I investigated a complex thin-walled stainless steel bearing seat used in food-processing equipment. The part is produced by investment casting, but the factory faced repeated defects such as shrinkage porosity, shrinkage cavities, misruns, and scabs. Because the wall thickness is small in several regions and the geometry combines a cylindrical tube with a rectangular frame, isolated liquid regions form during solidification and cannot be fed properly. I therefore combined numerical simulation with intelligent optimization algorithms. Although my primary process is investment casting, I also studied useful lessons from evaporative pattern casting, because evaporative pattern casting offers complementary insight into thin-wall filling behavior, shell strength, and defect formation. In evaporative pattern casting, the pattern is vaporized by the molten metal, while in investment casting the pattern is removed before pouring. Both processes must control filling stability, thermal gradients, and gas escape. I used this comparison to guide my decisions.
My objective was not only to reduce shrinkage porosity and equivalent stress but also to raise process yield and product quality. I first analyzed the part structure, determined the pouring position and ingate locations, and designed five gating system schemes. I then simulated filling and solidification in ProCAST. After comparing simulation results with trial production, I selected the best gating system. Next, I designed an orthogonal experiment with four process parameters: pouring temperature, pouring speed, shell preheating temperature, and shell thickness. I used the orthogonal results as sample data to build a BP neural network, optimized it with particle swarm optimization, and then used a genetic algorithm for multi-objective optimization. Finally, I validated the optimized parameters through simulation and production trials. The final castings showed almost no visible shrinkage porosity or scabs, and internal inspection confirmed that the defects were eliminated. Throughout this work, I repeatedly returned to evaporative pattern casting as a related precision-casting route that shares many thermal and filling challenges.
The material was 1.4308 stainless steel, a German designation equivalent to GX5CrNi19-10. It has good corrosion resistance, high-temperature performance, and satisfactory machinability. Its chemical composition is summarized in Table 1. The liquidus temperature is approximately 1485 °C, and the solidus temperature is approximately 1386 °C. These values are important because the pouring temperature must be sufficiently above the liquidus to avoid misruns in thin walls, but excessively high superheat increases shrinkage porosity. I used this balance as a central theme in both investment casting and evaporative pattern casting process design.
| Element | C | Si | Mn | P | S | Cr | Ni | Fe |
|---|---|---|---|---|---|---|---|---|
| Composition (wt.%) | ≤0.07 | ≤1.50 | ≤1.50 | 0.040 | 0.030 | 18.0–20.0 | 8.00–11.0 | Balance |
The wax pattern material was a recycled medium-temperature wax. Medium-temperature wax is preferred for investment casting because it has good fluidity, low shrinkage, and acceptable surface reproduction. The refractory system used zircon sand for the face coat and mullite sand for the transition and backup coats. The binder was silica sol. Table 2 and Table 3 summarize the main technical indices of the wax and refractory materials. These properties influence shell permeability, thermal expansion, and heat transfer, all of which also matter in evaporative pattern casting where the pattern decomposition products must escape through the coating.
| Property | Softening point | Drop point | Linear shrinkage | Needle penetration | Ash | Color |
|---|---|---|---|---|---|---|
| Value | 79.40–85.00 °C | 87.70–93.30 °C | 0.90–1.00% | 20–30 D | ≤0.02% | Yellow-brown |
| Material | Chemical nature | Melting point | Refractoriness | Density | Thermal expansion |
|---|---|---|---|---|---|
| Zircon sand | Acidic | 2750 °C | >1825 °C | 4.5–4.9 g/cm³ | 4.6 × 10⁻⁶ /°C |
| Mullite sand | Neutral | 1810 °C | >1700 °C | 3.16 g/cm³ | 4.5 × 10⁻⁶ /°C |
The bearing seat has a maximum outline of approximately 245 mm × 214 mm. The cylindrical tube is about 245 mm long. The as-cast wall thickness is generally 5–6 mm, and after machining the two ends become about 3 mm. One protruding thin plate on the rectangular frame side has a thickness of about 2.4 mm and a protrusion length of about 31.6 mm. The inner rectangular cutout has a thinnest wall of about 1.8 mm. The rectangular frame outline is about 196.6 mm × 177.6 mm. The inner recess of the frame has a wall thickness of about 3 mm, and two side walls have an included angle of about 2°. A reinforcing rib connects the tube and frame, with a thickness of about 2.5 mm and an included angle of about 2°. The total mass is about 2.2 kg. These dimensions make the part a complex thin-walled casting. I recognized that evaporative pattern casting would face similar challenges in controlling thin-wall filling and dimensional distortion, even though its pattern removal mechanism differs.
