I have focused my research on the prediction and analysis of casting defects, especially shrinkage cavities and shrinkage porosity in castings. In my experience, casting defects are not isolated events. They emerge from the coupled effects of heat transfer, fluid flow, alloy solidification, mold reaction, and process design. Among all casting defects, shrinkage cavity and shrinkage porosity are among the most common and most damaging because they directly reduce density, pressure tightness, fatigue resistance, and service reliability. I therefore designed and developed a practical software system for casting defect prediction and casting defect analysis. My system uses a general finite element analysis platform as a secondary development environment, and I organized it through a three-layer development architecture. The system combines graphical operation, parametric design, modular functions, numerical simulation, and expert-system reasoning. Its purpose is to help foundry engineers predict casting defects, understand their causes, and improve casting process design before expensive trial pours are made.

My motivation for studying casting defects. Casting production has traditionally depended on accumulated workshop experience. That experience is valuable, but it is difficult to transfer, difficult to quantify, and sometimes unreliable when alloy grade, casting geometry, gating system, or molding method changes. When I analyzed the solidification of castings, I saw that casting defects often appear at locations where the local thermal condition prevents adequate feeding. In particular, shrinkage cavity and shrinkage porosity form when liquid and solidification contraction cannot be compensated by liquid metal from a riser, gate, or feeder. The problem becomes more severe in alloys with wide freezing ranges, in thick sections, and in components with isolated hot spots. I wanted to build a system that could connect temperature-field simulation with defect criteria and with expert knowledge, so that casting defects could be predicted and analyzed in a systematic way.
Numerical simulation and my system scope. The solidification of a casting is a transient heat-transfer problem with phase change. I treated the casting, mold, core, chill, riser, and gating system as a thermal system. I calculated the temperature field as a function of time, extracted thermal gradients and cooling rates, and then applied defect criteria to identify potential casting defects. My system includes four major subsystems: casting process design, solidification temperature-field simulation, shrinkage cavity and shrinkage porosity prediction, and casting defect analysis. These subsystems are independent in function but connected in workflow. The process design subsystem defines geometry and process parameters. The simulation subsystem computes the thermal history. The prediction subsystem applies criteria. The analysis subsystem explains why casting defects occur and suggests corrective measures.
Main physical fields and casting defects. I classify casting defects by the dominant physical field that controls their formation. This classification helps me connect simulation results to defect mechanisms. Table 1 summarizes the relationship between physical fields, typical casting defects, simulation basis, and prediction difficulty.
| Physical field | Typical casting defects | Simulation basis | Prediction difficulty |
|---|---|---|---|
| Flow field | Cold shut, entrained gas, slag inclusion, sand wash | Momentum, continuity, volume-of-fluid, turbulence | High because free surfaces and turbulence are complex |
| Temperature field | Shrinkage cavity, shrinkage porosity | Transient heat conduction, latent heat, boundary conditions | Moderate if thermal properties and criteria are reliable |
| Stress and strain field | Hot tearing, cold cracking, distortion | Thermo-elastic-plastic, creep, viscoplastic models | Very high because high-temperature behavior is uncertain |
| Coupled flow and temperature | Misrun, cold shut, incomplete filling | Coupled momentum and energy equations | High because flow and heat transfer interact strongly |
| Microstructure field | Grain-size defects, segregation, undesirable phases | Nucleation, growth, solute diffusion, stochastic models | Very high for industrial castings |
Thermal analysis as the foundation. I treat solidification simulation as a transient thermal analysis. In a closed system with no mass flow across the boundary, the first law of thermodynamics gives the energy balance. For most casting solidification problems, kinetic and potential energy changes are negligible, and no external work is considered. The energy balance then reduces to the balance between heat input, heat output, and internal energy change. A steady thermal analysis satisfies an equilibrium equation, while a transient analysis satisfies a time-dependent equation. Because solidification involves cooling, latent heat release, and changing boundary conditions, it is inherently transient.
