Investment casting is often the preferred route for aerospace casings because it can produce integral, thin-walled geometries with complex flow surfaces and local mounting features. The wax pattern is the dimensional master of that route: its shrinkage, warpage, and joint mismatch are transferred to the ceramic shell and then to the metal casting. For a large casing, a pattern that is out of tolerance cannot be corrected by downstream casting adjustments. Conventional wax injection requires a split metal die and a demolding strategy; for a 620 mm class casing with freeform surfaces, undercuts, and many bosses, the die becomes complex and the injection-warpage problem becomes difficult to control.
Selective laser sintering (SLS) builds the pattern directly from a polymer powder, so tooling cost is decoupled from geometric complexity. The process is not, however, dimensionally neutral. Each layer is heated locally by a moving laser, cools against a powder bed and previously sintered material, and contracts under constraints that vary across the part. For large, thin-walled investment-casting patterns, the limiting factor is usually not laser resolution but thermal distortion.
The case examined here is a Boeing compressor casing with an outer diameter of 620.4 mm and a height of 438.64 mm. The wall thins to 5.95 mm, and the pattern is split into left and right half casings joined by only ten small cylindrical connectors. Those open gaps and the low stiffness of the half-casings make the pattern sensitive to nonuniform cooling. The pattern material is polystyrene (PS) powder, a common SLS material for investment-casting patterns because it can be burned out cleanly and sintered at relatively low temperatures.
Process window and material response
The builds used an HK S500 SLS system with a 30 W CO2 laser and a 320 mm × 320 mm × 450 mm build envelope. The PS powder had a mean particle size of 35 μm and a density of 1.03 g/cm3. Its reported sintering range was 92–102 °C, its glass-transition temperature was 102 °C, and its thermal conductivity was 0.08 W/(m·K). The low thermal conductivity is important: heat introduced by the laser is not rapidly redistributed, so the top of a layer cools faster than the material below it, while the previously sintered solid constrains contraction in the plane of the layer.
| Property | Value |
|---|---|
| Density | 1.03 g/cm3 |
| Mean particle size | 35 μm |
| Sintering temperature | 92–102 °C |
| Preheat temperature | 90–95 °C |
| Glass-transition temperature | 102 °C |
| Thermal conductivity | 0.08 W/(m·K) |
| Specific heat | 1.3 kJ/(kg·K) |
| Parameter | Value |
|---|---|
| Laser power | 17 W |
| Scan speed | 4000 mm/s (4 m/s) |
| Layer thickness | 0.2 mm |
| Hatch spacing | 0.2 mm |
| Initial preheat setpoint | 110 °C |
| Processing temperature | 90 °C |
| Preheat time | 10 min |
| Laser spot diameter (simulation) | 0.2 mm |
The same power, scan speed, layer thickness, and hatch spacing were held constant in the feature trials and in the numerical model. This is important because the comparison is not a search for an optimum process window; it is an assessment of how a fixed, workable window performs when the geometry becomes large and structurally uneven.
Feature-level dimensional response
To isolate the influence of geometry, six representative feature families were built: thin cylinders, draft-angle cylinders, a segmented cylinder, rings, circular bosses, and square bosses. Simple features were measured with a vernier caliper; the full pattern required three-dimensional scanning. The feature set was chosen to reproduce the local geometries of the casing pattern, including thin walls, draft surfaces, open gaps, and local bosses.
