In modern foundry engineering, the ability to quantitatively predict and analyze internal casting defects, especially shrinkage cavities and porosity, is of paramount importance. This work focuses on the development of a specialized computer-aided system that integrates finite element thermal simulation, shrinkage defect criteria, and expert system technology for the diagnosis of sand foundry defects. The system is built on a general-purpose finite element analysis platform using a three-layer architecture. It incorporates parametric design, graphical user interfaces, modular programming, and an open knowledge base structure. The primary objectives are to simulate the solidification process of castings, predict the formation and distribution of shrinkage cavities and micro-porosity, and provide defect cause analysis and corrective actions based on accumulated foundry knowledge. Two principal prediction criteria are implemented: the temperature gradient method and the Niyama criterion. In addition, the system features a defect classification scheme based on physical fields, a knowledge-based reasoning mechanism, and a maintenance interface for the continuous updating of expert knowledge. This paper discusses the theoretical foundations, system structure, development tools, implementation algorithms, and practical validation using a high-chromium cast iron component. The results demonstrate that the temperature gradient method, coupled with a carefully calibrated critical value, can accurately predict shrinkage defects in large wear-resistant castings. The expert system module effectively assists foundry engineers in identifying the root causes of defects and recommending process improvements. The developed framework significantly enhances the efficiency of casting process design and quality control, offering a practical pathway toward intelligent foundry manufacturing.
1. Introduction
The foundry industry has long been a fundamental pillar of manufacturing. However, casting processes are notoriously complex, involving simultaneous heat transfer, fluid flow, phase transformation, and stress evolution. Defects such as shrinkage cavities, porosity, hot tears, and inclusions frequently arise due to improper gating, risering, or solidification conditions. Among these, shrinkage cavities and macro-porosity account for a significant proportion of rejected castings. In the context of sand casting, these defects are commonly referred to as sand foundry defects when the root cause is related to mold filling, solidification feeding, or sand-material interaction. Traditional trial-and-error approaches for defect elimination are time-consuming and costly. The advent of computer numerical simulation has enabled foundry engineers to visualize the solidification process, compute temperature fields, and predict defect locations before physical trials are undertaken.
The work described in this dissertation aims to develop an integrated system for the prediction and analysis of shrinkage cavities and porosity. A key contribution is the seamless integration of three critical components: (a) three-dimensional solid model construction of the casting, gating, risering, and mold system; (b) finite element analysis of the transient temperature field during solidification; and (c) a knowledge-based expert system for defect classification, cause analysis, and remedial action recommendation. The system provides two well-established shrinkage criteria — the temperature gradient criterion and the Niyama criterion — to assess the feeding capability of the liquid metal and the likelihood of defect formation. Furthermore, the system incorporates a novel classification scheme based on physical fields (flow, thermal, stress), allowing the user to trace a detected defect back to its root causes in a systematic and logical manner.
Throughout the development, the focus is placed on practical usability, scientific rigor, and extensibility. The software environment is anchored by a widely used finite element analysis platform, which supplies robust solvers and pre/post-processing capabilities. Internal programming languages (parametric design language and user interface design language) are utilized to customize the interface and automate simulation workflows. External development tools, along with a relational database management system, are used to construct the knowledge base and facilitate data storage. The resulting system supports the complete workflow from computer-aided design of the casting and feeding system, to thermal simulation, defect prediction, and finally, expert-based defect analysis. The entire environment is intended to assist foundry engineers in making reliable decisions, thereby reducing scrap rates and optimizing casting processes.
In this paper, we first provide an overview of the fundamental theory behind solidification thermal analysis. Next, we describe the overall architecture of the system and the functions of its sub-modules. Then we elaborate on the development principles and technologies, including expert system implementation and software integration. After that, we present the detailed algorithms for shrinkage defect prediction, covering the temperature gradient method and the Niyama criterion. We then discuss the design of the defect analysis expert system, emphasizing the physical-field classification and the reasoning process. Finally, we validate the system using a high-chromium cast iron hammerhead casting, present the calibration of critical values, and demonstrate the process improvement achieved through the system’s recommendations.