The finite element form of the transient heat balance that I used can be written as:
$$[C]\{\dot{T}\} + [K]\{T\} = \{Q\}$$
where \([C]\) is the specific heat matrix, \([K]\) is the conduction matrix including convection and radiation effects, \(\{T\}\) is the nodal temperature vector, \(\{\dot{T}\}\) is the time derivative of temperature, and \(\{Q\}\) is the nodal heat-flow vector. For steady-state analysis, the time-dependent term vanishes and the equation becomes:
$$[K]\{T\} = \{Q\}$$
When material properties depend on temperature, boundary conditions vary with temperature, or radiation is included, the problem becomes nonlinear. I therefore used an iterative solution strategy for the nonlinear transient thermal problem.
Heat transfer modes in my casting model. Inside the casting and mold, heat transfer is dominated by conduction. At the casting-mold interface, I assumed intimate contact and no interface gap, so conduction was the main mode. At the outer mold surface, I considered convection and radiation. Because radiation analysis is highly nonlinear and expensive, I often lumped radiation into an equivalent convective boundary. Table 2 summarizes the heat-transfer modes I considered.
| Mode | Governing relation | Where I applied it | My modeling choice |
|---|---|---|---|
| Conduction | $$q = -k \nabla T$$ | Casting, mold, core, chill, riser | Solved with finite element transient heat conduction |
| Convection | $$q = h(T_s – T_f)$$ | Outer mold surface and exposed boundaries | Used constant or temperature-dependent convection coefficient |
| Radiation | $$q = \epsilon \sigma A_1 F_{12}(T_1^4 – T_2^4)$$ | Hot surfaces facing environment | Often converted to equivalent convection for efficiency |
The transient heat conduction equation with latent heat release that I solved has the following general form:
$$\rho c_p \frac{\partial T}{\partial t} = \frac{\partial}{\partial x}\left(k_x \frac{\partial T}{\partial x}\right) + \frac{\partial}{\partial y}\left(k_y \frac{\partial T}{\partial y}\right) + \frac{\partial}{\partial z}\left(k_z \frac{\partial T}{\partial z}\right) + q_{lat}$$
where \(\rho\) is density, \(c_p\) is specific heat, \(k_x\), \(k_y\), and \(k_z\) are directional thermal conductivities, and \(q_{lat}\) is the volumetric latent heat release rate. This equation is the core of my solidification temperature-field simulation.
Latent heat and solid fraction models. During solidification, latent heat is released. If I ignore latent heat, the predicted cooling curve is too fast and the hot spot is underestimated. I therefore incorporated latent heat through enthalpy, equivalent specific heat, or a temperature-recovery method. For alloys with a narrow freezing range, the temperature-recovery method is often suitable. For alloys with a wide freezing range, equivalent specific heat or enthalpy is more appropriate. In my system, I used enthalpy-based treatment in the finite element solver because it is robust for mushy-zone solidification.
The latent heat contribution can be written as:
$$\Delta Q = L \Delta f_s$$
where \(L\) is latent heat and \(\Delta f_s\) is the change in solid fraction. The effective specific heat is:
$$c_{eff} = c_p – L \frac{\partial f_s}{\partial T}$$
I also considered two solid-fraction distribution models. The linear model is:
$$f_s = \frac{T_L – T}{T_L – T_S}$$
The parabolic model is:
$$f_s = \left(\frac{T_L – T}{T_L – T_S}\right)^2$$
Here \(T_L\) is liquidus temperature and \(T_S\) is solidus temperature. In my system, I applied the linear distribution method because it is simple, stable, and sufficiently effective for the casting defects prediction problems I studied. Table 3 compares the two models.
| Model | Expression | Behavior near liquidus | Behavior near solidus | My application |
|---|---|---|---|---|
| Linear solid fraction | $$f_s = \frac{T_L – T}{T_L – T_S}$$ | Gradual increase | Gradual approach to one | Used in the prediction subsystem |
| Parabolic solid fraction | $$f_s = \left(\frac{T_L – T}{T_L – T_S}\right)^2$$ | Slow initial increase | Faster final increase | Implemented as an alternative model |
Solidification contraction and feeding resistance. Shrinkage cavity and shrinkage porosity are driven by volume contraction during liquid cooling, solidification, and solid-state cooling. If the liquid contraction and solidification contraction cannot be fed by liquid metal, a cavity or porosity forms. In my analysis, I distinguish three contraction stages: liquid contraction, solidification contraction, and solid contraction. The first two stages are most important for casting defects such as shrinkage cavity and shrinkage porosity. When the solid fraction reaches a critical value, the mushy zone becomes a network of dendrites and liquid flow becomes very difficult. This critical solid fraction is a key parameter for casting defects prediction.
