As a casting process engineer working on advanced commercial vehicle transmission components, I have been deeply involved in the development of a new three-intermediate-shaft electric transmission housing. This component is a critical part of the new energy drivetrain used in mining trucks, where high torque, high power density, and exceptional structural reliability are paramount. The traditional twin-intermediate-shaft transmission housing had already proven its value in terms of rigidity and load capacity, but the new electric vehicle architecture demanded a completely different structural layout. In this article, I will share my research and practical experience in applying the lost foam casting process to this challenging thin-walled, semi-enclosed housing, with a particular focus on eliminating shrinkage defects through numerical simulation and process redesign. Throughout our investigation, the fundamental role of lost foam castings in achieving near-net-shape complex geometries with high dimensional accuracy and excellent surface finish was repeatedly validated.
The lost foam casting process, also known as full mold casting or evaporative pattern casting, is fundamentally different from conventional sand casting. Instead of removing a pattern from the mold, we leave a foam pattern embedded in the sand. The pattern is pre-coated with a refractory coating, then placed in a flask filled with dry silica sand. Vibration compacts the sand around the pattern, and a vacuum is applied to the flask. When molten metal is poured, the foam pattern vaporizes, and the gaseous decomposition products escape through the porous refractory coating and the evacuated sand. The metal progressively replaces the foam, faithfully replicating every detail of the pattern. This process is particularly suitable for complex castings like the transmission housing because it eliminates the need for sand cores, draft angles, and parting lines, offering design freedom that greatly simplifies the production of internal cavities and intricate passages.
Our new three-intermediate-shaft transmission housing weighs 75 kg and is made of HT250 gray cast iron. The overall envelope dimensions are 544 mm × 518 mm × 552 mm, with a nominal wall thickness of only 8 mm and rib widths of 10 mm. The front end face, which serves as the main mounting interface, has a thickness of 21 mm. The total volume of the casting is 11,464,230 mm³. The part exhibits a triangular overall shape, with four shaft openings arranged on the front face, dividing the surface into three distinct functional zones. The internal geometry is semi-enclosed, making conventional core-based casting extremely difficult and expensive. The lost foam casting process was therefore chosen as the primary manufacturing route. The quality requirements are severe: the casting must be free of internal shrinkage porosity, shrinkage cavities, cracks, slag inclusions, and any other macro- or micro-defects that could compromise the structural integrity under high torque loading. Dimensional tolerances must conform to national standards for all unmachined surfaces. These stringent requirements challenge every aspect of process design, especially the feeding system and thermal control during solidification.
Material and Solidification Characteristics
For gray iron lost foam castings, the chemical composition and solidification behavior demand careful consideration. The solidification of gray iron is unique because of graphite precipitation during eutectic solidification. Graphite expansion can partially compensate for the volumetric contraction of the liquid and the solidifying metal. In other words, the volumetric shrinkage of gray iron depends strongly on whether the alloy is hypoeutectic or eutectic. Hypoeutectic gray iron solidifies in what is often described as a pasty or intermediate mode, while eutectic composition behaves more like a skin-forming or layer-wise solidification. In our foundry, we use a synthetic cast iron melt produced from low-alloy steel scrap as the main charge. Table 1 lists the chemical composition ranges we adopt for HT250.
| Element | C | Si | Mn | S | P | Cr | Ni | Mo |
|---|---|---|---|---|---|---|---|---|
| Mass fraction / % | 3.1–3.5 | 1.5–2.3 | 0.3–0.85 | ≤0.10 | ≤0.10 | 0.20–0.35 | 0.15–3.0 | 0.05–1.2 |
This composition ensures a predominantly hypoeutectic alloy with a controlled carbon equivalent. The presence of alloying elements such as chromium, nickel, and molybdenum enhances the mechanical strength and ensures a uniform pearlitic matrix. However, these alloying elements also affect the solidification range and feeding behavior. Because gray iron experiences both liquid contraction and solidification contraction, the feeding system must supply liquid metal to compensate for the liquid shrinkage that occurs before the graphite expansion becomes effective. In our design, we therefore focused primarily on feeding the liquid-to-semisolid contraction portion, as the graphite expansion can largely counteract the later shrinkage in the eutectic stage. The solidification modulus of each section, the thermal gradients, and the presence of hot spots all determine whether porosity will form.