2. Basic Theory of Temperature Field Simulation
2.1 Heat Transfer Modes
During the solidification of a casting, heat is transferred within the casting and across the mold boundaries through three fundamental mechanisms: conduction, convection, and radiation. Conduction occurs within the solid and liquid phases; convection takes place at the external mold surfaces due to air movement; radiation is significant at high temperatures but is often modeled via an equivalent convection coefficient to simplify nonlinear calculations. The governing equation for transient heat conduction in a solidifying casting is the Fourier partial differential equation:
$$ \rho c \frac{\partial T}{\partial t} = \frac{\partial}{\partial x} \left(k_x \frac{\partial T}{\partial x}\right) + \frac{\partial}{\partial y} \left(k_y \frac{\partial T}{\partial y}\right) + \frac{\partial}{\partial z} \left(k_z \frac{\partial T}{\partial z}\right) + \dot{q}_L $$
where ρ is density, c is specific heat capacity, T is temperature, t is time, k is thermal conductivity, and q̇L is the latent heat release rate per unit volume. This equation is solved using the finite element method with appropriate initial and boundary conditions. The boundary conditions include the casting–mold interface, often assumed to be perfectly contacting, and the outer mold surface where convective and radiative heat losses are combined.
2.2 Solid Fraction
For alloys that solidify over a temperature range, the solid fraction fs is a crucial parameter for the prediction of shrinkage defects. The solid fraction characterizes the mushy zone, in which liquid and solid coexist. In this work, a linear relationship between solid fraction and temperature is adopted for simplicity and computational efficiency. The linear law is written as:
$$ f_s = \frac{T_L – T}{T_L – T_S} $$
where TL and TS are the liquidus and solidus temperatures, respectively. A parabolic distribution is also possible when more accurate experimental data are available. The choice of the linear model has been successfully verified in the present system.
2.3 Phase Change and Latent Heat
The release of latent heat during solidification must be carefully treated. Several methods exist, including the temperature recovery method, the equivalent specific heat method, and the enthalpy method. The enthalpy method treats the total heat content using an enthalpy function H(T), which is integrated with the energy equation. The present system utilizes the enthalpy formulation because it is robust for alloys with a wide solidification range. The enthalpy is expressed as:
$$ H = \int_{T_0}^{T} \rho c(T) dT + \rho L f_s $$
where L is the latent heat of fusion. The finite element solver computes the temperature field at each time step, automatically accounting for the latent heat through the enthalpy–temperature curve.
3. Overall Architecture and Functional Modules of the System
The developed system, named the Casting Shrinkage Defect Prediction and Analysis System, is designed as a modular and integrated framework. It consists of four major sub-systems: (1) casting process CAD sub-system; (2) temperature field simulation sub-system; (3) shrinkage defect prediction sub-system; and (4) defect analysis expert system. The tree structure of the complete system is shown schematically in the following table, which outlines the functions of each module.
| Sub-system | Main Functions | Key Outputs |
|---|---|---|
| Casting Process CAD | Parting line, casting shrinkage allowance, machining allowance, draft angle; riser design; gating system design; mold layout; chills and sand cores | 3D solid models of casting, risers, gating, chills, cores, mold |
| Thermal Simulation | Material property database, mesh generation, initial/boundary conditions, nonlinear transient thermal analysis | Temperature field, temperature gradient field, solid fraction distribution, cooling curves |
| Defect Prediction | Temperature gradient method, Niyama criterion; flexible selection of full or partial load-step analysis | Spatial distribution of predicted shrinkage cavities and porosity, defect intensity |
| Defect Analysis Expert System | Defect classification, cause analysis, corrective actions, knowledge base maintenance, coding and query | Defect diagnosis report, process recommendations, knowledge updates |
3.1 Casting Process CAD Sub-system
The CAD sub-system assists the engineer in defining the initial casting process plan. It includes tools for determining the pouring position, selecting the parting surface, computing casting contraction, and designing risers and gating. The riser design module incorporates a database of riser types and dimensions, and it checks the feeding capacity based on the modulus method. The gating design module calculates the choke area using empirical equations and balances the cross-sections of the runner and gates. The mold design module allows the placement of chills, cores, and the selection of molding materials. All 3D models are constructed using the commercial CAD engine embedded in the finite element package, and they are automatically converted into the finite element mesh.