For flow through a porous mushy zone, I used a Darcy-type relation:
$$\nabla P = -\frac{\mu}{K} f_L \rho g$$
where \(\mu\) is dynamic viscosity, \(K\) is permeability, \(f_L\) is liquid fraction, \(\rho\) is density, and \(g\) is gravity. The pressure loss over a distance \(\Delta x\) can be approximated as:
$$\Delta P = -\frac{\mu}{K} \frac{f_s^2}{(1-f_s)^3} v \Delta x$$
This relation shows that as solid fraction \(f_s\) increases, the resistance to feeding increases sharply. When the pressure loss becomes too large, liquid feeding stops and casting defects form. I therefore used a critical solid fraction \(f_{s,cr}\) to mark the transition from a feedable mushy zone to a non-feedable region.
My system architecture. I designed the system as a modular and expandable architecture. The process design subsystem handles casting geometry, parting plane, machining allowance, draft angle, shrinkage allowance, risers, gating, and mold design. The simulation subsystem handles material properties, mesh generation, initial conditions, boundary conditions, solution controls, and post-processing. The prediction subsystem applies shrinkage criteria. The defect analysis subsystem uses expert-system reasoning to explain causes and suggest remedies. Table 4 lists the main modules and their functions.
| Subsystem | Main function | Input | Output |
|---|---|---|---|
| Process design | Define casting, riser, gating, mold, core, and chill geometry | Part geometry, alloy, process parameters | CAD and finite element model data |
| Temperature-field simulation | Solve transient heat conduction with latent heat | Mesh, thermal properties, initial and boundary conditions | Temperature field, gradient field, cooling rate |
| Shrinkage prediction | Apply temperature-gradient and gradient-cooling-rate criteria | Temperature results, critical solid fraction, critical criterion value | Potential casting defects locations and severity |
| Defect analysis | Query, reason, explain, and suggest remedies | Defect type, process route, user selections | Causes, countermeasures, knowledge-base updates |
Three-layer development architecture. I used a three-layer structure: system layer, tool layer, and application layer. The system layer includes the operating system and the general finite element platform. The tool layer includes internal development tools, external development tools, databases, and knowledge bases. The application layer is the casting defect prediction and analysis system itself. This structure improves compatibility, openness, and maintainability. It also allows me to extend the system without modifying the finite element solver kernel.
Internal development tools. I used the parametric design language and the user-interface design language provided by the finite element platform. With the parametric design language, I defined macros, parameters, arrays, loops, branches, and custom commands. This allowed me to automate mesh generation, material property assignment, load-step control, result extraction, and criterion evaluation. With the user-interface design language, I created menus, dialogs, prompts, warnings, and help files. These tools made the system graphical, parametric, and user-oriented.
For example, I used macro constructs to define geometry, define parameters, pass arguments, and perform Boolean operations. I used array parameters to store nodal temperatures, element temperatures, solid fractions, gradients, and cooling rates. I used looping and branching to scan elements over load steps and substeps. This is essential because casting defects prediction requires time-dependent evaluation of thermal conditions.
External development tools and databases. I used an object-oriented programming environment for the expert-system interface and database operations. I selected a relational database as the back-end repository for knowledge, defect classifications, material data, and process recommendations. The external program communicates with the finite element environment through tool buttons and data exchange. This integration lets the user move from simulation to defect analysis without leaving the working environment.
