In my research, I focused on the optimization of metal core bars used in sand mould casting. Sand mould casting is one of the most important manufacturing processes for producing complex metal components in industries such as aerospace, automotive, energy, and heavy machinery. As castings become larger and more integrated, the requirements on sand moulds and cores—such as strength, collapsibility, and gas permeability—vary significantly across different regions. Metal core bars are embedded within sand cores to enhance permeability and collapsibility, improve core strength, and meet the diverse demands of complex moulds. However, traditional design of metal core bars relies heavily on empirical knowledge. In complex sand cores, the distribution of core bar material lacks theoretical guidance, often resulting in over-design, material waste, and increased cost. Topology optimization offers a powerful method to find the optimal material layout under given mechanical performance constraints. Therefore, I proposed a topology optimization-based method for designing metal core bars in sand mould casting. This article presents my mathematical model, numerical implementation, experimental validation, and practical applications in large castings.

I began by reviewing existing topology optimization methods. Among them, the Solid Isotropic Microstructures with Penalization (SIMP) model is widely used due to its simplicity and efficiency. The SIMP model introduces a continuous density variable for each finite element, ranging from 0 to 1, and uses a penalty factor to push intermediate densities toward 0 or 1. This approach converts the discrete 0-1 optimization problem into a continuous one, which can be solved using gradient-based algorithms. I adopted the SIMP model for metal core bar optimization. The material interpolation is expressed as:
$$E(x_e) = x_e^p E_0$$
where \(E(x_e)\) is the Young’s modulus of element \(e\), \(x_e\) is the relative density, \(p\) is the penalty factor (typically \(p \ge 3\)), and \(E_0\) is the solid material modulus. The global stiffness matrix \(K\) is assembled as:
$$K = \sum_{e=1}^{N} x_e^p k_e$$
Here \(k_e\) is the element stiffness matrix for solid material. The optimization problem aims to minimize the structural compliance \(C\), which is equivalent to maximizing stiffness. The mathematical formulation is:
$$\begin{aligned}
\min_{x} \quad & C = U^T K U = \sum_{e=1}^{N} u_e^T k_e u_e \\
\text{s.t.} \quad & K U = F \\
& \sum_{e=1}^{N} v_e x_e \le V^* \\
& 0 < x_{\min} \le x_e \le 1
\end{aligned}$$
where \(U\) is the global displacement vector, \(F\) is the external load vector, \(v_e\) is the element volume, \(V^*\) is the prescribed volume constraint, and \(x_{\min}\) is a small lower bound (e.g., \(10^{-3}\)) to avoid singularity. I used the Optimality Criteria (OC) method to update the design variables. The OC update scheme is:
$$x_e^{\text{new}} = \begin{cases}
\max(x_{\min}, x_e – m) & \text{if } x_e B_e^\eta \le \max(x_{\min}, x_e – m) \\
x_e B_e^\eta & \text{if } \max(x_{\min}, x_e – m) < x_e B_e^\eta < \min(1, x_e + m) \\
\min(1, x_e + m) & \text{if } \min(1, x_e + m) \le x_e B_e^\eta
\end{cases}$$
where \(m\) is a positive move limit (e.g., 0.2), \(\eta\) is a damping coefficient (e.g., 0.5), and \(B_e\) is given by:
$$B_e = \frac{-\frac{\partial C}{\partial x_e}}{\lambda \frac{\partial V}{\partial x_e}}$$
Here \(\lambda\) is a Lagrange multiplier for the volume constraint, updated via a bisection algorithm. The sensitivity of compliance with respect to density is:
$$\frac{\partial C}{\partial x_e} = -p x_e^{p-1} u_e^T k_e u_e$$
To prevent numerical instabilities such as checkerboarding and mesh dependence, I employed a mesh filtering technique. The filtered sensitivity is:
$$\frac{\partial \hat{C}}{\partial x_e} = \frac{\sum_{f=1}^{N} H_f x_f \frac{\partial C}{\partial x_f}}{x_e \sum_{f=1}^{N} H_f}$$
where the convolution operator \(H_f\) is defined as:
$$H_f = \max(0, r_{\min} – \text{dist}(e,f))$$
with \(r_{\min}\) being the minimum filter radius and \(\text{dist}(e,f)\) the distance between element centers. This filtering ensures that the optimization result is mesh-independent and manufacturable.