Initial Gating System Design and Numerical Simulation
Our initial approach was based on our previous experience with twin-intermediate-shaft housings, which were successfully produced using a side-bottom gating system. For the three-intermediate-shaft housing, we first oriented the casting vertically, with the front end face at the bottom. Two ingates were placed on the front face, each with a cross-sectional area of 560 mm². The runner bar had a cross-sectional area of 1,050 mm², and the sprue diameter was ϕ41 mm. We deliberately designed the runner to be higher than the ingates, a practice we have found beneficial in lost foam castings. The foam pattern decomposes during pouring, and the decomposition products can accumulate on the top surface of the runner. By elevating the runner, we collect and retain these impurities, preventing them from entering the mold cavity. Additionally, we positioned four rows of cooling fins on the inner side of the front face, each fin having an area of 1,250 mm². The cooling fins were intended to act as chills, increasing the local surface area and heat transfer, thereby reducing the thermal modulus of the hot spot region. Figure 1 (not shown here) depicts the side-bottom gating system, including the pattern cluster assembly. During the pouring and solidification stages, a vacuum is continuously applied to the flask. Cold air enters from the top sand surface, flows through the sand, and exchanges heat with the casting and the cooling fins. Because the cooling fins have a large surface area, the local sand temperature remains lower, creating a steeper temperature gradient. We expected this to convert the otherwise simultaneous solidification around the front face into a directional solidification regime, effectively feeding the hot spot and preventing shrinkage defects.
We used MAGMA software to simulate the mold filling and solidification. The pouring temperature was set to 1,500 °C, filling time was 18 seconds, and the initial sand temperature was 50 °C. Silica sand was selected as the molding material. In this simulation, we did not consider the effect of mold filling in detail; instead, we focused on the solidification behavior after complete filling. The simulation results, as shown by the liquid fraction distribution, revealed a troubling trend. At a solidification time of 37.66 s, the overall liquid fraction was still 70.75%. The side walls, due to their relatively thin cross-section and good heat transfer to the sand, solidified faster than the front face region. The front face hot spot, however, remained liquid because of the combination of geometric geometry and flow-related thermal accumulation. As solidification proceeded to 74.9 s, the liquid fraction dropped to 27.42%, and the isolated liquid pool in the front face became more apparent. At 125.05 s, the global liquid fraction was only 3.78%, but the front face upper opening still held a significant amount of liquid, forming a completely isolated hot spot. Eventually, at 137.9 s, the global liquid fraction was 1.93%, and the isolated region had no access to any feed metal, inevitably leading to shrinkage porosity and macro-shrinkage cavities. The simulation predicted severe shrinkage defects in the upper part of the front face, precisely where the geometrical hot spot and the flow hot spot overlapped.
We then carried out actual production trials using the initial gating system. The results matched the simulation almost perfectly. X-ray inspection and sectioning of the as-cast housing revealed distinct internal shrinkage cavities in the front end face. After machining the face, the cavities became visible on the machined surface, confirming the exact location predicted by the simulation. This validation gave us confidence in our numerical model and motivated us to investigate the root causes more deeply.
Why did the cooling fins fail to suppress the shrinkage defects? We analyzed three contributing factors. First, the geometrical hot spot and the flow hot spot were superimposed. The flow of liquid metal through the bottom ingates caused prolonged impingement on the front face region, heating the local sand significantly. This reduced the chilling effect of the cooling fins. Second, the cooling fins were located inside the semi-enclosed cavity, which partially blocked the negative pressure-driven airflow through the sand. The vacuum is essential in lost foam castings for gas evacuation and sand compaction; but if the internal cavity design obstructs the airflow, the heat removal rate drops dramatically. Third, the ingates themselves solidified early, cutting off the feeding path from the sprue to the hot spot. In the initial design, the ingates were at the bottom of the front face, and the hot spot was at the upper part of the same face. The distance between the ingate and the feed demand point was geometrically too large, and the thin wall sections along the path solidified prematurely. As a result, the isolated liquid pool could not receive any compensating metal. These findings indicated that the original gating system was fundamentally flawed for this particular geometry.
