (up 181.3% and 166.4%). The mechanism is physical rather than chemical — boric acid does not change gypsum crystal size, but it precipitates from the supersaturated pore solution between grains during setting and fills the interstices, raising density. Consistent with a pore-filling mechanism, the compressive gain consistently exceeded the flexural gain except at the lowest dosage.
| Additive | Practical dosage | Effect on fluidity | Effect on setting | Effect on fired strength |
|---|---|---|---|---|
| Polycarboxylate superplasticizer | 0.4% (saturation) | 200 mm at 0.1%; 220 mm at 0.15% | Severe retardation; 140 / 220 min at 0.4% | Peak at 0.4%; declines above saturation |
| Melamine superplasticizer | 0.8% (saturation) | 150 mm at 0.1%; 220 mm at saturation | Mild; maximum final-set extension 47.1% | Peak at 0.8%; lower than polycarboxylate at equal dosage |
| Colloidal silica sol | Water : sol 4:1 to 1:1 | Maximum at 4:1; no flow below 1:1 | Initial set +214.3%, final set +70.6% (1:0 → 4:1) | +51.4% compressive, +44.1% flexural at 4:1; +231.8% and +153.3% at 1:1 |
| Boric acid | 3–10% | None | Marginal at low dosage | +58.9% compressive at 3%; +181.3% compressive and +166.4% flexural at 10% |
Glass-fiber reinforcement and interface control
Gypsum is brittle, and the failure mode of a gypsum mold under pouring pressure or thermal shock is crack initiation followed by rapid propagation. Glass fiber, with a Young’s modulus near 70 GPa and a tensile strength around 1280 MPa, is an obvious candidate, but smooth fiber surfaces bond poorly to an inorganic matrix. The fibers were therefore sized at 450 °C for 3 h, etched in 1 mol/L HCl for 30 min to remove inorganic residue and roughen the surface, rinsed, dried at 120 °C, and treated for 30 min in a solution of 3 wt% KH550 (3-aminopropyltriethoxysilane) in 95:5 ethanol–water adjusted to pH 4–6 with HCl, then cured at 120 °C. An L9(34) orthogonal array varied fiber length (3, 6, 10 mm), fiber content (0.2, 0.4, 0.6 wt%) and KH550 curing time (3, 6, 9 h); specimens were cast with a water-to-silica-sol ratio of 4:1 and calcined on the 2 °C/min schedule.
| Run | Combination | Flexural (MPa) | Compressive (MPa) |
|---|---|---|---|
| 1 | A1B1C1 (3 mm, 0.2%, 3 h) | 0.55 | 1.57 |
| 2 | A1B2C2 (3 mm, 0.4%, 6 h) | 0.58 | 1.79 |
| 3 | A1B3C3 (3 mm, 0.6%, 9 h) | 0.90 | 1.70 |
| 4 | A2B1C2 (6 mm, 0.2%, 6 h) | 0.63 | 1.85 |
| 5 | A2B2C3 (6 mm, 0.4%, 9 h) | 0.65 | 2.04 |
| 6 | A2B3C1 (6 mm, 0.6%, 3 h) | 1.25 | 2.00 |
| 7 | A3B1C3 (10 mm, 0.2%, 9 h) | 0.67 | 1.58 |
| 8 | A3B2C1 (10 mm, 0.4%, 3 h) | 0.95 | 2.00 |
| 9 | A3B3C2 (10 mm, 0.6%, 6 h) | 1.84 | 1.91 |
Range analysis ranks fiber content first for flexural strength (range 0.71 MPa), length second (0.47 MPa) and curing time third (0.28 MPa); for compressive strength, length and content tie at 0.27 MPa and curing time again contributes least (0.09 MPa). The best flexural result, 1.84 MPa at 10 mm and 0.6%, is 178.8% above the unreinforced mold; the best compressive result, 2.04 MPa at 6 mm and 0.4%, is 25.9% above it. Fiber reinforcement is thus far more effective against bending than against compression, as expected for a crack-bridging mechanism.
