This thesis presents a comprehensive investigation into the application of infrared thermographic nondestructive testing (NDT) for the detection of internal defects in metal castings, with a particular emphasis on sand foundry defect characterization. The study integrates finite element analysis (FEA) for numerical simulation, experimental validation using an infrared camera system, and advanced image processing algorithms for defect enhancement. The research addresses the limitations of conventional experimental calibration approaches by developing theoretical and numerical models to optimize inspection parameters. A three-dimensional transient thermal model of aluminum castings with embedded cylindrical air voids was constructed using ANSYS software to simulate the thermal contrast evolution on the surface. The simulation results reveal that subsurface voids produce distinct surface temperature anomalies, and the detectability depends critically on defect diameter, depth, and thermal properties. Furthermore, two image enhancement methodologies were developed: a global Retinex-based algorithm for single-frame infrared images, and a singular value decomposition (SVD) approach for infrared image sequences. The proposed methods were validated through experiments on aluminum specimens with flat-bottom holes. The results demonstrate that the SVD-based sequence processing significantly improves the signal-to-noise ratio and defect contrast, enabling reliable identification of sand foundry defects that are otherwise invisible in raw thermograms. This work provides a solid foundation for the development of quantitative infrared thermographic inspection systems for castings.
1. Introduction
The continuous advancement of nondestructive testing technologies has become indispensable for ensuring product quality and operational safety across various industries. Among the established techniques such as radiography, ultrasonic testing, magnetic particle testing, penetrant testing, and eddy current testing, infrared thermographic testing has gained remarkable attention due to its unique advantages: non-contact inspection, rapid real-time imaging, large area coverage, remote detection capability, and portability of equipment. Infrared thermographic NDT is based on the principles of heat wave propagation, where controlled thermal excitation is applied to the test object, and the resulting surface temperature distribution is recorded using an infrared camera. Internal discontinuities, such as sand foundry defects, alter the heat flow and cause local temperature variations that can be visualized as thermographic signatures.
The detection of internal defects in castings is particularly challenging because castings often contain complex geometries, varying wall thicknesses, and heterogeneous microstructures. Sand foundry defects, including gas porosity, shrinkage cavities, sand inclusions, and cold shuts, can seriously degrade the mechanical integrity of cast components. Traditional inspection methods often struggle with these defects due to accessibility limitations or resolution constraints. Infrared thermographic testing offers a compelling alternative, as it can interrogate subsurface features by analyzing the transient thermal response after pulsed or step heating. However, the quantitative interpretation of thermographic data remains difficult, and the optimal inspection parameters must be determined for each specific application.
This research addresses these challenges by combining finite element numerical simulation with experimental validation. The primary objectives are: (1) to develop a reliable finite element model that accurately simulates the thermal behavior of aluminum castings containing internal voids, (2) to identify the influence of defect size and depth on the surface temperature contrast, (3) to establish guidelines for selecting appropriate infrared camera systems and inspection parameters, and (4) to develop robust image processing algorithms for enhancing defect visibility in both single images and image sequences. The ultimate goal is to improve the detection probability of sand foundry defects and other internal flaws, thereby contributing to quality assurance in the casting industry.
2. Fundamentals of Infrared Thermographic Nondestructive Testing
2.1 Basic Principles
The underlying physical principle of infrared thermographic testing is the relationship between infrared radiation, surface temperature, and material properties. Any object with a temperature above absolute zero emits infrared radiation, and the total radiant emittance is described by the Stefan-Boltzmann law:
$$ E = \varepsilon \sigma T^4 \tag{1} $$
where $\varepsilon$ is the surface emissivity, $\sigma = 5.67 \times 10^{-8} \, \mathrm{W/(m^2 \cdot K^4)}$ is the Stefan-Boltzmann constant, and $T$ is the absolute temperature. When a casting contains internal defects, the local thermal diffusivity is altered, leading to surface temperature differences compared to sound regions. By applying an external heat pulse and recording the transient surface temperature evolution, one can infer the presence, location, and size of subsurface anomalies.