Based on these theories, I developed a topology optimization program using C++. The program includes modules for finite element analysis, sensitivity computation, filtering, OC update, and convergence checking. The overall flowchart consists of initializing the design domain, applying boundary conditions and loads, solving the finite element problem, computing compliance and sensitivities, applying mesh filtering, updating densities via OC, and checking convergence. I validated the program with two simple cylindrical examples. In the first example, a cylinder of diameter 20 and length 80 was fixed at one end and loaded downward at the other end. The volume constraint was 40% of the original volume. After 26 iterations, the optimized structure resembled a cantilever beam with truss-like members. In the second example, the cylinder was fixed at both ends and loaded at the middle. After 43 iterations, the optimized structure exhibited an arch-like shape. These tests confirmed the correctness and stability of my program.
To further verify the algorithm, I examined the convergence behavior. The compliance decreased monotonically and stabilized after a number of iterations. The volume constraint was satisfied within a tolerance of \(10^{-3}\). The filtered sensitivity effectively suppressed checkerboard patterns. I also tested different penalty factors \(p = 3, 4, 5\) and found that \(p = 3\) provided a good balance between discreteness and convergence. The move limit \(m = 0.2\) and damping \(\eta = 0.5\) were selected based on standard practice. The filter radius was set to 1.5 times the element size to ensure smooth boundaries.
Next, I conducted a series of experiments to validate the effect of topology-optimized metal core bars on the mechanical properties of sand cores. I designed three sets of specimens: without any metal core bar, with a conventional cylindrical core bar (diameter 6 mm, height 46 mm), and with a topology-optimized core bar. The sand cores were made of silica sand (50–100 mesh) mixed with an inorganic binder and a hardener. The mass ratio of sand to binder was 100:3, and the binder to hardener ratio was 100:18. The mixture was cured for 24 hours before testing. All specimens were prepared in batches to ensure consistency. The metal core bars were manufactured via 3D printing using stainless steel to avoid wear during repeated experiments. The optimized core bar geometry was obtained from the topology optimization program with the load case shown in Figure 3-2 of the original study, which corresponds to a three-point bending configuration: fixed at both ends and loaded at the center.
For the bending test, I used a three-point bending setup. The sand core specimen had dimensions 160 mm × 20 mm × 22 mm. The metal core bar was placed longitudinally inside the core. The loading speed was 10 mm/min. I prepared 20 specimens for each group. After excluding outliers caused by surface defects or incomplete curing, I selected 10 specimens per group for analysis. The results are summarized in Table 1.
| Group | Average Bending Strength (MPa) | Increase vs. No Core Bar | Mass of Core Bar (g) |
|---|---|---|---|
| No core bar | 0.79 | — | — |
| Conventional core bar | 1.07 | 33.6% | 9.62 |
| Optimized core bar (as-optimized) | 0.93 | 16.3% | 3.85 |
| Optimized core bar (smoothed ends) | 1.08 | 34.9% | 3.85 |
I observed that the as-optimized core bar had sharp edges at the ends, causing stress concentration and premature fracture. After smoothing the ends, the bending strength increased to 1.08 MPa, slightly higher than the conventional core bar. More importantly, the mass of the optimized core bar was reduced by 59.9% compared to the conventional one. The fracture location shifted from the core bar ends to the middle of the sand core, indicating a more uniform stress distribution.
For the compression test, I used cylindrical sand cores of diameter 20 mm and height 50 mm. The loading direction was axial. I prepared 30 specimens per group and selected 10 for analysis. The results are shown in Table 2.
| Group | Average Compressive Strength (MPa) | Change vs. No Core Bar |
|---|---|---|
| No core bar | 1.06 | — |
| Conventional core bar | 0.93 | -12.3% |
| Optimized core bar (smoothed) | 0.94 | -11.3% |
The addition of any core bar slightly reduced the axial compressive strength because the core bar orientation aligned with the crack propagation direction. However, the difference between conventional and optimized core bars was negligible. Since axial loads are small in sand mould casting, this reduction is acceptable.
For the tensile test, I used an “8”-shaped standard sand core. The loading direction was axial. I prepared 30 specimens per group and selected 10 for analysis. The results are given in Table 3.
| Group | Average Tensile Strength (MPa) |
|---|---|
| No core bar | 0.86 |
| Conventional core bar | 0.84 |
| Optimized core bar (smoothed) | 0.84 |
The tensile strength was almost identical across all groups, as fracture occurred at the narrowest section perpendicular to the core bar. Thus, the optimized core bar does not degrade the axial tensile strength of sand cores.
I then applied my method to two large industrial castings: a crown casting and a guide vane casting. For the crown casting, the maximum diameter was 2200 mm. The original process used a sand core with a metal core bar frame consisting of two rings and nine cylindrical bars, weighing 41 kg. I performed casting numerical simulation using a commercial CAE software to obtain the filling, solidification, and deformation behavior. The mesh size was 10 mm, resulting in 7,057,570 elements, with 1,057,246 elements for the casting body. The material was stainless steel with a pouring temperature of 1580°C. The filling simulation showed that the molten metal impacted the bottom sand core mainly below three risers. I extracted the pressure distribution from the simulation and converted it into concentrated forces at the corresponding locations. These forces served as the load conditions for topology optimization.