3.2 Temperature Field Simulation Sub-system
This sub-system is the numerical core. It manages the thermal physical properties of casting alloys and molding materials, generates finite element grids, and applies the initial and boundary conditions. The solver supports both linear and nonlinear transient heat conduction. The simulation outputs include nodal temperatures, thermal gradients, and solid fractions. These results are stored in binary arrays for subsequent post-processing. The sub-system also provides visualization capabilities, such as color contour maps of the temperature field and animated sequences, enabling the engineer to observe the solidification dynamics.
3.3 Shrinkage Defect Prediction Sub-system
This is the central innovation of the system. Two methods are implemented: the temperature gradient method and the Niyama criterion. The algorithms are programmed using the internal scripting language of the finite element platform, allowing them to directly access the temperature field results. The user can specify the liquidus temperature, solidus temperature, critical solid fraction, and the critical temperature gradient (or critical Niyama value). The system then loops through all finite elements, evaluates the solid fraction, determines the moment when the critical solid fraction is reached, and computes the ambient temperature gradient. If the gradient is below the critical threshold, or if the Niyama parameter is below the critical value, the element is flagged as a potential defect element. The final output is a graphical representation of the defect-prone zones overlaid on the casting geometry.
3.4 Defect Analysis Expert System
The expert system module is developed using external programming tools and a relational database. It contains a structured knowledge base with defect categories, causation rules, and remedial actions. The defect classification follows the international standard of the International Committee of Foundry Technical Associations (ICFTA), which organizes defects into seven main categories with four-level alphanumeric codes. In addition, the system introduces a physical-field-based classification, linking defects to the flow field, temperature field, and stress field, respectively. This classification enables a direct association with simulation results. The expert system reasoning mechanism adopts a backward-chaining approach, where the user answers a series of questions about the visible characteristics of a defect, and the system narrows down the possible identities and causes. The knowledge base is open and expandable, allowing users to add new rules and modify existing ones through a maintenance interface.
4. Development Tools and Integration Strategy
Developing a professional engineering system on top of general-purpose finite element software requires careful selection of internal and external tools. The system employs a three-layer architecture: the application layer, the tool layer, and the system layer. This architecture ensures compatibility, flexibility, and ease of expansion.
| Layer | Components | Role |
|---|---|---|
| Application Layer | Defect prediction and analysis system, user interfaces, expert system modules | Direct interaction with foundry engineers |
| Tool Layer | Finite element software, internal scripting languages, external database programming tools, relational database | Thermal analysis, custom algorithm execution, data management, knowledge representation |
| System Layer | Operating system (Windows NT/2000/XP) | Platform support |
4.1 Internal Development Tools
The internal tools are the parametric design language (PDL) and the user interface design language (UIDL) of the finite element package. PDL is a powerful scripting language similar to Fortran, which supports definition of macros, parameters, arrays, loops, and conditional branches. With PDL one can create custom macros to automate geometry creation, mesh generation, load application, and post-processing. For example, a macro can construct a rectangular solid with given dimensions, and then copy and rotate it to form a complete 3D model. Parameters can be passed in a command line, enabling parametric design of the casting and mold components. PDL also provides commands to extract nodal results from the database. For instance, the command *GET can retrieve the temperature value at node n and store it in a variable for further computations. This capability is essential for implementing the defect prediction algorithms.
UIDL, on the other hand, is dedicated to creating and modifying the graphical user interface. With UIDL, the developer can customize the main menu, create dialog boxes, and provide context-sensitive help files. This allows the system to present the user with a user-friendly, wizard-driven interface. For example, a dialog box may appear to ask for the liquidus and solidus temperatures, and the system automatically checks whether the liquidus temperature is higher than the solidus temperature. If not, a warning message is displayed. The help system is also built using UIDL, providing guidelines and explanations for each step.
4.2 External Development Tools
The external development environment is based on a visual programming language and a relational database management system. The visual programming language offers excellent database connectivity and rapid application development, making it suitable for constructing the expert system shell, the knowledge base editor, and the user-friendly query interfaces. The database stores defect descriptions, causal factors, and recommended actions. Tables are normalized to avoid redundancy and to facilitate maintenance. The database is accessed via ODBC from the application, which runs independently of the finite element package but is integrated through a button on the tool bar. This button invokes the external executable, thereby achieving a tightly coupled workflow.