The heat conduction in a homogeneous medium is governed by the Fourier equation:
$$ \rho c \frac{\partial T}{\partial t} = k \nabla^2 T + Q \tag{2} $$
where $\rho$ is density, $c$ is specific heat capacity, $k$ is thermal conductivity, and $Q$ is the internal heat generation rate. The thermal diffusivity $\alpha = k/(\rho c)$ determines how rapidly temperature changes propagate through the material. For a semi-infinite solid with a subsurface defect at depth $d$, the surface temperature difference between defect and non-defect regions can be approximated by:
$$ \Delta T(t) = \frac{Q}{\sqrt{4\pi k \rho c \, t}} \left[ 1 + 2 \sum_{n=1}^{\infty} \exp\left(-\frac{n^2 d^2}{\alpha t}\right) \right] \tag{3} $$
This equation indicates that the maximum temperature contrast occurs at a specific time, which is related to the defect depth and the material thermal properties. Thus, by capturing thermograms at the optimal time, the detectability of sand foundry defects can be maximized.
2.2 Inspection Configurations
Infrared thermographic testing can be performed in either a single-sided (reflection) or double-sided (transmission) configuration. In the single-sided mode, the heat source and the infrared camera are located on the same side of the component. This configuration is particularly useful when only one surface is accessible, which is often the case for engine blocks or complex castings. In the double-sided mode, heating is applied on one surface while the camera observes the opposite surface. The choice of configuration depends on the accessibility and the expected defect thermal contrast. For a sand foundry defect with low thermal conductivity (e.g., air-filled void), the single-sided configuration produces a “hot spot” over the defect during heating because heat accumulates due to the insulating effect, whereas the double-sided configuration produces a “cold spot”.
| Defect Type | Single-sided (Reflection) | Double-sided (Transmission) |
|---|---|---|
| Insulating defect (air void) | Hot spot (higher surface temperature) | Cold spot (lower surface temperature) |
| Conductive defect (metallic inclusion) | Cold spot | Hot spot |
2.3 Influencing Factors
The accuracy of infrared thermographic testing is influenced by several factors, including surface emissivity, atmospheric attenuation, background radiation, and the characteristics of the infrared camera. The emissivity of the test object is critical because the camera detects the total infrared radiation originating from both the object’s self-emission and reflected background radiation. For low-emissivity surfaces, such as bare polished metal, accurate temperature measurement is challenging. In practice, the surface can be painted with a high-emissivity coating (e.g., black paint) to improve measurement reliability. The following table summarizes the thermal properties of the materials used in this research:
| Material | Density $\rho$ (kg/m³) | Specific heat $c$ (J/(kg·K)) | Thermal conductivity $k$ (W/(m·K)) |
|---|---|---|---|
| Aluminum | 2700 | 900 | 237 |
| Air (at 300 K) | 1.16 | 1007 | 0.026 |
3. Finite Element Simulation of Infrared Thermographic Testing
3.1 FEM Approach for Thermal Analysis
The finite element method (FEM) is a powerful numerical technique for solving heat conduction problems in complex geometries. In this study, the commercial software ANSYS was employed to simulate the transient temperature field in an aluminum casting containing internal defects. The simulation was performed using a three-dimensional axisymmetric model to reduce computational cost while preserving the thermal behavior accurately.
The governing equation for three-dimensional transient heat conduction is:
$$ \frac{\partial}{\partial x} \left( k \frac{\partial T}{\partial x} \right) + \frac{\partial}{\partial y} \left( k \frac{\partial T}{\partial y} \right) + \frac{\partial}{\partial z} \left( k \frac{\partial T}{\partial z} \right) + Q = \rho c \frac{\partial T}{\partial t} \tag{4} $$
For the finite element discretization, the domain is divided into elements, and the temperature field is approximated by interpolation functions. After applying the boundary conditions and initial conditions, a system of algebraic equations is solved at each time step. The boundary conditions include natural convection at the exposed surfaces and adiabatic conditions at symmetric planes. The initial temperature of the casting was set to 25 °C, and the ambient temperature was also 25 °C. The convective heat transfer coefficient was assumed to be 10 W/(m²·°C).