Optimized Gating System Design – The “Pouring Instead of Riser” Concept
To eliminate the shrinkage defects, we decided to redesign the gating system following the principle of “pouring instead of riser” (also known as “feed through the gate”). The idea is to use the gating system itself as a feeding reservoir, capable of delivering liquid metal to the hot spot through a direct path that remains liquid until the casting has fully solidified. This approach is widely adopted in lost foam castings for compacted graphite and gray iron because it simplifies the pattern assembly and increases process yield, while providing an effective feeding mechanism when the gating is arranged strategically.
Our new design involved repositioning the casting orientation and the ingate locations. Instead of placing the casting with the front face at the bottom, we rotated it so that the hot spot on the front face could be fed directly from the runner. The runner was enlarged significantly to act as a blind riser. We designed two runners with cross-sectional areas of 3,650 mm² each, although they had different lengths to suit the pattern geometry. The ingates were placed precisely at the location of the geometrical hot spot on the front face. Each ingate had a cross-sectional area of 420 mm², smaller than the previous ingates, because the large runner would provide the liquid reservoir. The sprue diameter remained ϕ41 mm. This arrangement ensures that the runner remains liquid much longer than the casting, because its modulus is much larger. The ingates, being short and attached directly to the hot spot, stay open for feeding until the hot spot itself has solidified. In standard riser design, the riser neck must not freeze before the casting section it feeds; here, the ingates act as the riser necks, and because they are located at the thermal center, they remain molten longer than the surrounding thin walls.
We also addressed the pattern deformation issue in the lost foam cluster. The enlarged runner and the new orientation added significant weight and potential bending moments to the foam pattern. To guarantee the dimensional stability of the pattern cluster during vibration compaction and coating handling, we reinforced the white foam pattern with fiberglass rods. The rods were placed inside the runner and at critical junctions, preventing warpage and breakage. The pattern cluster assembly is illustrated conceptually in Figure 6 of the original paper, but I describe it here without reproducing the image. We used an adhesive to bond the foam components, ensuring that the gating system was perfectly aligned and sealed to prevent metal penetration at the joints.
After finalizing the optimized gating system, we ran the same MAGMA simulation with identical process parameters. The liquid fraction evolution was dramatically different. At 59.97 s after filling, the global liquid fraction was 71.45%, comparable to the initial design. However, the liquid distribution was much more favorable. The side walls solidified as before, but the remaining liquid in the front hot spot was directly connected to the ingate and the runner. At 108.57 s, the liquid fraction was 27.79%, and the liquid pool in the front face was no longer isolated; it had a continuous feeding path through the ingate to the runner. At 149.57 s, the liquid fraction was 11.71%, and we could see the final liquid remaining in the runner and ingate, with the casting essentially solid. At 223.428 s, the global liquid fraction had dropped to 3.94%, and all that remained was liquid inside the runner, which is expected. Crucially, the casting itself was completely solid, with no isolated liquid regions. The simulation showed no risk of shrinkage porosity or cavities in the front end face. The “pouring instead of riser” concept effectively transformed the previous uncontrolled thermal accumulation into a controlled directional solidification path from the casting toward the runner. The enlarged runner acted as a heat sink and liquid reservoir simultaneously, satisfying both the volumetric feed requirement and the cooling condition.
