In my work, I focus on the detection of internal casting defects using infrared thermal-wave testing. I treat casting defects as hidden discontinuities that disturb heat flow and therefore change the surface temperature field in a measurable way. My objective is to combine heat-transfer theory, finite element simulation, experimental infrared imaging, and image enhancement so that internal casting defects can be located, compared, and evaluated with better reliability. I am specifically interested in casting defects such as gas holes, shrinkage cavities, porosity, inclusions, and flat-bottom artificial defects that mimic internal voids in metal castings. Because casting defects can reduce load-bearing capacity, fatigue life, pressure tightness, and dimensional stability, I consider their non-destructive evaluation to be a critical part of quality assurance.
I begin from a simple physical premise: when I apply a controlled thermal stimulus to a cast component, the heat diffuses through the material. If the material is homogeneous and free of casting defects, the surface temperature evolves in a relatively smooth and predictable manner. If internal casting defects are present, however, the local thermal properties or the local geometry of the heat path is altered. A defect filled with air, for example, has much lower thermal conductivity and much lower volumetric heat capacity than aluminum. As a result, heat accumulates above or below the casting defects depending on the viewing configuration. I observe this accumulation as a local temperature contrast on the surface. This contrast is the raw signature that I use for detection.
In my experimental and numerical studies, I use a cylindrical or block-shaped aluminum specimen because aluminum casting defects are common in engine blocks, housings, and structural components. I also use flat-bottom holes of different diameters and depths to represent internal casting defects in a controlled manner. I blacken the inspected surface to increase and stabilize emissivity. I then heat the surface with a controlled source and record the cooling or heating sequence with an infrared camera. The resulting thermal images are not simple photographs of temperature; they are indirect projections of internal casting defects through the diffusion process. This indirect nature is exactly why I need both simulation and enhancement algorithms.
My Motivation and Scope for Detecting Casting Defects
I have found that conventional non-destructive testing methods each have strengths and limitations when applied to internal casting defects. Radiography can reveal density changes, but it may miss tight cracks or defects oriented unfavorably. Ultrasonic testing is sensitive to internal reflectors, but it requires coupling and can be challenging for complex cast geometries. Magnetic particle and penetrant testing are mainly surface methods and therefore are not directly suited to deep internal casting defects. Eddy current testing is effective for conductive surface and near-surface defects, but its depth penetration is limited. Infrared thermal-wave testing attracts me because it is non-contact, rapid, full-field, and capable of inspecting large areas. I can also move the equipment to the component rather than moving the component to a fixed inspection station.
| Method | Primary Principle | Contact Requirement | Typical Strength | Limitation for Internal Casting Defects |
|---|---|---|---|---|
| Radiography | Differential attenuation of penetrating radiation | Usually non-contact | Good volumetric sensitivity | Safety requirements; orientation-dependent for some casting defects |
| Ultrasonic testing | Elastic wave reflection and scattering | Couplant often needed | Good depth resolution | Geometry and coupling sensitivity; difficult for complex cast surfaces |
| Magnetic particle testing | Magnetic flux leakage | Contact | Excellent for surface and near-surface defects | Ferromagnetic materials only; not for deep internal casting defects |
| Penetrant testing | Capillary action of dye | Contact | Simple and visual | Surface-breaking defects only |
| Eddy current testing | Electromagnetic induction | Close proximity | Fast for conductive surfaces | Limited depth; mainly surface and near-surface casting defects |
| Infrared thermal-wave testing | Heat diffusion and surface temperature contrast | Non-contact | Full-field, rapid, movable | Depth inversion is indirect; image processing is essential |
In my view, the most important challenge is not simply to obtain a thermal image, but to make internal casting defects visible and quantifiable. Thermal diffusion acts as a low-pass filter. It blurs sharp defect boundaries, reduces high-frequency spatial information, and lowers contrast. Noise from the camera, environmental reflections, non-uniform heating, and emissivity variation further complicates the interpretation. Therefore, my research combines three layers: first, I model the thermal physics; second, I simulate the inspection to choose parameters; third, I enhance the thermal images or image sequences to reveal casting defects more clearly.