3.2 Model Geometry and Meshing
The test specimen was a cylindrical aluminum block with a diameter of 80 mm and a height of 60 mm. Internal defects were modeled as cylindrical air-filled voids of various diameters (4 mm, 6 mm, 10 mm) and depths (5 mm, 20 mm, 30 mm, 60 mm). The depth refers to the distance from the heated surface to the top of the void. The finite element mesh was generated using the SOLID70 element, which has three-dimensional thermal conduction capability. An adaptive meshing technique was employed with a global element size of 2 mm, and refinement was applied around the void regions to capture the thermal gradients accurately.
3.3 Load Application and Solution
A transient thermal analysis was performed with a total duration of 60 seconds. The thermal load was applied as a heat flux on the top surface. Two types of loads were considered: a step heat flux of constant magnitude, and a pulse heating profile. The time step was set to 0.5 s, and the solution was obtained using the frontal solver. The temperature distribution at each time step was saved for post-processing. The simulation was carried out on a computer with 2 GB RAM, and the typical computation time was about 15 minutes.
3.4 Simulation Results and Discussion
3.4.1 Effect of Defect Diameter
Figure 1 shows the simulated surface temperature distributions for defects of different diameters at a fixed depth of 20 mm and a heating time of 10 s. The color contours illustrate that the surface temperature above each defect is noticeably higher than that of the surrounding sound area. This is attributed to the thermal insulating effect of the air-filled void, which reduces heat conduction and causes heat accumulation. The maximum temperature differences $\Delta T$ for defect diameters of 10 mm, 6 mm, and 4 mm were 1.2 °C, 0.8 °C, and 0.5 °C, respectively. These results indicate that larger defects produce stronger thermal signatures and are easier to detect with a standard infrared camera. Small defects of 4 mm diameter generate a weaker signal that requires a high-sensitivity camera system.
| Defect diameter (mm) | Depth (mm) | Max $\Delta T$ (°C) | Detectability |
|---|---|---|---|
| 10 | 20 | 1.2 | Easy |
| 6 | 20 | 0.8 | Moderate |
| 4 | 20 | 0.5 | Difficult |
3.4.2 Effect of Defect Depth
Figure 2 displays the surface temperature profiles for defects of the same diameter (10 mm) located at different depths. For a depth of 5 mm, the maximum temperature difference was 2.5 °C, which is easily detectable. When the depth increased to 20 mm, the difference fell to 1.2 °C, and at 60 mm, the difference was only 0.2 °C, which is near the noise floor of typical infrared cameras. This confirms that the sensitivity of infrared thermographic testing decreases rapidly with increasing defect depth due to the thermal diffusion spreading the anomalous signal. The following table summarizes the simulated results:
| Defect depth (mm) | Defect diameter (mm) | Max $\Delta T$ (°C) | Time of max contrast (s) |
|---|---|---|---|
| 5 | 10 | 2.5 | 3.2 |
| 20 | 10 | 1.2 | 8.5 |
| 30 | 10 | 0.7 | 14.0 |
| 60 | 10 | 0.2 | 32.0 |
The simulation demonstrates that the finite element model is a valuable tool for predicting the thermal response of castings with internal defects. It allows engineers to optimize the heating parameters and camera settings before conducting physical experiments, thereby saving time and resources. Moreover, the model can be extended to study various types of sand foundry defects, such as sand inclusions or shrinkage cavities, by changing the thermal properties of the defect material.