Expert-system principles in my casting defect analysis. Casting defect analysis often cannot be reduced to a single mathematical formula. It depends on knowledge, experience, and heuristics. I therefore incorporated expert-system concepts into my system. The knowledge base stores rules about casting defects. The inference engine applies forward reasoning, backward reasoning, or mixed reasoning. The explanation mechanism answers why and how questions. The knowledge-acquisition mechanism allows new knowledge to be added. The working database stores facts and intermediate conclusions. Table 5 summarizes the expert-system components.
| Component | Role in my system | Example content |
|---|---|---|
| Knowledge base | Stores rules and relationships for casting defects | If a cavity is irregular, rough-walled, and located at a hot spot, then it is likely a shrinkage cavity |
| Working database | Stores current facts and reasoning state | Defect shape, location, alloy, molding method, process parameters |
| Inference engine | Selects and applies rules | Backward chaining from defect name to causes |
| Explanation mechanism | Explains conclusions and rule use | Shows why a cause was selected |
| Knowledge acquisition | Adds or modifies knowledge | Insert new defect rule or update countermeasure |
| User interface | Communicates with the engineer | Menus, dialogs, query forms, help texts |
Prediction criteria for shrinkage cavity and shrinkage porosity. I implemented multiple criteria. The temperature-gradient criterion is simple and effective. The gradient-cooling-rate criterion, often called the Niyama-type criterion, considers both temperature gradient and cooling rate. I also considered solid-fraction-based criteria, time-based criteria, and direct feeding-resistance criteria. Table 6 gives the main criteria I examined.
| Criterion | Condition for casting defects | Main variables | My assessment |
|---|---|---|---|
| Temperature-gradient criterion | $$|\nabla T| \le C_{cr}$$ when $$f_s \ge f_{s,cr}$$ | Temperature gradient, critical solid fraction | Simple and practical; critical value depends on geometry and alloy |
| Gradient-cooling-rate criterion | $$\frac{G}{\sqrt{R}} \le C_{N,cr}$$ | Gradient \(G\), cooling rate \(R\) | Better for porosity but sensitive to mesh size and alloy type |
| Solid-fraction gradient criterion | $$\left|\frac{\partial f_s}{\partial x}\right| \le C_{f,cr}$$ | Solid fraction gradient | Useful for eutectic alloys; equivalent to thermal gradient when temperature and solid fraction are one-to-one |
| Critical solid-fraction time criterion | Late arrival at \(f_{s,cr}\) | Time to reach critical solid fraction | Qualitative but intuitive |
| Feeding-resistance criterion | Pressure loss exceeds critical value | Permeability, viscosity, flow velocity | Physically meaningful but requires permeability data |
Temperature-gradient method implementation. My implementation evaluates each element when its solid fraction reaches a critical value. For each load step or substep, I extract the temperature field and compute the element temperature. I calculate the solid fraction using the linear model. If the solid fraction is at or above the critical value, I compute the temperature gradient. If the gradient is below the critical gradient, I mark the element as a potential casting defect. I store the element number in a defect array. After all elements are checked, I display the defect locations and severity. The user can choose to analyze the whole load-step history or only a selected load step.
The core condition I used is:
$$f_s \ge f_{s,cr} \quad \text{and} \quad |\nabla T| \le C_{cr}$$
where \(f_{s,cr}\) is the critical solid fraction and \(C_{cr}\) is the critical temperature gradient. The first condition ensures that the local region has lost most of its feeding capability. The second condition indicates that the thermal gradient is too weak to maintain a feeding channel.
Gradient-cooling-rate method implementation. For the gradient-cooling-rate criterion, I calculate both the temperature gradient and the cooling rate for each element. The cooling rate is obtained from the change in temperature over a time step:
$$R = \left|\frac{\partial T}{\partial t}\right|$$
The criterion is:
$$\frac{G}{\sqrt{R}} \le C_{N,cr}$$
I implemented this criterion in a flexible way. The user can scan the entire simulation history or restrict the analysis to a specific load step. The system then stores potential casting defects and displays them. I found that this criterion is effective for some alloy systems but not universally applicable. In particular, for high-chromium cast iron, graphite and austenite formation make the volume change behavior much more complex than in steel. Therefore, I treat the gradient-cooling-rate criterion as an alternative tool rather than a universal solution.