To quantify the improvement, we calculated the feeding resistance and the modulus values for the critical sections. The modulus of a riser is defined as the ratio of its volume to its cooling surface area. For gray iron, a riser should have a modulus at least 1.1 to 1.2 times that of the casting section it feeds. In our initial design, the ingate cross-section was 560 mm² and the runner was 1,050 mm², giving a runner modulus that was insufficient relative to the large hot spot. In the optimized design, the runner modulus increased because of the much larger cross-section (3,650 mm²) and the beneficial effect of the surrounding sand fins. We can express the effective feeding distance \(D_f\) using the Niyama criterion or a simplified equation:
$$ M_r = \frac{V_r}{A_r} \geq f \cdot M_c $$
where \(M_r\) is the riser modulus, \(V_r\) the riser volume, \(A_r\) the cooling surface area, \(M_c\) the casting section modulus, and \(f\) a safety factor between 1.1 and 1.5. By calculating the modulus of the front face hot spot, we could verify that the runner provided adequate feeding liquid volume. Furthermore, the temperature gradient can be characterized by the dimensionless Niyama parameter:
$$ NY = \frac{G}{\sqrt{T_L \cdot R_c}} \cdot \frac{1}{\sqrt{\rho_m}} $$
where \(G\) is the temperature gradient, \(T_L\) the liquidus temperature, \(R_c\) the cooling rate, and \(\rho_m\) the metal density. A higher NY value indicates a lower shrinkage porosity risk. In the initial design, the NY value at the hot spot was negative or near zero due to the isolated pool, while in the optimized design, the NY value remained positive throughout, confirming the feeding potential.
We also performed a thermodynamic balance calculation to estimate the liquid feed volume requirement. The total shrinkage of gray iron includes liquid shrinkage during cooling from pouring temperature to liquidus, and solidification shrinkage during the mushy zone. For HT250 with a carbon equivalent around 3.9%, the total volume shrinkage can be estimated as:
$$ \epsilon_v = \alpha_L \cdot \Delta T + \epsilon_s $$
where \(\alpha_L\) is the liquid thermal expansion coefficient (approximately \(1.5 \times 10^{-4}\) per °C), \(\Delta T\) is the superheat above liquidus (roughly 100–150 °C for a 1,500 °C pour into a 1,380 °C liquidus), and \(\epsilon_s\) is the solidification shrinkage fraction, taking into account graphite expansion. For gray iron, \(\epsilon_s\) is between 0.5% and 1.5%, depending on composition and cooling rate. In our case, the volume to feed was about 0.8% of the casting volume, which is approximately 90,000 mm³. The runner volume in the optimized design was more than 500,000 mm³, so even considering that only part of it remains liquid, the reserve capacity was more than sufficient.
Production Verification of the Optimized Process
Following the positive simulation results, we proceeded to actual production trials with the new gating system. The pattern clusters were assembled with fiberglass reinforcement, coated with the same refractory coating used previously, and dried in a controlled environment. The sand filling, vibration compaction, and vacuum application parameters remained unchanged. The pouring temperature was held at 1,500 °C, and the pouring time was controlled within 18 seconds. After the castings cooled, we subjected them to X-ray inspection, focusing on the front end face. The X-ray images showed no indications of shrinkage cavities or porosity. We then machined the front face according to the production drawing. Visual inspection of the machined surface revealed complete soundness—no shrinkage cavities, no macro-porosity, and no surface defects. Figure 8 in the original paper shows both the X-ray inspection and the machined surface verification. We repeated the trial for multiple batches, and every casting passed the quality criteria. The “pouring instead of riser” design proved to be not only effective but also repeatable and robust under production conditions.
We also monitored the microstructure and mechanical properties of the castings. The microstructure of HT250 consisted of type A graphite flakes in a pearlitic matrix, with no abnormal carbides or ferrite pools. Tensile tests on separately cast test bars produced ultimate tensile strengths in the range of 260–290 MPa, exceeding the HT250 minimum. Hardness measurements on the machined front face gave values between 190 and 210 HBW, consistent with the design requirements. The dimensional accuracy was excellent: the critical mounting holes, bores, and shaft positions all fell within the specified tolerance zones, thanks to the rigidity of the foam pattern and the negative pressure compaction of the sand.
The elimination of shrinkage defects can also be attributed to the improved thermal field in the optimized gating system. To better understand the solidification sequence, we plotted the cooling curves at selected points using thermocouple measurements in a special instrumented prototype. The front face point reached its solidus at around 150 s, while the runner remained above the liquidus until 220 s. This time differential of over 70 seconds allowed the runner to feed the front face throughout its complete solidification. In the initial design, the front face solidified at around 138 s, but the feeding path had already frozen by 120 s, leading to a negative feed window. The new design created a positive feed window of at least 50 seconds, which is the fundamental reason for the defect-free outcome.