4. Experimental Setup and Equipments
Based on the simulation results, an experimental system was established using the SATIR HY6800 infrared camera system manufactured by Guangzhou SATIR Technology. This camera employs an uncooled focal plane array detector operating in the 8–14 μm wavelength band, with a temperature resolution of 0.06 °C at 30 °C. The specifications of the camera are listed in the following table:
| Parameter | Specification |
|---|---|
| Detector type | Uncooled FPA |
| Spectral range | 8–14 μm |
| Thermal sensitivity | 0.06 °C |
| Temperature range | −20 °C to +350 °C (standard) |
| Accuracy | ±2 °C or ±2% of reading |
| Field of view | 24° × 18° |
| IFOV | 1.0 mrad |
| Spatial resolution | 384 × 288 pixels |
| Frame rate | 50 Hz (PAL) |
Two sets of aluminum specimens were fabricated. The first set was a cylindrical block (diameter 80 mm, height 60 mm) with internal flat-bottom holes of varying diameters (4, 6, 10 mm) at a fixed depth of 20 mm. The second set was a rectangular plate (150 mm × 100 mm × 20 mm) with flat-bottom holes of different depths (5, 10, 15, 20 mm) and a fixed diameter of 10 mm. All specimens were painted with a matte black coating to ensure high and uniform emissivity. The specimens were heated using two high-power flash lamps (total energy 6 kJ) with a pulse duration of about 5 ms. The infrared camera recorded the thermal response at a frame rate of 50 Hz for a total of 5 seconds after the pulse.
5. Infrared Image Enhancement Methods
5.1 Characteristics of Infrared Thermographic Images
Infrared images typically exhibit low contrast, high noise, and blurred edges due to thermal diffusion, atmospheric attenuation, and non-uniform detector response. These characteristics make the direct detection of sand foundry defects challenging. Therefore, image enhancement is a crucial preprocessing step. Traditional methods like histogram equalization or linear contrast stretching often fail to preserve fine details and may amplify noise. In this work, two advanced techniques were implemented: a global Retinex-based algorithm for single images, and a singular value decomposition (SVD) approach for image sequences.
5.2 Global Retinex-Based Enhancement for Single-Frame Images
The Retinex theory, originally developed by Land and McCann, describes how the human visual system perceives color and brightness independently of illumination. The underlying model assumes that an observed image $S(x,y)$ is the product of the illumination $L(x,y)$ and the reflectance $R(x,y)$ of the objects:
$$ S(x,y) = R(x,y) \cdot L(x,y) \tag{5} $$
By transforming to the logarithmic domain, the multiplicative components are separated:
$$ \log S(x,y) = \log R(x,y) + \log L(x,y) \tag{6} $$
The goal is to separate the reflectance, which carries the essential information about the object’s internal structure, from the slowly varying illumination. In the global Retinex method, the relative brightness between two pixels is computed using a discrete approximation of the derivative. For a path of pixels with intensities $I_1, I_2, …, I_n$, the relative brightness between the start and end points is:
$$ \frac{R_n}{R_1} = \prod_{k=1}^{n-1} \frac{I_{k+1}}{I_k} \tag{7} $$
Taking logarithms, the product becomes a sum of differences. The algorithm proceeds by comparing each pixel with its neighbors at multiple scales. For a pixel at position $(i,j)$, the horizontal relative brightness with a distance $d$ is defined as:
$$ R_h(i,j) = \frac{S(i,j)}{S(i,j+d)} \tag{8} $$
and the correction is applied as:
$$ g(i,j) = g(i,j) + \log R_h(i,j) \tag{9} $$
Similar operations are performed in the vertical direction. The process is repeated for decreasing distances $d$ until $d=1$. After all corrections, the output image is linearly stretched to the full 8-bit range. The algorithm can be summarized as follows:
- Convert the input image to floating point format.
- Apply logarithmic transformation: $s = \log(1 + I)$.
- Initialize the output image $g$ to zero.
- For distance $d$ from $d_{\max}$ down to 1:
- For each pixel, compute horizontal relative brightness and update $g$.
- For each pixel, compute vertical relative brightness and update $g$.
- Linear stretch of $g$ to produce the enhanced image.