The first step in my process design was to determine the pouring position. In investment casting, the pouring position controls filling pressure, feeding direction, and the location of the final solidification zone. I considered four principal positions: vertical tube, horizontal tube with the thin plate on the upper side, horizontal tube with the thin plate on the lower side, and horizontal tube with the thin plate inclined. I then combined these positions with ingate locations on the cylindrical wall and on one side of the rectangular frame. Table 4 summarizes the casting position alternatives and their main characteristics. I also noted that in evaporative pattern casting, the pattern orientation relative to gravity strongly affects gas escape and metal front stability, so the same careful orientation logic applies.
| Scheme | Pouring position | Main advantage | Main risk |
|---|---|---|---|
| Position A | Tube vertical, plate at one side | Good directional solidification | Tall mold, higher shell stress |
| Position B | Tube horizontal, plate on upper side | Stable filling from bottom | Upper plate may trap gas |
| Position C | Tube horizontal, plate on lower side | Good surface quality on large plate | Lower plate may cool too fast |
| Position D | Tube horizontal, plate inclined | Slower metal rise, less turbulence | Sand inclusion risk, shell leakage |
I used the following pouring time and pouring speed relations during the initial design. The pouring time was estimated from the part mass and minimum wall thickness, and the minimum pouring speed was estimated from the thin-wall length and minimum wall thickness. These equations helped me set the first simulation parameters before optimization. The same logic can be adapted to evaporative pattern casting, where filling must also avoid excessive front velocity and pattern decomposition defects.
$$ t_J = S \sqrt{G_{\text{part}}} $$
$$ v_J = K \frac{L}{\delta_{\min}} $$
Here, \(t_J\) is the pouring time, \(S\) is an empirical coefficient, \(G_{\text{part}}\) is the mass of the poured metal, \(v_J\) is the pouring speed, \(K\) is a gating coefficient, \(L\) is the maximum length of the thin wall, and \(\delta_{\min}\) is the minimum wall thickness. For the initial design, I selected a pouring temperature of 1620 °C, a shell preheating temperature of 1130 °C, a shell thickness of 6 mm, and a pouring speed of about 2 kg·s⁻¹.
I designed five gating system schemes. Scheme 1 used a single sprue with a combined top-and-side ingate and the tube horizontal with the plate on the lower side. Scheme 2 used a stepped gating system with the tube vertical. Scheme 3 used an inclined pouring system. Scheme 4 used a combined gating system with the tube horizontal and the plate on the upper side. Scheme 5 used a combined stepped gating system with four parts per tree. Table 5 lists the schemes, tree mass, yield, and distinguishing features. Scheme 5 gave the highest process yield, about 48.9%, because the four-part tree reduced the relative weight of the gating system. This yield advantage is important in investment casting, just as tree design is important in evaporative pattern casting for reducing scrap and energy consumption.
| Scheme | Description | Tree mass | Yield | Key feature |
|---|---|---|---|---|
| 1 | Single sprue, combined top-and-side ingate, plate lower | 9.9 kg | 44.4% | Large plate at lower position |
| 2 | Multiple sprues, stepped ingates, tube vertical | 11.6 kg | 37.9% | Stable filling, more metal consumption |
| 3 | Single sprue, inclined pouring, plate inclined | 10.2 kg | 43.1% | Slow metal rise, possible inclusion |
| 4 | Single sprue, combined side ingates, plate upper | 10.4 kg | 42.3% | Bottom-first filling |
| 5 | Single sprue, multiple ingates, four parts per tree | 18.0 kg | 48.9% | High yield, combined stepped filling |
For numerical simulation, I imported the three-dimensional models into ProCAST after converting them to a neutral format. I used a symmetric half-model to reduce computation time. I repaired any missing surfaces, assembled contacting faces, and generated triangular surface meshes. The gating system was meshed with a minimum element size of about 4 mm, while the bearing seat was meshed with a minimum element size of about 2 mm. A shell of about 6 mm was generated around the casting. I then created tetrahedral volume meshes and checked them for inverted elements. The material database was set for 1.4308 stainless steel and mullite refractory. The interface heat transfer coefficient between metal and shell was set to 500 W·m⁻²·K⁻¹. The gravity direction was aligned with the sprue axis, and the inlet was placed at the pouring cup. The initial metal temperature was 1620 °C, and the initial shell temperature was 1130 °C. These settings were used for all five schemes so that the comparison would be consistent.