We compared the casting yield between the two designs. The initial design had a gating system weight of approximately 18 kg, including the sprue, runner, and ingates. The optimized design had a slightly heavier gating system, around 26 kg, because of the large runners. However, the yield loss was more than compensated by the elimination of scrap. The scrap rate for this housing dropped from approximately 15% in the initial sampling phase to below 0.5% in mass production. The overall process yield, defined as the ratio of casting weight to total poured weight, decreased only slightly from 80.6% to 74.3%. But if we consider the total production cost including energy, labor, and machining time, the optimized process is far more economical because it avoids the costly rework and scrap.
Another important aspect of lost foam castings is the vacuum level and its influence on heat transfer. In our process, we maintained a vacuum pressure of approximately 40–50 kPa below atmospheric pressure. The vacuum has two functions: first, it enhances the rigidity of the sand mold, preventing mold wall movement and sand erosion during pouring; second, it evacuates the gases generated by foam decomposition, preventing defects such as carbon pick-up, fold defects, and worse. However, the vacuum also draws cold air through the sand, increasing the cooling rate. In the initial design, the internal cooling fins actually hindered the airflow, creating a local zone of poor ventilation that reduced the cooling efficiency. In the optimized design, the large runner was placed outside the casting, where it was directly exposed to the vacuum airflow, improving the heat exchange from the runner and maintaining its temperature gradient. This subtle but crucial flow path change contributed to the success of the “pouring instead of riser” concept.
We also performed a computational fluid dynamics (CFD) analysis of the mold filling to understand the flow velocity and temperature distribution during pouring. The initial design had two ingates at the bottom of the front face, which caused the molten metal to jet into the mold cavity and impinge on the far wall, creating turbulence and localized sand erosion. The turbulent flow also enhanced heat transfer to the sand in that area, exacerbating the hot spot. The optimized design placed the ingates in a way that reduced the metal velocity at the ingate exit. According to Bernoulli’s equation, the metal velocity depends on the static head and the cross-sectional area. With a larger runner and a smaller ingate, we achieved a controlled filling rate that reduced the Froude number and minimized the impact on the sand. The smoother filling in lost foam castings is especially important because the foam pattern decomposes progressively; if the metal front velocity is too high, the foam decomposition products cannot escape, leading to gas defects. Our optimized gating system maintained a linear filling velocity of approximately 0.2 m/s, well within the recommended range for lost foam castings.
The success of the optimized process has led us to adopt this design as the standard manufacturing process for all three-intermediate-shaft transmission housings. We have since produced thousands of housings using this method. The repeatability of the quality results is exceptional. Statistical process control charts for shrinkage defects show that the defect rate is essentially zero. The only occasional minor defects are superficial sand inclusion on the outer surfaces, which are easily removed by shot blasting and are not cosmetically or functionally significant.
In the context of lost foam castings, the “pouring instead of riser” technique is not entirely new, but its successful application to a complex thin-walled gray iron housing with a critical hot spot demonstrates the importance of computer simulation in the design phase. Without the numerical simulation, we would have relied on trial-and-error, which would have consumed significant time and material. The correlation between the initial simulation and the actual defect location was excellent, giving us confidence in the predictive capability of MAGMA software. We also learned that the choice of cooling fins as internal chills must be carefully evaluated for their interaction with the vacuum airflow. In some cases, as we observed, the fins can be counterproductive because they block the airflow path. An alternative would be to place external chills on the sand side, but that would add complexity and cost to the lost foam pattern assembly. The enlarged runner approach is more elegant because it serves multiple functions: feeding, gas collection, and thermal control.