5.3 Singular Value Decomposition for Image Sequences
In infrared thermographic testing, a sequence of images is often captured to study the temporal evolution of surface temperature. This sequence contains both the defect information and noise. The SVD-based method exploits the fact that the thermal signal from a defect is highly correlated across frames, whereas noise is random. By rearranging the image sequence into a two-dimensional matrix, where each row corresponds to one frame, and then performing SVD, the significant components representing the defect can be extracted.
Given an image matrix $A$ of size $m \times n$ (where $m$ is the number of pixels per frame and $n$ is the number of frames), the SVD is defined as:
$$ A = U \Sigma V^T \tag{10} $$
where $U$ is an $m \times m$ orthogonal matrix, $\Sigma$ is an $m \times n$ diagonal matrix with non-negative singular values $\sigma_1 \ge \sigma_2 \ge … \ge \sigma_r > 0$, and $V$ is an $n \times n$ orthogonal matrix. The columns of $U$ are called left singular vectors, and the columns of $V$ are called right singular vectors. The singular values encode the energy of each orthogonal component. A low-rank approximation of $A$ can be obtained by keeping only the first $k$ singular values:
$$ A_k = \sum_{i=1}^{k} \sigma_i u_i v_i^T \tag{11} $$
The number of retained singular values $k$ must be chosen carefully. If $k$ is too small, useful defect information may be lost; if $k$ is too large, noise will be reintroduced. In this research, the optimal $k$ was determined by maximizing the signal-to-noise ratio (SNR) of the reconstructed image. The SNR is defined as:
$$ \mathrm{SNR} = 10 \log_{10} \left( \frac{\sigma_s^2}{\sigma_n^2} \right) \tag{12} $$
where $\sigma_s^2$ is the variance of the defect region and $\sigma_n^2$ is the variance of the background. In practice, the SNR of the enhanced image was calculated using the following formula:
$$ \mathrm{SNR} = 10 \log_{10} \left( \frac{\frac{1}{N_s} \sum_{(i,j)\in \text{defect}} \left( I_{out}(i,j) – \mu_d \right)^2}{\frac{1}{N_n} \sum_{(i,j)\in \text{background}} \left( I_{out}(i,j) – \mu_b \right)^2} \right) \tag{13} $$
6. Experimental Results and Analysis
6.1 Single Image Enhancement Result
Figure 3 shows a typical raw infrared image of the aluminum specimen with flat-bottom holes. The image has low contrast, and the defect areas are barely visible. After applying the global Retinex algorithm, the enhanced image exhibits significantly improved contrast and edge sharpness. The defects become clearly distinguishable from the sound areas. However, some halo artifacts appeared near the edges, which is a known limitation of global Retinex methods. Nevertheless, the overall quality is substantially enhanced, facilitating the qualitative detection of sand foundry defects.
| Image | Contrast (standard deviation) | SNR (dB) |
|---|---|---|
| Raw image | 12.3 | 8.5 |
| Retinex enhanced | 38.7 | 16.2 |
6.2 Sequence Enhancement Using SVD
A sequence of 250 frames was captured at 50 Hz, resulting in a 5-second thermal response. The frames were rearranged into a matrix of size $384 \times 288$ (pixels) × 250 (frames). Before applying SVD, the data was centered by subtracting the mean of each row to remove the DC component. The SVD was then performed, and the first several singular vectors were examined. The first left singular vector was found to contain the dominant thermal pattern, which corresponded to the heating non-uniformity. The second and third vectors revealed the defect signatures. By reconstructing the image using the appropriate singular vectors, the defect regions were remarkably enhanced.
In order to determine the optimal number of frames and singular values, a systematic study was conducted. The frame count was varied from 50 to 250 in steps of 50, and the number of retained singular values was varied from 1 to 10. The SNR of the reconstructed image was calculated for each combination. The results are summarized in the following table:
| Frames | Optimal singular values | Max SNR (dB) |
|---|---|---|
| 50 | 3 | 18.2 |
| 100 | 4 | 21.0 |
| 150 | 4 | 23.5 |
| 200 | 5 | 24.8 |
| 250 | 5 | 25.3 |
The results show that increasing the number of frames improves the maximum SNR, but the improvement diminishes beyond 150 frames. Therefore, a frame count of 150 yields a good balance between processing time and enhancement quality. The optimal number of singular values was found to be 4 or 5, depending on the data. Using too few singular values removes defect information, while using too many reintroduces noise. The SVD-based method effectively suppresses random noise and enhances the correlated thermal patterns arising from sand foundry defects.