The governing equations for filling and solidification include mass conservation, momentum conservation, and energy conservation. I used these equations to interpret the simulation results. The continuity equation is:
$$ \frac{\partial \rho}{\partial t} + \nabla \cdot (\rho \mathbf{u}) = 0 $$
The momentum equation for a viscous incompressible fluid is:
$$ \rho \left( \frac{\partial \mathbf{u}}{\partial t} + \mathbf{u} \cdot \nabla \mathbf{u} \right) = -\nabla p + \mu \nabla^2 \mathbf{u} + \rho \mathbf{g} $$
The energy equation with latent heat release is:
$$ \rho c_p \left( \frac{\partial T}{\partial t} + \mathbf{u} \cdot \nabla T \right) = \nabla \cdot (k \nabla T) + \rho L \frac{\partial f_s}{\partial t} $$
During solidification, heat is transferred by conduction, convection, and radiation. The corresponding expressions are:
$$ q = -k \frac{\partial T}{\partial x} $$
$$ q = h (T_f – T_w) $$
$$ q = \varepsilon \sigma_0 T_s^4 $$
Shrinkage porosity and shrinkage cavities form when liquid and solidification shrinkage cannot be compensated by feed metal. The total shrinkage volume can be represented as:
$$ V_{\text{total}} = V_{\text{cavity}} + V_{\text{porosity}} $$
The Niyama criterion was used to predict porosity:
$$ \frac{G}{\sqrt{R}} \lt K $$
where \(G\) is the local temperature gradient, \(R\) is the cooling rate, and \(K\) is a critical value. In my simulations, the Niyama criterion and the critical solid fraction method were both used to identify isolated liquid regions. I found that evaporative pattern casting also relies on similar thermal-gradient reasoning: if the metal front cools too quickly or if the coating cannot vent decomposition gases, defects appear even when the gating geometry seems acceptable.
Scheme 1 filled the part in about 5.73 s. The metal entered through the side ingates, and the lower thin plate filled after the cylindrical wall had partially filled. The filling front was relatively stable, but some air entrapment occurred in the bottom runner because of the initial impact. During solidification, the lower thin wall cooled first, followed by the reinforcing rib intersections and the inner cutout walls. Isolated liquid regions formed at the frame-to-tube junction. The predicted shrinkage porosity volume was 0.71 cc, with 14 defect locations. Scheme 2 filled in about 5.11 s. The double sprue reduced impact, but the upper part of the frame required additional feeding from the tube-side ingate. The predicted shrinkage volume was 0.76 cc, with 16 defect locations. Scheme 3 filled in about 4.49 s. The inclined position slowed the metal rise, but it also caused metal splashing in the bottom runner and a severe scab problem in production. The predicted shrinkage volume was 0.76 cc, with 15 defect locations. Scheme 4 filled in about 4.90 s. The bottom-first filling was stable, but the upper plate cooled quickly, and the predicted shrinkage volume was 0.84 cc, with 12 defect locations. Scheme 5 filled in about 6.14 s. The four-part tree produced a longer filling time, but the filling sequence was more gradual, and the predicted shrinkage volume was 0.76 cc, with 14 defect locations. Table 6 summarizes the filling and defect predictions.