Critical value determination. The accuracy of casting defects prediction depends strongly on the critical values. I proposed two methods to determine them: the forward method and the reverse method. In the forward method, I start with an initial critical value, run the simulation, predict casting defects, compare with experimental or production results, and adjust the value until agreement is acceptable. In the reverse method, I start from an optimized process or known defect-free casting, assume a critical value, and adjust it until the prediction matches the known result. Table 7 compares these methods.
| Method | Starting point | Procedure | Advantage | Limitation |
|---|---|---|---|---|
| Forward method | Original process and initial critical value | Simulate, predict, compare, adjust | Directly calibrated to production or experiment | Requires reliable defect observations |
| Reverse method | Optimized or defect-free process | Assume critical value, simulate, adjust until no defects | Useful when an optimized process is known | May not capture defect formation fully |
Using the forward method, I determined a critical temperature-gradient value for a high-chromium cast iron component. That critical value is not universal. It depends on alloy composition, casting section size, mold material, and the definition of the gradient. However, it provides a practical reference for similar high-chromium cast iron castings and for further optimization.
Defect classification for analysis. Correct classification is the first step in casting defect analysis. I studied three classification approaches: appearance-based classification, mechanism-based classification, and physical-field-based classification. For my system, I used a physical-field-based classification as the bridge between simulation and expert reasoning. This allows me to connect a predicted casting defect to a physical cause and then to a process remedy. Table 8 shows the physical-field classification I adopted.
| Physical field | Casting defects | Main simulation output | Typical remedy |
|---|---|---|---|
| Flow field | Cold shut, entrained gas, slag inclusion, sand wash | Velocity, pressure, free surface, filling time | Modify gating, reduce turbulence, improve venting |
| Temperature field | Shrinkage cavity, shrinkage porosity | Temperature, gradient, cooling rate, solid fraction | Add riser, chill, padding, or insulation |
| Stress field | Hot tearing, cold cracking, distortion | Stress, strain, displacement | Improve constraint, revise mold collapsibility, change alloy |
| Coupled fields | Misrun, cold shut | Coupled temperature and velocity | Increase pouring temperature, redesign gating |
Reasoning flow in my defect analysis subsystem. My defect analysis subsystem asks the user to select the defect category, defect name, casting alloy, molding method, and process route. It then follows a reasoning path:
First, I identify the defect type from its morphological features and classification code. Second, I match the defect to possible causes stored in the knowledge base. Third, I filter causes by the selected process route and alloy system. Fourth, I rank causes by confidence and relevance. Fifth, I present countermeasures and suggestions. Finally, I allow the user to update the knowledge base if new experience is obtained.
For shrinkage cavity, a typical rule can be expressed in first-person reasoning as follows: if the defect is an irregular closed or exposed cavity, if the wall is rough and dendritic, if it occurs at a hot spot or last-solidifying region, and if it is large and concentrated, then I classify it as a shrinkage cavity with high confidence. For shrinkage porosity, the reasoning focuses on dispersed small voids, interdendritic location, and a wide mushy zone.
Case study: high-chromium cast iron casting. I applied the system to a high-chromium cast iron component used under impact wear. The component has a thick rectangular body with an internal cavity. The alloy is hypoeutectic and solidifies over a wide temperature range. The original process used a three-box molding arrangement, clay-bonded sand for the mold, resin-bonded sand for the core and padding, carbon-steel chills in the lower box, and a riser in the upper box. The gating system also acted as a feeding path. The main quality risk was shrinkage cavity and shrinkage porosity at the inner cavity corners and thick walls.
I measured and analyzed the thermal behavior. The pouring temperature was high, but the measured maximum temperature in the mold system was lower because of heat loss during filling. I set the initial temperature of the casting and gating system to a high value and the mold, core, and chills to ambient temperature. The boundary condition at the outer mold surface was equivalent convection, including the effect of radiation. Table 9 lists the thermo-physical parameters I used.
| Material | Density | Specific heat | Thermal conductivity | Latent heat |
|---|---|---|---|---|
| High-chromium cast iron | High | Temperature dependent | Temperature dependent | Included through enthalpy |
| Carbon-steel chill | High | Temperature dependent | Temperature dependent | Not dominant |
| Clay-bonded sand | Low | Moderate | Low | Not dominant |
| Resin-bonded sand | Low | Moderate | Low | Not dominant |
Simulation results and casting defects prediction. The temperature-field simulation showed that the inner cavity corners and thick-wall regions remained hot while the thinner sections cooled faster. The temperature gradient in those hot regions was low. The feeding channel from the riser became blocked before the hot spot fully solidified. This is a classic condition for casting defects such as shrinkage cavity and shrinkage porosity.