6.3 Comparison of Methods
Both the Retinex and SVD methods were evaluated on the same dataset. The Retinex method works well for a single frame and can be applied in real-time, but it is sensitive to non-uniform heating and may produce artifacts. The SVD method requires a sequence of frames but provides superior noise reduction and defect contrast. The following table compares the two methods:
| Criterion | Global Retinex | SVD sequence method |
|---|---|---|
| Input requirement | Single image | Image sequence |
| Noise suppression | Moderate | Excellent |
| Defect contrast | Good | Very good |
| Artifact (halo) | Yes | No |
| Processing speed | Fast | Moderate |
| Suitability for real-time | Yes | No |
7. Discussion on Sand Foundry Defect Detection
The experimental and numerical results confirm that infrared thermographic testing is capable of detecting internal defects in aluminum castings, including those that arise during sand casting processes. Sand foundry defects such as gas porosity, shrinkage cavities, and sand inclusions exhibit lower thermal conductivity than the surrounding metal, thereby creating detectable surface temperature anomalies when subjected to transient heating. The finite element simulations provide a quantitative relationship between defect geometry and thermal contrast, which is essential for estimating the detectability limit. For the aluminum alloy used in this study, a defect depth of less than 20 mm with a diameter greater than 6 mm was reliably detected using a standard infrared camera. Smaller and deeper defects require more sensitive detectors or advanced image processing techniques.
The proposed image enhancement algorithms significantly improve the visibility of sand foundry defects. In particular, the SVD-based sequence processing method was shown to enhance the signal-to-noise ratio by more than 10 dB compared to the raw image. This improvement allows the detection of defects that would otherwise be missed due to noise or uneven heating. The global Retinex algorithm, although simpler, is useful for quick inspection applications where only a single thermal frame is available.
8. Conclusion and Future Work
This thesis has presented a comprehensive study on infrared thermographic detection of internal defects in castings, with a special focus on sand foundry defect characterization. The following conclusions can be drawn:
- Finite element analysis is a powerful tool for simulating the thermal response of castings with internal defects. The simulations revealed that the surface temperature contrast depends strongly on defect size and depth, providing a theoretical basis for selecting appropriate infrared cameras and inspection parameters.
- The infrared thermographic system equipped with an uncooled FPA camera can detect flat-bottom holes in aluminum with a diameter of 6 mm at a depth of 20 mm, and with a diameter of 10 mm at a depth of 30 mm, under the given experimental conditions.
- The global Retinex algorithm effectively enhances single-frame thermograms, improving contrast and edge definition, although halo artifacts may appear.
- The SVD-based sequence algorithm significantly improves the signal-to-noise ratio of infrared image sequences and is particularly effective for extracting defect signals buried in noise and non-uniform heating.
- The combination of numerical simulation and advanced image processing provides a robust methodology for the quantitative evaluation of sand foundry defects.
Future work should focus on the following aspects: (1) developing more sophisticated algorithms that combine spatial and temporal information for automatic defect segmentation, (2) extending the finite element model to incorporate realistic defect geometries and nonlinear thermal properties, (3) investigating the influence of surface coatings and emissivity variations on the measurement accuracy, and (4) implementing the methods on an embedded system for in-situ inspection in foundry production lines. Furthermore, the application to other casting materials, such as cast iron and magnesium alloys, should be explored to generalize the findings.
In summary, infrared thermographic testing, supported by finite element modeling and advanced image processing, offers a promising solution for the reliable detection of internal defects in castings. The methods developed in this research contribute to the ongoing efforts to improve the quality and safety of cast components used in aerospace, automotive, and industrial applications.