5. Prediction of Shrinkage Cavities and Porosity
5.1 Mechanisms of Shrinkage Defects
Shrinkage cavities and porosity in castings are direct consequences of volume contraction during liquid cooling and solidification. Three volumetric contraction stages occur: liquid contraction, solidification contraction, and solid-state contraction. The first two stages are primarily responsible for shrinkage cavities. Liquid metal in the mold tends to feed the contracting regions. If the feeding path is interrupted, for instance by the formation of a solid skeleton or by an unfavorable thermal gradient, the remaining liquid becomes isolated and its subsequent solidification creates voids. The critical solid fraction fs,c is the parameter that describes the state at which the semi-solid slurry loses its ability to flow sufficiently. In the system, this parameter is user-defined and is usually between 0.5 and 0.8 for steel and high-chromium cast iron, based on literature and experimental data.
To mathematically quantify contraction and feeding, the liquid pressure drop in the mushy zone can be described by Darcy’s law:
$$ \frac{dP}{dx} = -\frac{\mu}{K} \frac{u}{f_l} $$
where μ is the dynamic viscosity, u is the velocity, fl is the volume fraction of liquid, and K is the permeability of the dendritic network, which strongly depends on the solid fraction. When the solid fraction rises, the permeability decreases dramatically, causing a steep pressure drop. When the pressure falls below the vapor pressure of dissolved gases, porosity nucleates and grows.
5.2 Temperature Gradient Method
The temperature gradient method is based on the empirical observation that the centerline feeding of a casting is controlled by the temperature gradient in the mushy zone. A steeper gradient leads to a larger expansion angle of the solidification front toward the riser, which keeps the feeding channel open. Conversely, a shallow gradient can cause the channel to be blocked before the complete solidification of the isolated liquid, resulting in shrinkage defects. In our implementation, the algorithm first determines the time step at which the solid fraction in a given element reaches the critical value fs,c. Then it calculates the temperature gradient G using the nodal temperature results at that instant. The criterion may be written as:
$$ \text{defect} \quad \text{if} \quad G \le G_{cr} $$
where Gcr is the critical temperature gradient. The value of Gcr depends on the casting shape, size, and alloy composition, and cannot be determined solely from theory. The system therefore proposes two calibration approaches: the forward method and the backward method. The forward method starts from a reasonable initial value, conducts simulations, compares predicted defects with experimental observations, and adjusts the critical value until agreement is reached. The backward method, by contrast, starts from an optimized process (without defects) and increases the critical value until predictions match the boundary of defect formation. In this research, the forward method was employed to calibrate a high-chromium cast iron hammerhead. After multiple iterations, the critical gradient was determined to be 0.25 °C/mm for that particular casting. This is a useful reference for future similar castings.
5.3 Niyama Criterion
The Niyama criterion, proposed by Niyama in Japan, is widely used for predicting micro-porosity in steel castings. It is defined as the ratio of the temperature gradient to the square root of the cooling rate:
$$ N = \frac{G}{\sqrt{\dot{T}}} $$
where G is the temperature gradient and Ṫ is the cooling rate at the moment when the solid fraction passes the critical value. When N falls below a critical value Ncr, the element is regarded as a potential defect site. Niyama proposed a universal critical value of 1 °C1/2·min1/2·mm−1, but subsequent studies have shown that the critical value depends on casting size, with larger castings requiring higher critical values. The system permits the user to input either a constant critical value or a value scaled according to casting modulus.
The implementation of the Niyama criterion in our system follows a similar framework to the temperature gradient method. The solver lopps through all elements, computes the cooling rate using the time derivative of temperature, and then calculates the Niyama parameter. The advantage of this criterion is that it inherently accounts for both the thermal gradient and the local solidification time. However, it is not suitable for alloys that exhibit significant graphitic expansion during solidification, such as flake and ductile irons. In our high-chromium cast iron validation, the Niyama criterion failed to correctly locate the actual defects, confirming that it should be applied only to alloys without severe volumetric expansion, or to those that solidify in a skin-forming mode.