| Feature | Nominal geometry | Measured response | Interpretation |
|---|---|---|---|
| Thin cylinder | Inner diameter 236 mm; wall 5 mm; height 20–50 mm | Inner diameter 235.40–235.55 mm; wall 4.84–4.96 mm; height 19.86–49.88 mm | Height shrinkage 0.12–0.14 mm, nearly independent of height; radial shrinkage small but systematic |
| Draft-angle cylinder | Inner diameter 100 mm; wall 10 mm; height 30 mm; draft 0–30° | Angular error 0–0.051°, relative error ≤0.17%; inner diameter 99.65–99.80 mm; wall 9.71–9.80 mm; height 29.82–29.86 mm | High angular fidelity; draft surfaces can be reproduced within the process window |
| Segmented cylinder | Inner diameter 100 mm; wall 10 mm; height 30 mm; one open gap | X-direction diameter 100.15 mm; Y-direction diameter 99.60 mm; 45° diameter 99.66 mm; closed reference 99.67, 99.67, and 99.65 mm | The gap removes constraint and produces nonuniform radial shrinkage |
| Ring | Inner diameter 270 mm; height 7.5 mm; width 14–20 mm | Inner diameter 269.40–269.53 mm; height 7.52–7.56 mm; width 13.82–19.62 mm | Width variation has no clear effect on inner diameter or height; width shrinkage is irregular |
| Circular boss | Outer diameter 22 mm; height 30–36 mm | Diameter 21.68–21.80 mm; bottom-edge Z offset 3.30–4.50 mm | Warpage increases with boss height; a 22 mm × 30 mm boss is expected to warp about 3.3 mm |
| Square boss | 50 mm × 30 mm; height 6–12 mm | Length 49.91–49.92 mm; width 29.78–29.83 mm; height 5.83–11.80 mm | Cooling shrinkage dominates; height shrinkage 0.17–0.20 mm is only weakly dependent on height |
Two results stand out. First, the draft-angle features are accurate. At a 30° draft, the absolute angular error was 0.051° and the relative error was 0.17%; at 10° and 20°, the relative error was about 0.05%. This indicates that the process window can reproduce inclined and freeform surfaces with good angular fidelity. The limiting distortion in the casing pattern is therefore not the laser’s ability to follow a contour, but the thermal and mechanical response of a thin, open structure.
Second, the segmented cylinder shows how strongly an opening changes the shrinkage field. The X-direction diameter increased to 100.15 mm, while the Y-direction diameter contracted to 99.60 mm; the closed reference cylinder measured about 99.67 mm in both directions. The opening relieves residual stress near the gap and allows the two sides to contract differently. The same mechanism appears in the full casing pattern, which is split into half-casings and joined by only ten small cylinders. That geometry is a direct predictor of a noncircular pattern.
The thin cylinders and square bosses show a different pattern. Their dimensions were consistently below nominal, indicating that cooling shrinkage dominates. Height shrinkage was almost constant as height increased: about 0.12–0.14 mm for the thin cylinders and 0.17–0.20 mm for the square bosses. Newly spread powder compensates the vertical shrinkage of earlier layers, so the measured top surface does not accumulate the full free-shrinkage strain. The circular bosses behave differently: their local Z offset increased from 3.30 mm at 30 mm height to 4.50 mm at 36 mm height. For the 22 mm × 30 mm boss used in the casing, the expected warp is therefore about 3.3 mm, a large fraction of the boss diameter.
Full-pattern trial and measured distortion
The full wax pattern was built with the same laser, scan, and layer settings. To reduce local warpage at the circular bosses, an additional high-temperature layer was introduced before each boss layer, intended to slow local cooling and reduce the thermal gradient. The finished pattern was measured with a PowerScan-Pro 15M noncontact scanner, which had a point spacing of 0.051–0.104 mm and a measurement accuracy of ±0.008 mm.
The scan gave an average positive dimensional deviation of 1.999 mm and an average negative deviation of −1.245 mm. The circular boss region and the bottom plane showed visible warpage, and the pattern was rejected. These averages should not be read as a complete tolerance assessment: they combine global shrinkage with local distortion, and the local flatness and roundness errors are more severe than a single mean value suggests. The result does show that a process window validated on small features is not sufficient for a large, thin-walled pattern with open gaps.
Numerical simulation and validation
The SLS process was modeled in Simufact Additive using the PS properties and the process settings above. The model used a 0.2 mm laser spot, 4 m/s scan speed, 17 W power, and 0.2 mm layer and hatch spacing. The simulated deformation field was compared with the three-dimensional scan at ten corresponding points.