I applied the temperature-gradient criterion first. The predicted casting defects appeared at the inner cavity corners and in the thick wall, which matched production statistics. The simulation also showed a primary shrinkage cavity in the riser and secondary shrinkage below it. The temperature-gradient method produced a reasonable match with observed casting defects. I then applied the gradient-cooling-rate criterion. The result did not match the high-chromium cast iron behavior well. I concluded that the gradient-cooling-rate criterion, although successful in some steels, cannot be directly transferred to high-chromium cast iron because graphite and austenite formation cause complex volume changes. Table 10 compares the predictions.
| Criterion | Predicted location | Agreement with observed casting defects | My conclusion |
|---|---|---|---|
| Temperature-gradient criterion | Inner cavity corners, thick walls, riser neck | Good | Suitable for the high-chromium cast iron case after calibration |
| Gradient-cooling-rate criterion | Scattered regions, some incorrect locations | Poor | Not directly applicable to high-chromium cast iron |
| Solid-fraction visual inspection | Isolated mushy zones | Qualitative agreement | Useful as a supporting method |
Process optimization. Based on the prediction and analysis, I proposed process improvements. The original riser had a rectangular shape. For the same volume, a square or cylindrical riser has a larger modulus, so it remains liquid longer and feeds more effectively. I changed the riser to a square shape with the same volume. I also added padding near the riser to increase the effective feeding distance and maintain a favorable expansion angle toward the riser. The chills were retained to promote directional solidification. Table 11 summarizes the original and optimized process.
| Process feature | Original design | Optimized design | Reason for change |
|---|---|---|---|
| Riser shape | Rectangular | Square | Increase modulus at the same volume |
| Padding | None | Added near thick walls | Improve feeding path and expansion angle |
| Chills | Carbon-steel chills | Retained and positioned to promote directional solidification | Increase local cooling and reduce hot spots |
| Core and mold | Resin-bonded core, clay-bonded mold | Same materials with revised geometry | Maintain process compatibility |
| Gating | Gating system also served as feeder | Revised to support riser feeding | Improve feeding efficiency |
After optimization, I ran the solidification simulation again. At the time when the casting interior was fully solidified, the riser and gate still contained high-temperature regions. This means that feeding could continue longer and compensate for contraction in the casting. The defect prediction array contained no defect elements for the optimized process. This indicated that the optimized process was better than the original process. Table 12 summarizes the predicted outcomes.
| Process | Hot spot severity | Feeding channel condition | Predicted casting defects | My assessment |
|---|---|---|---|---|
| Original | Strong at inner corners and thick walls | Blocked before final solidification | Shrinkage cavity and shrinkage porosity likely | Needs improvement |
| Optimized | Reduced by chills and padding | Maintained longer by larger-modulus riser | No defect elements in selected load step | Improved process |
What I learned about casting defects prediction. My work shows that casting defects prediction is not only a numerical problem. It is a knowledge-intensive engineering problem. A temperature field alone does not identify casting defects. A criterion alone does not guarantee accuracy. A knowledge base alone cannot replace physics. The best result comes from integrating numerical simulation, physically meaningful criteria, experimental calibration, and expert reasoning. I therefore designed my system to be open and extensible. The material database, process database, knowledge base, and rule base can all be updated. This is important because casting defects depend on alloy, geometry, mold, gating, pouring, and many other variables.
Why temperature-gradient calibration matters. The critical temperature gradient is not a universal constant. In my high-chromium cast iron case, I determined it through the forward method. I started with an initial value, predicted casting defects, compared with observed shrinkage cavity and shrinkage porosity, and adjusted the value until the prediction matched the production result. This calibrated value is useful for similar castings, but it should be re-evaluated for different section sizes and alloy compositions. The same caution applies to the critical solid fraction and to the gradient-cooling-rate constant. Table 13 lists the key calibration parameters in my system.