5.4 Software Algorithm for Defect Prediction
The flowchart of the defect prediction process is given below in a step-by-step format:
- Run the finite element thermal simulation and store the temperature field for each load step.
- Prompt the user to input the liquidus temperature TL, solidus temperature TS, critical solid fraction fs,c, and either the critical temperature gradient Gcr or the critical Niyama Ncr.
- Offer two analysis modes: Intelligent Analysis (all load steps) or Load Step Analysis (a specific load step). In the former, the algorithm scans every load step and finds the one where the solid fraction first exceeds the critical value. In the latter, only the selected load step is processed.
- For each element, compute the local temperature at the critical solid fraction. Calculate the temperature gradient by finite difference interpolation of the nodal temperatures. If using the Niyama method, also compute the cooling rate by differentiating temperature with respect to time.
- Compare the computed value with the critical threshold. If the condition is satisfied, store the element number in a defect array.
- After processing all elements, display the defect elements in the post-processor using a special contour. The user can inspect the location and intensity of predicted defects.
The following pseudo-code snippet illustrates the core loop for the temperature gradient method:
ALLSEL
*DO, ielem, 1, nelem
*GET, T_e, ELEM, ielem, TEMP ! temperature at element centroid
f_s = (TL - T_e) / (TL - TS)
*IF, f_s, GE, f_sc, THEN
*GET, G_e, ELEM, ielem, TGX
*IF, G_e, LE, G_cr, THEN
ELEM_FLAW(ielem) = 1
*ENDIF
*ENDIF
*ENDDO
This algorithm is flexible and efficient. It runs within the memory of the finite element package, avoiding time-consuming disk I/O.
6. Expert System for Defect Analysis
6.1 Defect Classification
A robust defect classification system is essential for the knowledge base. The ICFTA classification divides defects into seven main categories: metallic projections, cavities, cracks, surface defects, incomplete casting, incorrect dimensions or shape, and inclusions or abnormal metallurgical structure. Each category comprises groups, sub-groups, and specific defects. A four-character code is used, for example “B211” represents a hot tear in the code structure of ICFTA. This coding system enables computer storage and retrieval. In addition to the ICFTA standard, the present work introduces a classification based on the physical field that is predominantly responsible for the defect. For instance, flow field defects include cold shuts, mistuns, and entrained oxides; temperature field defects include shrinkage cavities, macro-porosity, and micro-porosity; stress field defects include hot tearing, cold cracking, and distortion. This classification helps the user directly link simulation results to potential defects.
| Physical Field | Typical Defects | ICFTA Code Example |
|---|---|---|
| Flow Field | Cold shut, misrun, gas entrapment, slag inclusion, mold erosion | C100, D600, etc. |
| Temperature Field | Shrinkage cavity, macro-porosity, micro-porosity, hot spot, open shrinkage | A200, B100, B200 |
| Stress/Strain Field | Hot tear, cold crack, distortion, residual stress | B300, B400 |
| Coupled Flow-Thermal | Misrun due to premature freezing, cold lap | D100, D300 |
6.2 Knowledge Representation
The knowledge base uses production rules of the form:
IF conditions THEN conclusion with confidence factor CF
For example, a rule to identify a shrinkage cavity is:
- IF the defect is a cavity, shaped irregularly, either open or closed, with a rough internal surface and dendritic crystals, located in the hot spot or the last solidified region, and is large and concentrated, THEN the defect is a primary shrinkage cavity. Confidence = 0.90.
Confidence factors incorporate uncertainty into the reasoning process. When multiple rules refer to the same conclusion, the system combines their confidence factors using a deterministic uncertainty theory. The rules are editable; users can add new rules via the maintenance interface.
6.3 Reasoning Mechanism
The expert system employs a backward-chaining strategy. The user is presented with a list of defect categories and is asked to confirm observed characteristics. The system traverses the rule base, firing rules whose antecedents are satisfied. This process continues until a conclusion is reached or no further questions remain. The system then displays the defect name, its ICFTA code, possible causes, and recommended corrective actions. The causes are organized according to the manufacturing process stages: pattern design, molding materials, core making, melting, pouring, gating/risering, solidification, and post-casting operations. The user can select from a list of likely causes, and the system generates a specific recommendation. For instance, if the cause is “inadequate pouring temperature,” the recommendation might be to increase the pouring temperature by 20–30 °C while monitoring metal fluidity.