The entire core bar frame was treated as a design domain (a cylinder with a diameter of 400 mm and a height of 200 mm). The central axis was fixed to represent the connection to the mould base. The volume constraint was set to 40% of the original volume. After 163 iterations, the optimized topology was obtained. The optimized structure featured a lattice-like pattern with thicker members along the load paths. I then designed a new core bar structure based on the optimization result, using a combination of curved bars and a central hub. The new frame weighed 27 kg, a reduction of 34.16%. The actual pouring experiment showed no damage to the sand mould or core, and a sound casting was produced.
For the guide vane casting, the maximum length was 1450 mm. The original metal core bars consisted of eight slender cylinders on both sides, weighing 28 kg. Numerical simulation indicated that the molten metal exerted pressure mainly on the side walls near the riser. The mesh size was 3 mm, with 5,668,488 elements and 1,632,252 elements for the casting. I converted the core bar assembly into a prismatic design domain of 1000 mm × 100 mm × 100 mm. The loads were applied as distributed pressures on the top and bottom surfaces, and the ends were fixed. After 53 iterations, the topology optimization yielded a new distribution. The optimized structure weighed 13 kg, a reduction of 51.24%. The pouring experiment was successful, with no sand mould or core failure, and the casting was complete.
To summarize the mass reductions and strength comparisons, I present Table 4.
| Application | Original Mass (kg) | Optimized Mass (kg) | Mass Reduction | Sand Mould/Core Integrity |
|---|---|---|---|---|
| Crown casting | 41 | 27 | 34.16% | No damage |
| Guide vane casting | 28 | 13 | 51.24% | No damage |
These results demonstrate that my topology optimization method can effectively reduce the mass of metal core bars while maintaining sufficient strength for sand mould casting. The combination of casting numerical simulation and topology optimization provides a scientific basis for core bar design, replacing empirical rules. This method contributes to lightweight, cost-effective, and sustainable sand mould casting production.
In conclusion, I proposed and validated a topology optimization method for metal core bars in sand mould casting. The method uses the SIMP model with OC update and mesh filtering. I developed a C++ program and verified it with simple examples. Sand core strength experiments showed that optimized core bars with smoothed ends perform as well as conventional bars, while reducing mass by over 59.9% in the laboratory tests. Full-scale applications in crown and guide vane castings achieved mass reductions of 34.16% and 51.24%, respectively, with successful casting outcomes. This work provides a new approach for optimizing metal core bars in sand mould casting and offers a foundation for further research on integrating temperature effects and more accurate sand core simulation.
Future work could extend the method to consider temperature-dependent material properties. In sand mould casting, the elastic modulus of the metal core bar changes with temperature during pouring and solidification. Incorporating this effect into the SIMP interpolation would yield more accurate results. Another direction is to develop a dedicated numerical simulation for sand cores to directly obtain the stress state of embedded core bars, rather than inferring from casting deformation. Additionally, the optimization of very slender core bars (aspect ratio > 20) requires further stabilization. The current program occasionally produced unstable results for such geometries, likely due to numerical issues in the finite element analysis. Improved element formulations or adaptive remeshing could address this limitation. Despite these challenges, the current method has proven effective for practical sand mould casting applications. I believe that topology optimization will become an indispensable tool for designing metal core bars and other sand mould components in the future.
To provide a complete picture of my methodology, I summarize the key parameters and their values in Table 5.
| Parameter | Value | Description |
|---|---|---|
| Penalty factor \(p\) | 3 | Exponent in SIMP interpolation |
| Move limit \(m\) | 0.2 | Maximum change in density per iteration |
| Damping coefficient \(\eta\) | 0.5 | OC update damping |
| Filter radius \(r_{\min}\) | 1.5 × element size | Mesh filtering radius |
| Volume fraction \(V^*\) | 0.4 | Target volume ratio |
| Minimum density \(x_{\min}\) | \(10^{-3}\) | Lower bound to avoid singularity |
| Convergence tolerance | \(10^{-3}\) | Change in compliance or density |
Finally, I note that the success of this method relies on accurate load identification. In sand mould casting, the loads on core bars come from the weight of the sand core, the pressure of the molten metal, and the thermal expansion. My numerical simulations captured the filling and solidification stages, but the thermal expansion effects were not fully included. Future work should incorporate thermo-mechanical coupling to obtain more realistic loads. Nevertheless, the mass reductions achieved in the crown and guide vane castings prove that the method is robust enough for industrial practice. I am confident that this approach will help foundries reduce material consumption, lower costs, and improve the sustainability of sand mould casting.