| Parameter | Meaning | How I determined it | Main influence on casting defects |
|---|---|---|---|
| Liquidus temperature | Start of solidification | Alloy database or thermal analysis | Defines start of latent heat and solid fraction |
| Solidus temperature | End of solidification | Alloy database or thermal analysis | Defines end of feeding |
| Critical solid fraction | Loss of feeding capability | Literature, experiment, or calibration | Determines when a region is evaluated for defects |
| Critical temperature gradient | Threshold for shrinkage cavity and porosity | Forward or reverse calibration | Controls defect classification |
| Critical gradient-cooling-rate value | Threshold for porosity | Literature or calibration | Controls porosity prediction |
| Heat-transfer coefficient | Boundary heat loss | Experiment or inverse estimation | Controls cooling rate and hot spot |
Software integration and user experience. I wanted the system to be usable by foundry engineers, not only by simulation specialists. Therefore, I used graphical menus, dialogs, parameter input forms, automatic prompts, warning messages, and embedded help. The user can select a prediction method, input liquidus and solidus temperatures, set critical solid fraction, set critical gradient or gradient-cooling-rate value, choose a load-step range, and run the prediction. The system then displays the result on the temperature field or gradient field. The user can also switch to the defect analysis subsystem to query causes and remedies. This integration reduces the gap between simulation and process decision-making.
Knowledge representation. I represented knowledge in production rules. A general rule has the form: if premise then conclusion with confidence. For casting defects, a rule can include multiple conditions joined by logical operators. Confidence factors allow the system to handle uncertainty. For example, a defect might be classified as shrinkage porosity if several conditions are satisfied, but the confidence may be lower if the location or shape is atypical. I also allowed the knowledge base to be edited outside the program. This makes the system maintainable and allows new foundry experience to be added without reprogramming.
Explanation and maintenance. The explanation mechanism helps the user understand why a conclusion was reached. It can show which rules were used, which facts were considered, and how confidence was combined. The knowledge-maintenance module allows the user to add, modify, or delete defect names, causes, countermeasures, and classification codes. This is important because casting defects terminology and process experience evolve over time. The system is therefore not a closed product but a growing knowledge platform.
Limitations I observed. My system has several limitations. First, the temperature-field simulation assumes that the mold is instantaneously filled or that filling effects are secondary. For thin-wall castings or fast-cooling processes, coupled flow and heat transfer is necessary. Second, the defect criteria are semi-empirical. They require calibration. Third, the gradient-cooling-rate criterion is not reliable for high-chromium cast iron because graphite expansion and austenite contraction interact. Fourth, the expert system covers a limited portion of the full casting defect spectrum. Fifth, stress-related casting defects such as hot tearing and cold cracking are not fully integrated. These limitations define the direction for future work.
Future development directions. I believe the next steps should include more prediction criteria, automatic critical-value calibration, coupling with flow and stress fields, microstructure-informed defect criteria, and a larger knowledge base. I also recommend improving the user interface with a type tree, parameter input area, schematic area, text explanation area, and online help. The system should remain parametric, graphical, and modular. It should be able to handle sand casting, investment casting, die casting, low-pressure casting, and other processes. It should also be able to exchange data with CAD and process-planning systems.
Final reflection. My work on casting defects prediction and analysis has convinced me that the most effective approach is integration. I integrate thermal theory with numerical simulation, simulation with defect criteria, criteria with calibration, and calibration with expert knowledge. This integration allows foundry engineers to move from experience-based trial and error toward science-based process design. The system I developed is not the end of the journey. It is a framework that can grow. As new alloys, new processes, and new simulation methods appear, the framework can be extended. The goal remains the same: reduce casting defects, improve casting quality, shorten development cycles, and lower production cost.
Summary of my main conclusions. I designed a casting defect prediction and analysis system using a general finite element platform as the secondary development base. I used a three-layer architecture and modular, parametric, graphical development methods. I implemented solidification temperature-field simulation with latent heat and solid-fraction models. I implemented temperature-gradient and gradient-cooling-rate criteria for shrinkage cavity and shrinkage porosity. I proposed forward and reverse methods for determining critical values. I built a defect analysis subsystem with expert-system reasoning, knowledge representation, explanation, and maintenance. I applied the system to a high-chromium cast iron component. The temperature-gradient method matched observed casting defects after calibration, while the gradient-cooling-rate method was not suitable for that alloy. The optimized process with a larger-modulus riser and padding reduced the predicted casting defects. These results show that the system is useful, extensible, and practically relevant for casting defects control.