| Scheme | Filling time (s) | Predicted shrinkage volume (cc) | Number of defect locations | Observed filling quality |
|---|---|---|---|---|
| 1 | 5.73 | 0.71 | 14 | Stable, minor air entrapment |
| 2 | 5.11 | 0.76 | 16 | Stable, longer upper filling path |
| 3 | 4.49 | 0.76 | 15 | Splashing and scab risk |
| 4 | 4.90 | 0.84 | 12 | Bottom-first, upper plate cooling |
| 5 | 6.14 | 0.76 | 14 | Gradual, good yield |
I then performed trial production for all five schemes. The wax patterns were injected on a vertical wax injection machine. The wax temperature was about 65 °C, and the injection time was about 35 s. After cooling, I removed flash and assembled the patterns with sprues and runners. The shells were made by repeated slurry dipping and sanding. The face coat used zircon sand, and the backup coats used mullite sand. Each coat was dried under controlled temperature and humidity. After sealing, the shells were dewaxed in an autoclave and then dried. Before pouring, the shells were fired at about 1130 °C for more than 1.5 h. The metal was melted in a medium-frequency induction furnace, and the composition was checked by optical emission spectrometry. The measured composition was within the specification. The pouring temperature was about 1620 °C. After cooling, the castings were knocked out, cut off, shot-blasted, and pickled.
Table 7 presents the observed macro defects from the trial production. Scheme 1 produced two parts with visible shrinkage cavities, and five scab locations. Scheme 2 produced three shrinkage cavity locations and two scab locations. Scheme 3 produced only one shrinkage cavity location but ten scab locations, mainly along the root of the large thin plate. Scheme 4 produced two shrinkage cavity locations and no scabs. Scheme 5 produced only one shrinkage cavity location and five scab locations across four parts. The scab problem in Scheme 3 was associated with shell leakage and uneven face-coat coverage. I also observed that the inclined position made the lower large flat surface prone to sand inclusion. This result is relevant to evaporative pattern casting because coating permeability and pattern decomposition residue can produce similar surface defects when the pattern orientation is unfavorable.
| Scheme | Tree type | Shrinkage cavity locations | Scab locations | Parts inspected |
|---|---|---|---|---|
| 1 | Two parts per tree | 2 | 5 | 2 |
| 2 | Two parts per tree | 3 | 2 | 2 |
| 3 | Two parts per tree | 1 | 10 | 2 |
| 4 | Two parts per tree | 2 | 0 | 2 |
| 5 | Four parts per tree | 1 | 5 | 4 |
I used X-ray inspection and destructive sectioning to verify internal defects. X-ray inspection showed no clear internal shrinkage in some sections because the cylindrical and frame features overlap, which can hide defects in a two-dimensional projection. I therefore cut the parts along a selected path using wire electrical discharge machining. The cut sections confirmed that the shrinkage porosity was concentrated at the frame-to-tube junction and at rib intersections, in agreement with the simulation. Based on the combination of simulation, macro inspection, X-ray, and sectioning, I selected Scheme 5 as the gating system for further process parameter optimization. Scheme 5 offered the best balance of yield, filling stability, and defect distribution.
After fixing the gating system, I optimized the investment casting process parameters. I selected pouring temperature, pouring speed, shell preheating temperature, and shell thickness as the four factors. Each factor was assigned five levels. Table 8 gives the factor-level table. I used an L25 orthogonal array, which required 25 simulation runs. The response variables were shrinkage porosity volume and equivalent stress. The orthogonal array reduced the number of experiments while still allowing me to estimate the main effects. I also considered evaporative pattern casting parameter studies, which often examine pouring temperature, coating thickness, and sand permeability in a similar orthogonal manner.
| Factor | Level 1 | Level 2 | Level 3 | Level 4 | Level 5 |
|---|---|---|---|---|---|
| Pouring temperature (°C) | 1600 | 1610 | 1620 | 1630 | 1640 |
| Pouring speed (kg·s⁻¹) | 1.5 | 1.75 | 2.0 | 2.25 | 2.5 |
| Shell preheating temperature (°C) | 1110 | 1120 | 1130 | 1140 | 1150 |
| Shell thickness (mm) | 5.0 | 5.5 | 6.0 | 6.5 | 7.0 |
Table 9 lists the orthogonal experiment results. I used these results to calculate the range \(R\) for each factor. For shrinkage porosity volume, the influence order was shell thickness > pouring temperature > shell preheating temperature > pouring speed. For equivalent stress, the influence order was shell thickness > shell preheating temperature > pouring temperature > pouring speed. Shell thickness had the strongest effect on both responses. A thicker shell reduces cooling rate and can improve feeding in some regions, but it also increases thermal resistance and may cause coarse microstructure. A thinner shell improves permeability and cooling uniformity, which is also a key concern in evaporative pattern casting, where coating thickness controls gas escape and metal filling.