Thermal-Wave Physics for Casting Defects
I use the Stefan-Boltzmann law as the starting point for relating surface temperature to infrared radiation. For a real surface with emissivity epsilon, the total emitted radiant exitance is
$$M = \epsilon \sigma T^4$$
where sigma is the Stefan-Boltzmann constant and T is the absolute temperature. In my infrared measurements, I do not measure temperature directly. I measure a camera signal that depends on emitted radiation, reflected radiation, atmospheric transmission, and the spectral response of the detector. For a surface with emissivity less than one, the apparent temperature is influenced by the surrounding environment. I therefore increase surface emissivity by blackening the specimen, and I keep the environment as stable as possible.
The main governing equation I use for internal casting defects is the transient heat conduction equation. For an isotropic material with thermal conductivity k, density rho, specific heat capacity c_p, and internal heat generation q, I write
$$\rho c_p \frac{\partial T}{\partial t} = \nabla \cdot (k \nabla T) + q$$
In my simulations, I usually set the internal heat generation to zero because the heating is applied externally as a boundary heat flux or convection. The diffusion of heat in the casting is then controlled by the material properties and the boundary conditions. For one-dimensional heat flow, the Fourier law is
$$q_x = -k \frac{\partial T}{\partial x}$$
At the surface, I apply convective and radiative boundary conditions. The convective heat flux is represented by Newton’s law of cooling:
$$q” = h(T_s – T_\infty)$$
where h is the convective heat transfer coefficient, T_s is the surface temperature, and T_infinity is the ambient temperature. The net radiative heat flux between the surface and its surroundings is
$$q”_{\mathrm{rad}} = \epsilon \sigma (T_s^4 – T_{\mathrm{sur}}^4)$$
For the purposes of my finite element model, I combine convection and radiation into an effective boundary condition when the temperature differences are moderate, or I include radiation explicitly when the surface temperature becomes high. The thermal diffusion coefficient is a convenient parameter because it determines how quickly a material responds to a thermal disturbance:
$$\alpha = \frac{k}{\rho c_p}$$
This parameter is especially important for internal casting defects. A defect filled with air has an effective thermal diffusion coefficient that is very different from that of the surrounding metal. The thermal-wave concept allows me to describe the periodic temperature field produced by a modulated heat source. I write a one-dimensional thermal wave as
$$T(z,t) = T_0 + \Delta T \exp\left(-\frac{z}{\mu}\right)\cos\left(\omega t – \frac{z}{\mu}\right)$$
where z is depth, omega is angular frequency, and mu is the thermal diffusion length. The diffusion length is
$$\mu = \sqrt{\frac{2\alpha}{\omega}} = \sqrt{\frac{\alpha}{\pi f}}$$
This equation tells me that higher modulation frequencies confine the thermal wave closer to the surface, while lower frequencies penetrate deeper. For internal casting defects located deeper below the surface, I must use a longer thermal wave or observe the temperature evolution at later times. The peak contrast time is often approximated by
$$t_{\mathrm{peak}} \approx \frac{d^2}{\alpha}$$
where d is the depth of the casting defect below the inspected surface. Equivalently, I estimate the depth from the observed peak time by
$$d \approx \sqrt{\alpha t_{\mathrm{peak}}}$$
The temperature contrast produced by a casting defect is the difference between the defect region and the sound region:
$$\Delta T = T_{\mathrm{defect}} – T_{\mathrm{sound}}$$
I use this contrast as a primary indicator. In my simulations, I observe that the contrast depends on the size, depth, shape, and filling material of the casting defect. It also depends on the heating duration, the boundary conditions, and the thermal properties of the casting material.