6.4 Integration with Simulation Results
A unique feature of this system is the integration between simulation and defect analysis. Once the thermal simulation predicts a defect zone, the user can click on a button to invoke the expert system and receive an explanation of why that zone is prone to defects. The system considers the thermal gradient, cooling rate, solid fraction, and the presence of isolated liquid islands. It then advises on possible modifications to the gating/risering design or the use of external chills to improve feeding. This tightly coupled loop enables rapid iteration between simulation and process improvement, greatly reducing the time required to find a sound casting process.
7. Case Study: High-Chromium Cast Iron Hammerhead
7.1 Component and Process
The validated object is a hammerhead from a crusher. The casting is a rectangular block with a central cavity, intended to operate under impact and abrasive wear. The alloy is hypoeutectic high-chromium cast iron with chemical composition close to Fe-2.8C-18Cr plus minor additions of Mn, Si, and Mo. The thickness of the wall is about 90 mm, which is considered a thick-section component. The original process design used a gating system that also acts as a riser, with the sprue providing feeding after mold filling. To promote directional solidification, four carbon-steel chills were placed at the bottom of the mold cavity, and a sand core formed the internal cavity. The mold was made from green sand, while the core and the sand bead (for clamping the core) were made from resin-bonded sand. The pouring temperature was measured around 1450 °C, and the highest recorded temperature inside the casting after filling was about 1410 °C.
The thermophysical properties used in the simulation are summarized in the table.
| Material | Density (kg/m³) | Specific Heat (J/(kg·°C)) | Thermal Conductivity (W/(m·°C)) | Enthalpy (J/m³) |
|---|---|---|---|---|
| Casting (high-Cr iron) | 7200 | 500 | 25 | 2.0e9 |
| Chill (medium-carbon steel) | 7850 | 486 | 45 | — |
| Green sand mold | 1600 | 800 | 0.8 | — |
| Resin sand core and bead | 1550 | 750 | 0.6 | — |
7.2 Thermal Simulation and Defect Prediction
The finite element model included the casting, the gating system, the chills, the sand core, and the mold. The initial temperature of the casting system was set to 1410 °C; the mold temperature was 30 °C. A heat transfer coefficient between the casting and the mold boundary was set to account for the interface resistance. A transient nonlinear thermal analysis was conducted with automatic time stepping.
The simulation results revealed the presence of hot spots in the inner corner regions of the rectangular cavity. The temperature gradient was remarkably low in those locations, indicating that feeding would be excluded from the central sections. The temperature gradient method with a calibrated critical value of Gcr = 0.25 °C/mm correctly predicted a distribution of shrinkage defects near the inner cavity corners, which matched the practical statistics from the production floor. The Niyama method, however, predicted a much broader region of porosity, with many false positives. The reason is that the high-chromium iron solidifies as a paste (mushy) alloy, and the Darcy flow resistance, not the Niyama parameter, controls defect formation. Furthermore, the cooling rate calculation in the Niyama method is sensitive to the element size and time step, leading to unreliable values. Therefore, the temperature gradient method was adopted as the primary criterion in this system for high-chromium cast iron.
This image above illustrates typical appearances of shrinkage cavities and porosity defects in ferrous castings. Such defects are common examples of sand foundry defects when they arise in sand-cast components. The cavity surfaces are often dendritic, indicating that they were formed during the final stage of solidification when the remaining liquid lost the ability to feed.
7.3 Process Optimization
Based on the defect analysis and the simulation results, the casting process was optimized. The original rectangular riser, with dimensions 210 × 130 × 150 mm, had a modulus of about 42 mm. In the revised design, the riser was changed to a square-section riser of 200 × 200 × 150 mm, which has a modulus of 47 mm while preserving the same volume. This increases the solidification time of the riser and improves its feeding efficiency. In addition, a pad (feeding aid) of 30 mm thickness was added to the wall of the casting near the riser contact. This pad increases the local heat accumulation, thereby keeping the feeding channel open for a longer time and extending the effective feeding distance of the riser.