The results also expose a threshold effect that matters for practical dosing. At 3 mm and 6 mm with 0.2–0.4% fiber, flexural strength fell below the unreinforced value of 0.66 MPa. At those loadings the fibers are too few and too far apart to bridge cracks; they act instead as discrete defects that sever the matrix. Once either the length or the content is increased sufficiently — 10 mm at any tested content, or 0.6% at any tested length — the reinforcement becomes positive, since a propagating crack must either debond and pull out fibers or consume energy circumventing them. Fractography confirms the mechanism: fibers protruding from fracture surfaces are longer for 10 mm than for 6 mm stock, and the matrix remains bonded to the fiber surfaces, indicating good interfacial contact rather than pull-out from a debonded sleeve. At the same time, fibers peel away from the matrix instead of fracturing, which shows that interfacial shear strength remains below the fiber’s own strength — the interface, not the fiber, is the limiting element.
The coupling agent contributes less than expected, and the reason is thermal. SEM shows KH550 covering the fiber surface as irregular nodules that roughen it and improve mechanical keying with the slurry. But KH550 boils at 217 °C, well below the calcination temperature, and after firing the surface is measurably smoother. The silane’s organophilic function therefore has no role to play in an all-inorganic system, and its residence time before firing — the curing time variable — has little influence on the fired strength. Surface treatment is still worth performing for green strength and handling, but it cannot be treated as a primary design variable.
Thermal crack resistance and the limits of reuse
Thermal crack resistance was assessed on L-shaped specimens fired on the standard schedule, held at 300 °C for 30 min and quenched to room temperature, then held at 700 °C for 5 min and quenched again. All three formulations tested survived the 300 °C cycle without visible cracking. At 700 °C the differences were pronounced: the 10 mm, 0.4% mold developed cracks from several origins that linked into a network covering the surface; the 6 mm, 0.6% mold developed a single crack running across the surface, wide at initiation and narrowing as it propagated, which is the signature of energy dissipation by fibers; and the 10 mm, 0.6% mold showed only a few short, narrow cracks. The combination of long fiber and high loading is therefore the most effective against thermal shock, consistent with its flexural ranking.
Recycling is the weak point of the technology. When calcined mold material was ground, screened to 60 mesh and blended into fresh powder at 10–50 wt%, the slurry lost fluidity at only 10% addition; raising the water-to-gypsum ratio to 0.67 restored workability up to 40% but not at 50%. Strength fell sharply across the 0–40% range examined, with flexural strength between 0.45 and 0.75 MPa and compressive strength between 0.57 and 1.5 MPa. The cause is mineralogical: during calcination the dihydrate converts through soluble anhydrite to insoluble anhydrite, whose water channels are destroyed, so the reclaimed powder has essentially no cementing capacity. Reuse therefore trades directly against both water demand and strength, and the material is better regarded as a filler-grade by-product than as a recyclable binder.
A working formulation and its operating envelope
Combining the three lines of work gives a formulation in which each component performs a specific function: α-hemihydrate 35%, quartz 8%, mullite 27%, coal gangue 20% and diatomite 10%, with 10 mm glass fiber at 0.6 wt% and a water-to-silica-sol ratio of 4:1. In this configuration the slurry spreads to 90 mm and sets in 7 min initially and 17 min finally; the green mold reaches 2.02 MPa flexural and 2.30 MPa compressive strength; after calcination at 2 °C/min it retains 0.458 MPa flexural and 1.07 MPa compressive strength with 0.320% linear shrinkage; and with the fiber addition the fired flexural strength rises to 1.84 MPa and compressive strength to 1.91 MPa, with the best thermal crack resistance among the formulations tested.
The trade-offs are explicit. The 90 mm spread is at the low end of what is pourable, and the porous fillers that deliver strength and dimensional stability are precisely what consume it; a silica sol addition or a superplasticizer at 0.1–0.4% recovers fluidity but at the cost of retardation, and the polycarboxylate option in particular pushes final set past three hours at its strength optimum. Boron-containing additions raise fired strength substantially without touching rheology, but require loadings of several percent. Fiber length and content must be chosen together to stay above the percolation threshold at which reinforcement turns positive, and the silane treatment should be justified by green-strength requirements rather than by fired properties. Within those constraints, the filler–additive–fiber framework described here provides a route to gypsum investment molds that combine sub-0.5% shrinkage with sufficient strength and thermal shock resistance for thin-walled, dimensionally critical castings.