Finite Element Modeling of Internal Casting Defects
I use finite element analysis because it can handle complex geometry, arbitrary boundary conditions, and temperature-dependent material properties. The basic idea is to discretize the continuous domain into finite elements, approximate the temperature field within each element, and assemble the element equations into a global system. For transient heat conduction, the finite element formulation leads to a system of ordinary differential equations in time:
$$[C]\{\dot{T}\} + [K]\{T\} = \{F\}$$
where [C] is the capacitance matrix, [K] is the conductivity matrix, {T} is the nodal temperature vector, and {F} is the load vector. I solve this system using an implicit time integration scheme because it is stable for the long cooling sequences I study. I use a general-purpose finite element package to build the model, assign material properties, apply boundary conditions, mesh the geometry, and post-process the temperature field.
In my model, I represent a cast aluminum component as a cylindrical or rectangular domain. I introduce internal casting defects as flat-bottom holes filled with air. I assume that the surrounding material is homogeneous and isotropic. I also assume that the lower and side boundaries are insulated or maintained at ambient temperature. The top surface exchanges heat with the environment by convection and radiation. Initially, I set the entire specimen to a uniform temperature. Then I apply a heat flux or a sudden convective heating condition to the inspected surface. I record the temperature field at each time step.
| Material | Density (kg/m3) | Specific Heat (J/kg K) | Thermal Conductivity (W/m K) | Thermal Diffusivity (m2/s) |
|---|---|---|---|---|
| Aluminum casting | 2700 | 900 | 237 | 9.75e-5 |
| Air inside casting defects | 1.2 | 1005 | 0.026 | 2.16e-5 |
| Black coating | 1200 | 1000 | 0.5 | 4.17e-7 |
I simplify the geometry in a way that preserves the essential thermal behavior. I use a three-dimensional thermal element with one degree of freedom per node, namely temperature. I apply adaptive meshing so that the mesh is finer near the casting defects and coarser in the far field. This reduces the computational cost while maintaining accuracy around the defect boundaries. I set the time step small enough to capture the early temperature evolution, because the most useful defect contrast often appears during the transition from heating to cooling.
| Parameter | Value Used in My Simulation | Reason |
|---|---|---|
| Initial temperature | 20 degrees C | Uniform initial condition |
| Ambient temperature | 20 degrees C | Natural laboratory environment |
| Convection coefficient | 10 W/m2 K | Natural air convection |
| Surface emissivity | 0.90 after blackening | Higher and more stable infrared signal |
| Heating pulse duration | 1 to 5 s | Controlled thermal excitation |
| Total observation time | 60 s | Capture cooling contrast |
| Defect diameters | 2, 4, 6, 8, 10, 12 mm | Size effect study |
| Defect depths | 1, 2, 3, 5, 8, 10 mm | Depth effect study |
I model several cases to understand how casting defects influence the surface temperature field. The first group varies the diameter while keeping the depth constant. The second group varies the depth while keeping the diameter constant. In each case, I extract the surface temperature above the defect and the surface temperature above a sound region. I then compute the contrast, the peak time, and the spatial extent of the thermal anomaly. These quantities help me determine the detection limits of my infrared system.
My Numerical Simulation Results for Casting Defects
I observe that after heating, the region above an internal casting defect cools more slowly than the sound region when the defect is filled with air. This is because the air-filled casting defect acts as a thermal barrier. Heat cannot easily diffuse downward through the defect, so more heat remains near the surface. In the cooling phase, the surface above the casting defect therefore remains warmer than the surrounding sound area. This produces a positive thermal contrast. The situation can reverse if the defect has higher thermal conductivity than the matrix, but for the air-filled casting defects I study, the contrast is usually positive during cooling.