| Point | Simulation (mm) | Measurement (mm) | Absolute error (mm) | Error rate (%) |
|---|---|---|---|---|
| 1 | 3.18 | 3.05 | 0.13 | 4.26 |
| 2 | 3.58 | 3.49 | 0.09 | 2.58 |
| 3 | 3.85 | 3.81 | 0.04 | 1.05 |
| 4 | 3.53 | 3.47 | 0.06 | 1.73 |
| 5 | 3.24 | 3.23 | 0.01 | 0.31 |
| 6 | 3.38 | 3.36 | 0.02 | 0.60 |
| 7 | 3.04 | 3.06 | 0.02 | 0.65 |
| 8 | 3.38 | 3.32 | 0.06 | 1.81 |
| 9 | 3.12 | 3.15 | 0.03 | 0.95 |
| 10 | 3.41 | 3.43 | 0.02 | 0.58 |
The mean error rate across the ten points was 1.45%; the maximum was 4.26% and the minimum was 0.31%. This level of agreement is adequate for ranking design variants and for identifying the dominant distortion modes. It does not eliminate the need for dimensional scanning of a production pattern, and it should not be interpreted as a general validation of the material model outside this geometry and process window. In particular, the bottom-plane warpage observed in the trial was not well reproduced by the simulation and remains an open modeling issue.
Distortion mechanisms
Using the validated model, three failure modes were examined: loss of cylindricity, mounting-face flatness, and local circular-boss warpage. They share a common origin in constrained cooling, but they appear at different length scales.
Cylindricity and the open-gap effect
Radial deviation was evaluated on quarter-sections of the inner arc at three heights. For a point with simulated coordinates \((X_1,Y_1)\) and deviations \(T_x,T_y\) from the nominal coordinates, the radial change was calculated as
$$\Delta r=\sqrt{X_1^2+Y_1^2}-\sqrt{(X_1-T_x)^2+(Y_1-T_y)^2}$$
where the origin is the arc center. The range \(\max(\Delta r)-\min(\Delta r)\) on a quarter-section is used here as the cylindricity index; it is an indicator of out-of-roundness rather than a full ISO 1101 evaluation over the complete cylindrical surface.
| Section | Minimum Δr (mm) | Maximum Δr (mm) | Cylindricity index (mm) |
|---|---|---|---|
| Z = 80 mm | −0.82 | −0.43 | 0.39 |
| Z = 230 mm | −1.21 | −0.68 | 0.53 |
| Z = 413.6 mm | −0.85 | 0.60 | 1.45 |
At Z = 413.6 mm, the radial deviation ranged from −0.85 mm to +0.60 mm, giving a cylindricity index of 1.45 mm. At Z = 230 mm and Z = 80 mm, the index fell to 0.53 mm and 0.39 mm. In all three sections, the maximum Δr occurred at X = 0, the gap between the half-casings. Near the gap, residual stress is relieved and cooling contraction is reduced, so the radius remains relatively large. Away from the gap, the material is more constrained and contracts more. A circular section therefore becomes oval, with its long axis at the gap.
The section at Z = 413.6 mm is more distorted than the lower sections because the ring is wider there and the gap-related Y-direction deviation is larger. The open gap is not a minor geometric detail; it sets the radial shrinkage pattern of the entire casing.
Mounting-face flatness
The pattern was sectioned along the YZ plane to expose the mounting face of one half-casing. Because that face is nominally parallel to the YZ plane, its X-direction deviation is a direct measure of local flatness. The maximum X deviation was +0.136 mm and the minimum was −0.449 mm, a range of 0.585 mm. The topmost point, which is farthest from the ten small connectors, showed the largest negative deviation. The connectors constrain the rest of the face, so its deviation stays closer to zero.
After the connectors are cut, the mounting face becomes an assembly datum. A flatness error of 0.585 mm is therefore significant for an aerospace casing, even though it is small compared with the overall pattern size. The result also explains why a process window validated on free-standing features cannot guarantee the accuracy of a constrained assembly surface.
Circular-boss warpage
The simulated local Z deviation at the circular bosses reached 3.5 mm. On a horizontal section, the deviation increased from the inner end to the outer end, and the slope increased outward. On a vertical section, the deviation decreased from the bottom to the top. The boss therefore warps most at its outer, lower region.
The mechanism is a combination of thermal gradient and mechanical constraint. The upper surface of a growing layer is exposed to air, while the lower surface is in contact with powder; the upper material cools and contracts faster. Within the same layer, the inner side of the boss is close to the already sintered pattern and is more constrained, while the outer side borders loose powder and is less constrained. The combination produces an outward-increasing warp. In the vertical direction, newly spread powder compensates part of the Z deformation of earlier layers, so the lower part accumulates a larger deviation than the upper part.