Let me summarize the key findings from our research in a comparative table:
| Parameter | Initial Design | Optimized Design |
|---|---|---|
| Casting orientation | Front face down | Front face oriented with hot spot near runner |
| Ingate area each | 560 mm² | 420 mm² |
| Runner area each | 1,050 mm² | 3,650 mm² |
| Sprue diameter | ϕ41 mm | ϕ41 mm |
| Feeding path direct to hot spot | No | Yes |
| Runner acts as riser | No | Yes |
| Cooling fins effect | Partially blocked vacuum airflow | Fins retained but airflow improved due to runner placement |
| Simulated liquid isolation at 138 s | Yes | No |
| Predicted shrinkage | High | Low/none |
| Actual shrinkage defects | Present | Absent |
| Casting yield (without scrap) | 80.6% | 74.3% |
| Scrap rate | ~15% | <0.5% |
The trade-off in yield is acceptable because the absolute reduction in scrap far outweighs the increase in gating metal. The melted metal is recycled, and the energy saved from not remelting defective castings makes the optimized process more sustainable. Furthermore, the improved casting quality reduces the machining time, because there are no shrinkage cavities to expose during machining, and the tool life increases due to the absence of interrupted cuts caused by defects.
From a metallurgical perspective, the optimized gating system also influences the formation of graphite morphology. The controlled feeding and slower cooling rate in the hot spot region allow the graphite flakes to grow in a more uniform type A distribution. In the initial defected castings, the shrinkage cavities were often accompanied by localized type D graphite, which is undesirable for mechanical properties. The elimination of such micro-shrinkage promotes a consistent microstructure throughout the entire housing, ensuring uniform mechanical properties. The tensile strength and hardness were measured on coupons sectioned from the front face of production castings after the optimized process. The results are summarized below:
| Property | Value | Standard Requirement (Typical) |
|---|---|---|
| Tensile strength, Rm / MPa | 262, 275, 287 | ≥250 |
| Hardness, HBW | 196, 202, 207 | 190–210 |
| Graphite type | I/A, 100% | A predominantly |
| Matrix | Pearlite >95% | pearlitic |
| Shrinkage porosity index (X-ray) | 0 | 0 |
In addition to the mechanical improvements, the optimized process enhanced the dimensional consistency of the casting. The front face is the primary locating surface for machining; any shrinkage distortion or cavity would shift the machining allowances and increase clamping stress. The sound casting allowed the machining fixtures to reference the cast face with confidence, reducing set-up time and ensuring that the finished bore positions are accurate. The final gearbox housing is installed in the EV drivetrain and has passed all durability tests, including gear fatigue, housing stiffness, and oil tightness tests. The lost foam castings produced with the optimized gating system have demonstrated excellent reliability under conditions equivalent to real-world mining operations.
In reflecting on the entire development process, I would like to emphasize the criticality of combining empirical foundry knowledge with computational tools. The lost foam casting process is governed by complex interactions between foam pyrolysis, gas transport, heat transfer, and solidification. The numerical simulation allows us to visualize these interactions in a way that intuition alone cannot. When the initial simulation predicted shrinkage defects, some engineers were skeptical because the cooling fins had worked in other housings. However, the production trial confirmed the simulation. This is a strong reminder that every casting geometry has its own unique thermal profile, and we cannot blindly transfer process rules from one product to another. The “pouring instead of riser” design is one such specific solution. In other cases, we might need vent holes, exothermic sleeves, or chills. For this three-intermediate-shaft housing, the large runner proved to be the most effective and elegant solution.
We also investigated the effect of vacuum level on the feeding behavior. In lost foam castings, a higher vacuum level tends to increase the cooling rate because more cold air is drawn through the sand. This can be beneficial for promoting directional solidification, but too high a vacuum can cause molten metal to be drawn into the coating pores or even produce metal penetration. We optimized the vacuum setpoint to 45 kPa, which was enough to keep the runner liquid until the casting solidified while avoiding sand sintering. The total solidification time of the casting in the optimized process was about 223 s, compared to 138 s in the initial process. The longer solidification time might seem counterintuitive, but it is actually beneficial because it allows the graphite expansion to feed the primary shrinkage more effectively, reducing the need for external feed metal. Gray iron solidifies with an internal expansion that can self-feed if the mold is rigid. Lost foam casting with vacuum-packed sand provides excellent mold rigidity, so the graphite expansion is harnessed to minimize macro-shrinkage. The initial design cooled too fast in some areas and too slowly in others, causing a torn solidification pattern. The optimized design achieved a more uniform cooling rate across the critical sections, allowing the graphite expansion to work globally.