The optimized design was simulated under the same conditions. The temperature field at the final stage of solidification showed that the riser and the sprue still contained a significant fraction of liquid, whereas the casting itself had completely solidified. The defect prediction algorithm placed no defect elements in the casting, indicating that the new process can eliminate shrinkage cavities and porosity. The improvement was validated further by the expert system module, which recommended exactly such modifications: increasing the riser modulus and adding a feeding pad on thick walls.
8. Knowledge Base Maintenance and System Extensibility
The open architecture of the system allows the knowledge base to be continuously refined. The maintenance interface consists of a set of forms where the user can browse, modify, or delete existing defect records, causes, and countermeasures. The system also supports the addition of entirely new defect categories. For example, a foundry that handles specialty alloys can add new defect names, assign ICFTA codes, and define custom reasoning rules. The database is structured as shown in the simplified entity-relationship diagram below, which is represented textually due to space constraints.
| Table Name | Fields | Purpose |
|---|---|---|
| DefectCategory | CategoryID, CategoryName, ICFTACode | Stores the hierarchical defect groups |
| DefectDetail | DefectID, DefectName, Description, Code | Describes individual defects |
| Causation | CauseID, CauseDescription, ProcessStage | Lists possible root causes for each defect |
| Remedy | RemedyID, RemedyDescription, RelatedCauseID | Provides countermeasures for each cause |
| InferenceRules | RuleID, Preconditions, Conclusion, Confidence | Production rules for backward chaining |
The user can access the maintenance forms from the main menu. This ensures that the system remains a living artifact that grows with the user’s experience. In addition, the defect coding scheme enables data exchange with other computer systems, facilitating the integration of quality management and numerical simulation across the foundry.
9. Results and Discussion
The combination of numerical simulation and expert systems provides a cost-effective methodology for reducing sand foundry defects. The temperature gradient method is particularly well suited for thick-section castings that solidify in a mushy mode. Calibration of the critical gradient is vital, and the developed forward/backward calibration approach is a practical alternative to expensive experiments. In the high-chromium cast iron example, the critical gradient was found to be 0.25 °C/mm. This value is in consistent with a few published data for similar alloys and section sizes. The system successfully predicted the defect locations in the original process and showed no defects after the recommended modifications. It should be noted that the temperature gradient method, despite its simplicity, is only an indicator of feeding efficiency. It does not quantitatively predict the volume of shrinkage cavities. For a quantitative prediction, one would need to solve the coupled equations of mass, momentum, and energy, which is computationally intensive and beyond the scope of this interactive system. Nevertheless, the qualitative information provided by the current criteria is sufficient for most engineering decision-making.
The expert system module, though limited by the initial knowledge base, proved valuable in guiding inexperienced engineers. The ability to trace a defect from its physical appearance to the underlying cause and to link the cause with the simulation results forms a powerful learning environment. Moreover, the system’s explanation mechanism makes the reasoning transparent, increasing user trust.
10. Conclusion
In this work, an integrated system for the prediction and analysis of shrinkage cavities and porosity has been designed and developed. The system is built on top of a finite element analysis platform, using internal scripting languages and external database tools. The main contributions are:
- A comprehensive and modular architecture that covers the entire workflow of casting process design, thermal simulation, defect prediction, and expert diagnosis.
- An efficient implementation of the temperature gradient method and the Niyama criterion, allowing the user to analyze all load steps or a specific load step for defect detection.
- A physical-field-based defect classification that connects numerical simulation to traditional defect knowledge, enabling traced reasoning.
- A knowledge-based defect analysis expert system with an open and maintainable structure.
- A forward calibration method for determining the critical threshold values, validated on a high-chromium cast iron hammerhead.
- The demonstration that the temperature gradient method is reliable for such alloys, whereas the Niyama criterion is not applicable due to the complex solidification behavior.
The developed system significantly reduces the time required to optimize feeding systems and eliminates many sand foundry defects. Future work will focus on extending the system to include stress analysis for hot tearing prediction, improving the quantitative prediction of shrinkage volume, and incorporating machine learning techniques to automatically refine the knowledge base from historical defect data. The continuing evolution of casting simulation and expert systems will definitely lead to the fully intelligent foundry of the future.