Structural compensation
The distortion modes point to two design levers: close the open gaps to make the shrinkage field more uniform, and stiffen the local boss regions that warp most. Both were evaluated by simulation before any further physical build.
Fully connected half-casings
The first scheme replaces the ten small connectors with full tie-bar connections, so the left and right half-casings form a closed ring rather than two weakly linked arcs. The open gap no longer exists, and the radial shrinkage is forced to be more uniform around the circumference.
| Section | Original Δr range (mm) | Original cylindricity (mm) | Fully connected Δr range (mm) | Fully connected cylindricity (mm) |
|---|---|---|---|---|
| Z = 80 mm | −0.82 to −0.43 | 0.39 | −0.93 to −0.66 | 0.27 |
| Z = 230 mm | −1.21 to −0.68 | 0.53 | −1.04 to −0.87 | 0.17 |
| Z = 413.6 mm | −0.85 to 0.60 | 1.45 | −0.66 to −0.61 | 0.05 |
At Z = 413.6 mm, the cylindricity index fell from 1.45 mm to 0.05 mm; at Z = 230 mm, from 0.53 mm to 0.17 mm; and at Z = 80 mm, from 0.39 mm to 0.27 mm. The largest improvement occurred where the original gap was widest. The closed ring also provides a stiffer datum that can be machined after building, so the mounting-face flatness can be controlled by a finishing operation rather than by as-built tolerance. The trade-off is additional material and a machining step, and the closed geometry must remain compatible with shell building and pattern removal.
Local support ribs at circular bosses
The second scheme adds a 5 mm × 5 mm support rib from the outer lower end of a circular boss to the casing body. A simplified coupon contained four 22 mm × 30 mm bosses: one unsupported and three supported at 20°, 30°, and 40° to the horizontal. The coupon was made hollow in the center to reduce powder consumption and simulation time.
All three ribs reduced the Z deviation in both the horizontal and vertical sections. The ranking was consistent: the unsupported boss had the largest deviation, followed by the 20°, 30°, and 40° cases, with the 40° rib giving the smallest deviation. The ribs did not reverse the basic gradients—the outer end still warped more than the inner end, and the lower part still warped more than the upper part—but they reduced the slope and the maximum deviation. The 40° rib was the most effective of the three because it restrained the outer lower region where the warpage was largest.
For the full pattern, the practical limitation is the number of bosses. Adding and removing hundreds of ribs would be labor-intensive and would complicate post-processing. The scheme is therefore most suitable for critical bosses or for a design in which the ribs can be integrated into the pattern and removed in a controlled finishing step.
Engineering implications for investment-casting patterns
The case shows a clear split between what SLS handles well and what it does not. Angular fidelity is good: draft surfaces were reproduced with a relative angular error of at most 0.17%, and simple cylindrical features held their dimensions within a few tenths of a millimetre. Large, thin-walled, open structures are different. Their accuracy is governed by constraint-induced shrinkage, not by the laser’s contouring ability. A process window that works for a free-standing boss or cylinder can still produce an unacceptable casing pattern when the same settings are applied to a split, low-stiffness ring.
Three design rules follow from the results. First, avoid open gaps in a large pattern unless they are structurally compensated; a closed ring or a well-designed tie-bar layout gives a more uniform shrinkage field. Second, place local stiffeners where the thermal gradient and the loss of constraint coincide, especially at the outer lower region of tall bosses. Third, use simulation to rank design variants, but validate the final pattern with three-dimensional scanning; the mean error rate of 1.45% is useful for comparison, not for tolerance sign-off.
The scope of the results is defined by the material and process: PS powder, 17 W laser power, 4 m/s scan speed, 0.2 mm layer thickness, and 0.2 mm hatch spacing. Different powders, layer thicknesses, or preheat strategies will change the shrinkage and warpage response. The case also stops at the wax pattern. Transfer to the ceramic shell and the metal casting, including shell cracking risk and dimensional inheritance, remains a separate qualification step.
For investment casting, the economic value of SLS is highest when the pattern is geometrically complex and the production volume is low. The cost then shifts from metal tooling to build time, powder, simulation, and post-processing. The casing example shows that this shift is worthwhile only when the pattern is designed for the thermal physics of layerwise sintering, not merely for the geometry of the final casting.