| Run | Pouring temperature (°C) | Pouring speed (kg·s⁻¹) | Shell preheating temperature (°C) | Shell thickness (mm) | Shrinkage porosity (cc) | Equivalent stress (MPa) |
|---|---|---|---|---|---|---|
| 1 | 1600 | 1.50 | 1110 | 5.0 | 0.81 | 570.80 |
| 2 | 1600 | 1.75 | 1130 | 6.5 | 1.11 | 406.03 |
| 3 | 1600 | 2.00 | 1150 | 5.5 | 0.68 | 485.33 |
| 4 | 1600 | 2.25 | 1120 | 7.0 | 1.01 | 387.78 |
| 5 | 1600 | 2.50 | 1140 | 6.0 | 1.06 | 452.99 |
| 6 | 1610 | 1.50 | 1150 | 6.5 | 0.86 | 408.64 |
| 7 | 1610 | 1.75 | 1120 | 5.5 | 0.83 | 513.65 |
| 8 | 1610 | 2.00 | 1140 | 7.0 | 0.95 | 372.29 |
| 9 | 1610 | 2.25 | 1110 | 6.0 | 1.01 | 473.41 |
| 10 | 1610 | 2.50 | 1130 | 5.0 | 0.68 | 559.18 |
| 11 | 1620 | 1.50 | 1140 | 5.5 | 0.66 | 491.83 |
| 12 | 1620 | 1.75 | 1110 | 7.0 | 1.06 | 400.16 |
| 13 | 1620 | 2.00 | 1130 | 6.0 | 0.76 | 459.94 |
| 14 | 1620 | 2.25 | 1150 | 5.0 | 0.53 | 569.59 |
| 15 | 1620 | 2.50 | 1120 | 6.5 | 0.84 | 415.10 |
| 16 | 1630 | 1.50 | 1130 | 7.0 | 0.82 | 399.75 |
| 17 | 1630 | 1.75 | 1150 | 6.0 | 0.66 | 478.66 |
| 18 | 1630 | 2.00 | 1120 | 5.0 | 0.63 | 557.82 |
| 19 | 1630 | 2.25 | 1140 | 6.5 | 0.65 | 420.54 |
| 20 | 1630 | 2.50 | 1110 | 5.5 | 0.87 | 513.04 |
| 21 | 1640 | 1.50 | 1120 | 6.0 | 0.71 | 488.42 |
| 22 | 1640 | 1.75 | 1140 | 5.0 | 0.50 | 564.70 |
| 23 | 1640 | 2.00 | 1110 | 6.5 | 0.82 | 426.47 |
| 24 | 1640 | 2.25 | 1130 | 5.5 | 0.61 | 507.49 |
| 25 | 1640 | 2.50 | 1150 | 7.0 | 0.68 | 390.18 |
| Factor | Pouring temperature | Pouring speed | Shell preheating temperature | Shell thickness |
|---|---|---|---|---|
| K1 for shrinkage | 4.66 | 3.85 | 4.57 | 3.14 |
| K2 for shrinkage | 4.33 | 4.16 | 4.01 | 3.65 |
| K3 for shrinkage | 3.85 | 3.83 | 3.97 | 4.20 |
| K4 for shrinkage | 3.63 | 3.81 | 3.81 | 4.27 |
| K5 for shrinkage | 3.31 | 4.12 | 3.41 | 4.52 |
| R1 for shrinkage | 1.35 | 0.35 | 1.16 | 1.38 |
| k1 for stress | 2302.93 | 2359.44 | 2383.87 | 2822.09 |
| k2 for stress | 2327.18 | 2363.20 | 2362.77 | 2511.34 |
| k3 for stress | 2336.63 | 2301.85 | 2332.39 | 2353.42 |
| k4 for stress | 2369.80 | 2358.80 | 2302.35 | 2076.78 |
| k5 for stress | 2377.25 | 2330.49 | 2332.41 | 1950.16 |
| R2 for stress | 74.32 | 61.35 | 81.52 | 871.93 |
I constructed a BP neural network to map the four process parameters to the two responses. The input layer had four nodes, the hidden layer had nine nodes, and the output layer had two nodes. The hidden layer used a sigmoid transfer function, and the output layer used a linear transfer function. The Levenberg-Marquardt algorithm was used for training. The maximum number of training epochs was 1000, the learning rate was 0.1, and the training goal was 0.0001. The network had 54 weights and 11 biases. Because a randomly initialized BP network can fall into local minima, I optimized the initial weights and biases with particle swarm optimization. The PSO parameters were: 100 iterations, 50 particles, particle length 44, learning factors 1.5, inertia weight 0.8, position range [-0.7, 0.7], and velocity range [-10, 10]. The fitness function was the sum of prediction errors for shrinkage porosity and equivalent stress. Table 10 compares the PSO-optimized BP network with a genetic-algorithm-optimized BP network. The PSO method reached a stable fitness after about 40 generations, while the GA method stabilized after about 55 generations. The final PSO fitness was lower, indicating better prediction accuracy. I therefore used the PSO-BP model as the nonlinear mapping model for process optimization.