For different defect diameters, I find that larger casting defects produce larger temperature contrasts and larger spatial anomalies. A larger defect interrupts a larger cross-section of the heat flow, and its thermal signature is easier for the infrared camera to resolve. Small casting defects, especially those below a few millimeters in diameter, produce weak contrast and require better camera sensitivity, better surface preparation, and more effective image enhancement.
| Defect Diameter (mm) | Depth (mm) | Maximum Temperature Contrast (degrees C) | Relative Visibility |
|---|---|---|---|
| 2 | 2 | 0.35 | Very low |
| 4 | 2 | 0.70 | Low |
| 6 | 2 | 1.15 | Moderate |
| 8 | 2 | 1.65 | Good |
| 10 | 2 | 2.20 | Good |
| 12 | 2 | 2.80 | Very good |
For different defect depths, I find that shallower casting defects are much easier to detect. The thermal wave is attenuated as it propagates through the material, and the temperature contrast decreases with depth. When the casting defect is deeper than a few millimeters, the surface anomaly becomes broad and weak. I also observe that the peak contrast time increases with depth. This is consistent with the diffusion relation between depth and time. Therefore, I can use the time of maximum contrast as an indicator of defect depth, although the inversion is not unique without additional constraints.
| Defect Depth (mm) | Diameter (mm) | Maximum Temperature Contrast (degrees C) | Peak Contrast Time (s) |
|---|---|---|---|
| 1 | 8 | 2.40 | 0.8 |
| 2 | 8 | 1.65 | 1.5 |
| 3 | 8 | 1.10 | 2.4 |
| 5 | 8 | 0.62 | 4.2 |
| 8 | 8 | 0.28 | 7.5 |
| 10 | 8 | 0.15 | 11.0 |
I also study the effect of the filling material. An air-filled casting defect produces a stronger thermal barrier than a defect filled with a resin or a lower-conductivity solid. If the defect is partially filled or if it has contact with the surrounding metal, the thermal contrast is reduced. This is important because real casting defects are rarely ideal voids. They may contain oxide films, gas, shrinkage porosity, or a mixture of air and metal. In my view, the finite element model is still useful because it gives a first-order estimate of the contrast and helps me choose the heating time, observation window, and camera settings before I run experiments.
Experimental Verification and Thermal Image Characteristics
For experimental verification, I use a commercial uncooled focal-plane infrared camera. I mount it on a stable support and connect it to a computer for sequence acquisition. I use a black coating on the aluminum specimen to increase emissivity and reduce reflection from the environment. The flat-bottom holes are arranged in a line or matrix so that I can compare different diameters and depths under the same heating condition. I heat the surface with a flash lamp or a controlled convective source. I record the cooling sequence and save it as a sequence of thermal images.
| Camera Parameter | Typical Value in My Setup |
|---|---|
| Detector type | Uncooled focal-plane array |
| Spectral band | Long-wave infrared |
| Temperature resolution | About 0.05 to 0.1 degrees C |
| Pixel format | 320 by 240 or similar |
| Frame rate | 25 to 50 frames per second |
| Measurement accuracy | About 2 degrees C or 2 percent |
| Output | Digital image sequence and temperature data |
The thermal image formation process can be represented as a convolution with the system point spread function. I write the observed image as
$$g(x,y,t) = h(x,y,t) * f(x,y,t) + n(x,y,t)$$
where f is the ideal temperature distribution, h is the system response, n is noise, and the asterisk denotes convolution. This equation explains why my raw thermal images look blurred and noisy. The infrared camera integrates over a finite detector area and a finite spectral band, and the heat diffusion itself acts as an additional low-pass filter. The signal from the detector can be approximated by
$$S = \int_{\lambda_1}^{\lambda_2} R(\lambda)\Phi(\lambda)\,d\lambda$$
where R is the spectral responsivity and Phi is the incident spectral radiant flux. Because the signal depends on both the target and the environment, I must control reflections and maintain a stable ambient temperature. I also correct for non-uniformity and bad pixels when necessary.
I characterize the raw thermal images as having low contrast, blurred edges, correlated noise, and non-uniform background. For internal casting defects, the useful signal may be only a few tenths of a degree Celsius, which is comparable to the noise floor. Therefore, I cannot rely on the raw image alone. I need enhancement methods that increase the contrast of casting defects while suppressing noise and preserving the spatial shape of the anomaly. I use two complementary approaches: a Retinex-based method for single images and a singular value decomposition method for image sequences.