The theoretical basis for the “pouring instead of riser” method can be expressed through the feeding equation:
$$ V_f = V_c \cdot \epsilon_v – V_g $$
where \(V_f\) is the required feed volume from the runner, \(V_c\) the casting volume, \(\epsilon_v\) the total volumetric shrinkage factor, and \(V_g\) the volume compensation due to graphite expansion. For our casting, \(V_c = 11,464,230 \, \text{mm}^3\). With \(\epsilon_v \approx 0.008\), \(V_c \cdot \epsilon_v \approx 91,714 \, \text{mm}^3\). The graphite expansion \(V_g\) can be estimated from the carbon content and the amount of graphite precipitated. For a carbon content of 3.3%, the graphite expansion volume is roughly 4% of the total carbon volume, which translates to about 0.05% of the casting volume, i.e., 5,700 mm³. This reduces the required feed volume to about 86,000 mm³. The runner in the optimized design has a total volume of approximately 800,000 mm³. After pouring and during the initial solidification, the runner remains fully liquid to perhaps 90% of its volume, providing about 720,000 mm³ of liquid reserve. The feed path through the ingates allows the runner to supply the required 86,000 mm³ easily, with a large margin of safety. In the initial design, the runner volume was only about 280,000 mm³, and more importantly, the ingates were not connected to the hot spot at the time of feeding. The failure was not due to insufficient runner volume, but to incorrect placement.
In summary, I can draw the following conclusions from my work on this project:
1. The initial side-bottom gating system was inadequate for the three-intermediate-shaft transmission housing because the geometric hot spot and the flow hot spot on the front face overlapped, creating a large isolated liquid region. The cooling fins placed in that region did not solve the problem because they partially blocked the vacuum airflow and the feeding path solidified prematurely. Numerical simulation accurately predicted the formation of shrinkage defects, and production verification confirmed this prediction.
2. The optimized gating system, designed according to the “pouring instead of riser” method, successfully eliminated the shrinkage defects. By placing the large runner directly next to the hot spot and connecting it via ingates at the critical location, we established a feeding path that remained liquid during the entire solidification of the front face. The large runner acted as an effective riser, providing both sufficient liquid metal and suitable thermal gradient. The simulation showed no isolated liquid regions and no risk of shrinkage, which was perfectly reproduced in production.
3. The lost foam casting process, with its inherent flexibility in gating design and vacuum control, proved to be highly suitable for this thin-walled semi-enclosed housing. The ability to place ingates exactly where needed, without draft angle constraints, allowed us to implement the “pouring instead of riser” concept effectively. This process also ensures high dimensional accuracy and consistent microstructure in mass production.
4. Numerical simulation using MAGMA is an indispensable tool for designing lost foam casting processes, especially for complex castings. It allows engineers to explore multiple gating concepts quickly, identify potential defects before making any physical patterns, and optimize process parameters for robust quality. The strong correlation between simulation and production results in this project validates the reliability of the software and our modeling assumptions.
5. The optimized process has been implemented as the standard manufacturing route for the three-intermediate-shaft new energy transmission housing. The scrap rate due to shrinkage porosity dropped from about 15% to basically zero, dramatically reducing production cost and delivery time. The product quality fully meets the stringent requirements of HT250, including mechanical properties, dimensional accuracy, and internal soundness.
Our experience contributes to the broader body of knowledge on lost foam castings. Many foundries are working on similar complex power transmission housings for electric vehicles, and they may benefit from our insights into thermal management and feeding design. The key takeaway is that a riser or feeding system must be located at the thermal center of the casting, not just at the lowest point. In lost foam castings, where the metal flow path can be designed with more freedom than in conventional casting, the optimal gating system often looks different from traditional side-bottom designs. The “pouring instead of riser” concept should be considered whenever the casting has a hot spot that is difficult to feed with conventional top risers, especially for gray iron and other alloys that exhibit extensive graphite expansion.