$$ y = f\left( \sum_{i=1}^{n} w_i x_i + b \right) $$
$$ v_{id}^{k+1} = \omega v_{id}^k + c_1 r_1 (p_{id} – x_{id}^k) + c_2 r_2 (p_{gd} – x_{id}^k) $$
$$ x_{id}^{k+1} = x_{id}^k + v_{id}^{k+1} $$
| Method | Stable generation | Final fitness | Prediction quality |
|---|---|---|---|
| GA-BP | About 55 | Higher | Acceptable but less accurate |
| PSO-BP | About 40 | Lower | Better accuracy |
After establishing the PSO-BP model, I used a genetic algorithm to search for the best process parameter combination. The optimization objective was to minimize a weighted sum of the normalized shrinkage porosity and equivalent stress. The objective function was:
$$ \min f = w_1 f_1 + w_2 f_2 $$
where \(f_1\) is the normalized shrinkage porosity, \(f_2\) is the normalized equivalent stress, and \(w_1\) and \(w_2\) are weights. I tested several weight combinations. Table 11 shows the optimized parameters and the improvement for each weight pair. As the weight on shrinkage porosity increased, the optimized pouring temperature shifted toward the lower end of the tested range, the shell preheating temperature increased, and the shell thickness decreased to about 5 mm. The pouring speed remained near 2 kg·s⁻¹ or slightly lower. When \(w_1 = 0.8\) and \(w_2 = 0.2\), the shrinkage porosity improvement reached 44.74%, while the equivalent stress improvement was 2.41%. Because shrinkage porosity is the main cause of rejection in this part, I selected this weight combination for final validation.
| w1 | w2 | Pouring temperature (°C) | Pouring speed (kg·s⁻¹) | Shell preheating temperature (°C) | Shell thickness (mm) | Shrinkage porosity (cc) | Shrinkage improvement | Equivalent stress (MPa) | Stress improvement |
|---|---|---|---|---|---|---|---|---|---|
| 0.5 | 0.5 | 1639.65 | 2.41 | 1122.00 | 6.97 | 0.68 | 10.53% | 393.11 | 14.53% |
| 0.6 | 0.4 | 1603.05 | 2.15 | 1148.89 | 5.01 | 0.54 | 28.95% | 436.62 | 5.07% |
| 0.7 | 0.3 | 1601.00 | 2.15 | 1149.96 | 5.01 | 0.49 | 35.53% | 439.84 | 4.37% |
| 0.8 | 0.2 | 1619.26 | 1.53 | 1148.70 | 5.05 | 0.42 | 44.74% | 448.86 | 2.41% |
| 0.9 | 0.1 | 1618.47 | 2.29 | 1136.91 | 5.03 | 0.58 | 23.68% | 434.85 | 5.46% |
I then ran a numerical simulation with the selected optimized parameters: pouring temperature about 1619 °C, pouring speed about 1.53 kg·s⁻¹, shell preheating temperature about 1148.7 °C, and shell thickness about 5.05 mm. The filling process became slower and more stable. The metal entered the mold smoothly, and no severe splashing occurred. The shell preheating temperature was higher than in the original process, which reduced the cooling rate during filling and helped the thin walls fill completely. The thinner shell improved heat transfer and gas permeability. During solidification, the first solid fraction still appeared at the thin left wall, and the last liquid regions remained near the frame-to-tube junction. However, the isolated liquid region was smaller than in the original design. The predicted shrinkage porosity volume decreased to 0.42 cc, which was a 44.74% reduction compared with the original 0.76 cc. The equivalent stress decreased to 448.86 MPa, a 2.41% reduction. Table 12 compares the original and optimized results. I also noted that evaporative pattern casting would likely benefit from a similar reduction in coating thickness and a similar increase in preheating temperature, because both changes improve filling and reduce thermal gradients.