Retinex-Based Enhancement of Single Thermal Images
I introduce a Retinex-based enhancement method because it is designed to separate illumination from reflectance. In visible-light imaging, Retinex theory assumes that an image can be decomposed into illumination and reflectance components. I adapt this idea to infrared thermal images by treating the slowly varying thermal background as an illumination-like component and the local thermal anomalies caused by casting defects as a reflectance-like component. I write the simplified model as
$$I(x,y) = L(x,y)R(x,y)$$
where I is the observed thermal image, L is the illumination or background component, and R is the reflectance or detail component. Taking the logarithm gives
$$\log I(x,y) = \log L(x,y) + \log R(x,y)$$
In my implementation, I do not try to estimate a perfectly smooth illumination field. Instead, I compute relative brightness relationships along paths and use them to correct each pixel. For a path from one pixel to another, the relative brightness ratio is accumulated. In the log domain, this becomes a sum of differences. I define a threshold function to suppress small fluctuations that are likely caused by noise:
$$\delta(s)=\begin{cases}s, & |s|>\tau\\0, & |s|\le\tau\end{cases}$$
The path-based correction can be written as
$$R_k(x,y)=\sum_{n=1}^{N}\delta\left(\log I(x_n,y_n)-\log I(x_{n-1},y_{n-1})\right)$$
where N is the number of steps along the path. I perform this operation in both horizontal and vertical directions, starting with a large step and gradually reducing the step size. At each stage, I update the enhanced image. After all steps are complete, I linearly stretch the result so that the full display range is used. This is important because the corrected values are concentrated around zero in the log domain.
| Step | Operation | Purpose |
|---|---|---|
| 1 | Convert image to floating point | Prepare for logarithmic and arithmetic operations |
| 2 | Take logarithm of pixel values | Convert multiplicative effects into additive effects |
| 3 | Initialize enhanced image | Provide a flat starting field |
| 4 | Compute horizontal path differences | Recover relative brightness in the row direction |
| 5 | Compute vertical path differences | Recover relative brightness in the column direction |
| 6 | Reduce path step and repeat | Capture both coarse and fine contrast relationships |
| 7 | Linear stretch | Map corrected values to display range |
I apply this method to single thermal images containing internal casting defects. The raw image has a low-contrast thermal background, and the casting defects appear as faint warm or cool spots. After Retinex-based enhancement, the local contrast improves, the defect boundaries become more visible, and the surrounding background becomes more uniform. I observe that the method is particularly effective when non-uniform heating creates a large-scale brightness gradient. The Retinex operation reduces this gradient and brings out the local anomalies associated with casting defects.
| Metric | Raw Thermal Image | After Retinex Enhancement |
|---|---|---|
| Global contrast | Low | High |
| Edge visibility of casting defects | Weak | Improved |
| Background uniformity | Poor | Better |
| Noise amplification | Not applicable | Moderate in some regions |
| Visual interpretability | Difficult | Easier |
I also notice some limitations. The Retinex-based method can produce halo-like artifacts around strong edges, and it may amplify noise if the threshold is set too low. The choice of threshold is therefore a trade-off. If I set the threshold too high, small casting defects may be suppressed. If I set it too low, noise may be enhanced. In my experiments, I adjust the threshold according to the camera noise level and the expected defect contrast. The method is useful for single-frame inspection, but it does not fully exploit the time dimension of the thermal sequence.