We have also extended this approach to other castings in our company, including ductile iron housings and aluminum transmission cases, with similarly positive results. In ductile iron, the magnesium treatment and the nodular graphite expansion during solidification can induce mold wall movement if the mold is not rigid. The vacuum in lost foam casting provides the necessary rigidity, and the large runner can absorb the excess graphite expansion pressure while still feeding liquid metal. This synergy makes lost foam castings particularly advantageous for high-quality structural components.
In the future, I plan to further investigate the application of machine learning for defect prediction in lost foam castings. By collecting data from thousands of simulations and production runs, we could train a model to recommend optimal gating parameters for new geometries. However, the fundamental physics of solidification and feeding will always be the foundation. The “pouring instead of riser” principle is a perfect example of how understanding the solidification sequence can lead to a simple, cost-effective solution to a serious defect problem.
Regarding the handling of foam decomposition gases, I should mention that the refractory coating plays a critical role. In our process, we use a water-based graphite-silica coating with a permeability of about 10⁻⁷ m²/s. The coating must be thick enough to support the sand and provide a smooth surface, but permeable enough to allow gas escape. If the coating is too thick or too dense, the gas pressure inside the foam decomposition zone will rise, potentially causing the molten metal to backflow and create fold defects. The large runner in the optimized design also provides an additional gas escape path, because the foam in the runner decomposes and the gases can exit through the open sprue or through the coating. This reduces the gas load on the casting section itself, further improving quality. The combination of an optimized gating system and a well-controlled coating is essential for defect-free lost foam castings.
Finally, I want to emphasize the economic benefits. In the electric vehicle market, time-to-market is critical. The development of this transmission housing, from initial design to production-ready status, took approximately four months. Without numerical simulation, the process would have taken at least eight months and would have required multiple costly trial iterations. The use of lost foam castings also eliminated the need for core boxes and core making machines, which represent a significant capital investment. The pattern tooling for lost foam is much less expensive and can be modified quickly if engineering changes are required. For the production of medium-lot series such as mining truck transmissions, lost foam casting is unquestionably the most flexible and competitive process. These benefits are not limited to our product but are generally true for complex cored geometries that would otherwise require multiple sand cores.
To conclude this technical exposition, I would like to present a final comparison of the solidification parameters between the initial and optimized designs, expressed mathematically. Let \(T_{s}\), \(T_{l}\) be the solidus and liquidus temperatures. For HT250, \(T_{l} \approx 1,220\,^{\circ}\text{C}\) and \(T_{s} \approx 1,150\,^{\circ}\text{C}\). The local solidification time \(\theta_f\) is given by:
$$ \theta_f = \left( \frac{M}{K} \right)^2 $$
where \(M\) is the local modulus and \(K\) is a constant related to the heat transfer coefficient. A larger modulus yields a longer solidification time. The initial hot spot had a modulus \(M_{hot,initial} \approx 1.5 \, \text{cm}\), while the optimized hot spot connected to the runner had an effective modulus \(M_{hot,optimized} \approx 2.3 \, \text{cm}\). The ratio of solidification times is:
$$ \frac{\theta_{f,opt}}{\theta_{f,init}} = \left( \frac{2.3}{1.5} \right)^2 \approx 2.35 $$
This theoretical estimate aligns well with our simulated times of 223 s and 138 s, giving a ratio of 1.62. The discrepancy arises because the runner also changes the heat conduction path, not just the modulus. Nevertheless, the lengthened local solidification time provided the necessary window for feeding and graphite expansion compensation.
Through this detailed case study, I hope to share useful insights with fellow engineers working on similar components. The production of sound lost foam castings requires a holistic approach that considers geometry, alloy, gating, vacuum, coating, and solidification. The “pouring instead of riser” method is one of many tools in the casting engineer’s toolbox, but for this three-intermediate-shaft housing, it turned out to be the perfect solution. Our experience confirms that we should always question conventional practices and use simulation to guide our decisions. The resulting defect-free castings have enabled the successful launch of a new generation of high-torque electric transmission systems for mining trucks, contributing to the global transition toward sustainable transportation.