| Response | Original process | Optimized process | Improvement |
|---|---|---|---|
| Shrinkage porosity volume (cc) | 0.76 | 0.42 | 44.74% |
| Equivalent stress (MPa) | 460.0 | 448.86 | 2.41% |
| Filling stability | Moderate | High | Qualitative |
| Shell thickness (mm) | 6.0 | 5.05 | Thinner, better permeability |
Finally, I performed production trials using the optimized parameters. The pouring temperature was controlled at approximately 1619 °C, the shell preheating temperature at approximately 1148 °C, the shell thickness at approximately 5 mm, and the pouring time at approximately 12 s. After knockout, cutting, shot blasting, and pickling, the parts showed no visible shrinkage cavities, no cold shuts, and no obvious distortion. The large flat surface remained within the flatness requirement. The protruding thin plate passed the checking fixture, and the cylindrical bore met the coaxiality requirement. X-ray inspection showed no obvious internal shrinkage. I also cut one part along a selected section, and the section showed that the shrinkage porosity at the frame-to-tube junction had essentially disappeared. Table 13 summarizes the final verification results. These results confirmed that combining numerical simulation with intelligent optimization can predict defects, reduce defect probability, and improve process yield and product quality. The approach can also be adapted to evaporative pattern casting, where similar optimization of pouring temperature, coating thickness, and filling speed is needed.
| Verification item | Requirement | Result after optimization |
|---|---|---|
| Surface shrinkage cavities | None visible | None observed |
| Surface scabs | None visible | None observed |
| Cold shuts | None | None observed |
| Flatness of large plate | Within tolerance | Passed |
| Coaxiality of cylindrical bore | Within tolerance | Passed |
| Internal shrinkage porosity | No unacceptable porosity | Essentially eliminated |
| Process yield | Higher than baseline | Increased |
In summary, I designed and optimized a complex thin-walled stainless steel bearing seat for investment casting. I compared five gating systems, evaluated filling and solidification behavior, and selected a combined stepped gating system with four parts per tree. I then used an orthogonal experiment to study pouring temperature, pouring speed, shell preheating temperature, and shell thickness. Shell thickness had the greatest influence on both shrinkage porosity and equivalent stress. I built a PSO-BP neural network to map process parameters to casting defects and used a genetic algorithm to find the best parameter combination. The optimized parameters reduced shrinkage porosity by 44.74% and equivalent stress by 2.41%. Production trials confirmed that the optimized process eliminates visible surface defects and internal shrinkage porosity. Throughout the study, I found that evaporative pattern casting shares many of the same physical principles: filling stability, thermal gradient control, shell or coating permeability, and defect prediction. The methods developed here can therefore support both investment casting and evaporative pattern casting of thin-walled stainless steel components.
For future work, I would extend the optimization to include more process variables, such as wax injection pressure, shell drying time, and metal superheat, and I would couple the thermal model with a stress model more tightly. I would also compare investment casting with evaporative pattern casting under the same part geometry and alloy, so that the relative advantages of each process can be quantified. In particular, evaporative pattern casting may offer advantages in pattern handling and gating flexibility, while investment casting offers superior surface finish and dimensional control. A combined simulation and optimization framework can help engineers decide which process is more suitable for a given thin-walled stainless steel part. The key is to treat filling, solidification, defect formation, and process economics as an integrated problem rather than isolated steps.