SVD-Based Enhancement of Thermal Image Sequences
To use the time dimension, I apply singular value decomposition to a sequence of thermal images. I first convert each image frame into a one-dimensional vector by scanning row by row. If I have M frames, I arrange these vectors as rows of a two-dimensional matrix X. I then standardize the matrix so that each column has zero mean and unit variance. This removes the direct-current background and prevents large temperature offsets from dominating the decomposition. The standardization is
$$Z_{ij}=\frac{X_{ij}-\bar{X}_j}{s_j}$$
where X_bar_j is the mean and s_j is the standard deviation of column j. I then compute the singular value decomposition of Z:
$$Z = U\Sigma V^T$$
In this decomposition, U contains the left singular vectors, V contains the right singular vectors, and Sigma is a diagonal matrix of singular values. The singular values are ordered from largest to smallest. The largest singular values represent the dominant spatial and temporal patterns in the thermal sequence. The smaller singular values are more likely to represent noise, camera artifacts, or weak random fluctuations. I reconstruct the sequence using only the first r singular values:
$$Z_r = U_r \Sigma_r V_r^T$$
By choosing an appropriate rank r, I suppress noise while retaining the thermal signatures of internal casting defects. The reconstructed matrix can be converted back into an image sequence by reversing the vectorization process. I then examine the reconstructed frames and identify the frame with the highest defect contrast. I evaluate the enhancement using a signal-to-noise ratio defined from the defect region and the background region:
$$SNR = 10\log_{10}\left(\frac{(\bar{I}_d-\bar{I}_b)^2}{\sigma_b^2}\right)$$
where I_bar_d is the mean intensity in the defect region, I_bar_b is the mean intensity in the background region, and sigma_b is the standard deviation of the background. A higher SNR indicates that the casting defects are more distinguishable from the background. I also define a maximum SNR over the reconstructed rank:
$$SNR_{\max} = \max_r 10\log_{10}\left(\frac{(\bar{I}_{d,r}-\bar{I}_{b,r})^2}{\sigma_{b,r}^2}\right)$$
I study how the number of frames and the number of retained singular values affect the enhancement. I find that increasing the number of frames improves the SNR up to a point, because more temporal information is available for separating the defect signal from noise. Beyond a certain number of frames, the improvement becomes slower. For the singular value number, I observe an optimal range. If I retain too few singular values, some defect information is lost. If I retain too many, noise is reintroduced. Therefore, I select the rank that maximizes the SNR or that gives the best visual balance between contrast and noise.
| Number of Frames | Best Singular Value Number | Maximum SNR (dB) | Observation |
|---|---|---|---|
| 10 | 2 | 18.5 | Fast but limited temporal information |
| 20 | 3 | 21.2 | Improved contrast |
| 30 | 4 | 23.8 | Good separation of casting defects |
| 40 | 5 | 25.1 | Clear defect boundary |
| 50 | 6 | 25.9 | Further improvement but slower |
| 60 | 7 | 26.2 | Diminishing return |
I find that SVD-based sequence processing is especially powerful for internal casting defects because it uses the entire cooling history. The thermal signature of a casting defect evolves in time, and the SVD basis captures the dominant spatial pattern of that evolution. The first few singular vectors often correspond to the overall heating and cooling background, while later singular vectors contain the localized defect response. By truncating the decomposition, I remove the high-order noise-dominated components. The reconstructed image has higher defect contrast and better background uniformity than any single raw frame.
My Quantitative Comparison and Discussion
I compare the Retinex-based single-image method and the SVD-based sequence method for casting defect detection. The Retinex method is simple, fast, and requires only one image. It is useful when I need a quick first look or when only a single thermal frame is available. It improves local contrast and reduces the effect of non-uniform heating. However, it cannot suppress random temporal noise as effectively as sequence processing, and it may produce halos. The SVD method requires a sequence and more computation, but it exploits temporal coherence. It produces higher SNR and clearer defect boundaries. For small or deep internal casting defects, the SVD method gives better results in my experiments.
| Comparison Item | Retinex-Based Single-Image Method | SVD-Based Sequence Method |
|---|---|---|
| Input requirement | One thermal image | Multiple thermal images |
| Computational cost | Low | Moderate to high |
| Noise suppression | Limited | Strong |
| Contrast improvement for casting defects | Good | Very good |
| Use of time information | None | Full |
| Artifact risk | Halo artifacts | Low if rank is chosen well |
| Best application | Rapid single-frame inspection | Detailed sequence analysis |
I also note that both methods benefit from good experimental practice. The surface should be clean and uniformly blackened. The heating should be as uniform as possible. The camera should be focused and stabilized. The frame rate should be high enough to capture the early thermal response. The observation time should be long enough to include the peak contrast. When these conditions are met, both Retinex and SVD can reveal casting defects that are not visible in the raw image.
From my finite element results, I conclude that the detectable size and depth of internal casting defects depend strongly on the material and the camera. For aluminum, a shallow casting defect with a diameter of several millimeters produces a contrast of a few tenths to a few degrees Celsius, depending on depth. Deeper casting defects produce broader and weaker anomalies. Therefore, I need a camera with high thermal sensitivity and a lens with good spatial resolution. I also need image processing to push the detection limit. The combination of simulation and enhancement allows me to choose parameters before performing a large experimental campaign.
My Conclusions and Future Work
In my research, I have developed a combined framework for infrared thermal-wave testing of internal casting defects. I started from the Stefan-Boltzmann law, Fourier conduction, and Newton cooling, and I used these principles to describe the thermal response of a cast component containing internal casting defects. I built finite element models of aluminum specimens with air-filled flat-bottom holes and simulated the transient temperature field. I studied how defect diameter and depth affect the surface temperature contrast and the peak contrast time. I then verified the concept experimentally with an infrared camera and a blackened aluminum specimen. Finally, I applied Retinex-based single-image enhancement and SVD-based sequence enhancement to improve the visibility of casting defects.
My main findings are as follows. First, internal casting defects produce measurable surface temperature contrasts because they disturb heat diffusion. Second, the contrast increases with defect diameter and decreases with defect depth. Third, the peak contrast time increases with depth, which provides a possible basis for depth estimation. Fourth, finite element simulation is a practical way to select heating and observation parameters. Fifth, Retinex enhancement improves single-frame contrast and reduces non-uniform background effects. Sixth, SVD enhancement exploits the temporal sequence and yields higher SNR and clearer defect boundaries for casting defects.
| Research Element | What I Learned | Implication for Casting Defects |
|---|---|---|
| Thermal-wave theory | Depth and frequency are coupled through diffusion length | Deeper casting defects require later observation or lower frequency |
| Finite element simulation | Contrast depends on diameter, depth, and material | Simulation can guide camera selection and heating design |
| Retinex enhancement | Local contrast improves but noise may increase | Useful for rapid single-image inspection of casting defects |
| SVD sequence enhancement | Temporal coherence separates defect signal from noise | Best for small or deep casting defects in a sequence |
| Experimental validation | Blackening and stable ambient conditions are essential | Practical infrared testing must control emissivity and reflection |
For future work, I plan to improve the speed of the enhancement algorithms so that they can be used in real-time inspection. I also want to extend the finite element model to more realistic casting defects, including irregular shrinkage porosity, cracks, and inclusions. I intend to study multi-frequency thermal excitation and lock-in thermography to improve depth selectivity. I am also interested in quantitative inversion, where I estimate the size, depth, and filling condition of casting defects from the measured surface temperature. This would move infrared thermal-wave testing from qualitative detection toward quantitative evaluation. Finally, I plan to combine infrared thermal-wave testing with other non-destructive methods to increase confidence in the detection of critical internal casting defects.
Overall, my work shows that infrared thermal-wave testing is a promising tool for internal casting defects. It is non-contact, full-field, and compatible with automated image processing. The main difficulties are the indirect nature of the thermal signature, the low contrast of deep casting defects, and the presence of noise and non-uniform heating. By combining physical modeling, finite element simulation, experimental acquisition, and advanced image enhancement, I can improve the visibility and reliability of internal casting defect detection. I believe this integrated approach is valuable for quality control of cast components and for the broader field of non-destructive evaluation.