In this paper, a comprehensive investigation into the audio frequency-based nondestructive inspection of casting defects is presented. The work focuses on the development of an acoustic resonance inspection system capable of rapidly identifying internal and surface discontinuities in cast metal components. The proposed system utilizes a fixed-energy steel ball impact excitation, an electret condenser microphone, a signal conditioning circuit, a computer sound card as the data acquisition hardware, and MATLAB-based signal processing algorithms. The fundamental principle relies on the fact that casting defects alter the mechanical resonance frequencies and internal friction (damping) characteristics of a component. By extracting the resonance frequency and the logarithmic decrement of the free-decay response, the presence and severity of casting defects can be evaluated. Experimental verification was carried out on artificially notched steel specimens and on ductile iron samples with various foundry defects, including surface cracks, inclusions, and shrinkage cavities. The results demonstrate that the acoustic inspection method provides a simple, low-cost, and highly accurate approach for online screening of casting defects. This article describes in detail the theoretical background, system hardware design, signal conditioning, data acquisition strategy, digital signal processing procedures, feature extraction techniques, and experimental outcomes. Several tables and mathematical formulations are included to summarize the relationships among physical parameters, defect geometry, and measured acoustic indicators.
Audio Frequency Inspection for Casting Defects
Nondestructive testing (NDT) has become an indispensable tool in modern manufacturing, especially for safety-critical components used in aerospace, automotive, marine, and energy industries. Among various flaws encountered during metal casting, cracks, inclusions, and shrinkage cavities are the most common. Early identification of these casting defects in the production line is essential to ensure product quality, reduce scrap, and minimize production costs. Conventional NDT techniques such as radiography, ultrasonic testing, magnetic particle testing, eddy current testing, and penetrant testing all have specific advantages, but they often require elaborate scanning procedures, skilled operators, and relatively expensive instrumentation. In contrast, acoustic resonance inspection, also known as audio frequency testing, offers a fast and cost-effective alternative that evaluates the structural integrity of the entire component in a single measurement. The method is based on the fact that every elastic structure possesses unique natural frequencies that depend on its geometry, material stiffness, density, and boundary conditions. When casting defects are present, the local stiffness or mass distribution is altered, thereby shifting the resonant frequencies and increasing the internal energy dissipation. By precisely measuring these acoustic parameters, the presence of casting defects can be inferred without damaging the inspected part.
This article reports on the design and implementation of an audio frequency inspection system specifically developed for casting defects identification. The system employs a small hard steel sphere with fixed gravitational potential energy to excite the test object. The resulting acoustic emission is captured by an electret condenser microphone, conditioned by a custom analog amplifier, and digitized using a standard personal computer sound card. Signal acquisition and subsequent processing are performed in the MATLAB environment, utilizing the Data Acquisition Toolbox and Signal Processing Toolbox. The key feature parameters extracted from the acquired audio signal are the fundamental resonance frequency and the internal friction value (or logarithmic decrement). These parameters are then used as indicators for casting defects. The experimental program involved two groups of specimens: first, steel bars with controlled artificial notches of varying depth and location, and second, ductile iron cylindrical bars containing different types of foundry defects. The results confirm the capability of the proposed audio inspection method to detect casting defects reliably with high sensitivity and repeatability.
1. Introduction and Background
In recent years, the increasing demand for high-performance cast components has intensified the need for reliable and efficient nondestructive evaluation techniques. Casting defects such as hot tears, cold shuts, shrinkage porosity, gas pores, slag inclusions, and surface cracks can significantly degrade the mechanical properties and service life of components. The purpose of my research is to develop an audio frequency-based inspection method that can be integrated into a casting production line for rapid screening of defective parts. The method exploits the natural vibration behavior of a casting when it is subjected to a mechanical impulse. All materials dissipate vibrational energy through internal friction, and the presence of discontinuities modifies both the stiffness and the damping characteristics of the structure. Consequently, the resonance frequencies and the decay rate of free vibrations provide useful signatures for detecting casting defects.
The history of acoustic testing dates back to the early practice of striking wheels and castings and listening to the resulting sound. Although this manual technique was subjective, it established the foundation for modern acoustic resonance analysis. In the 1960s, researchers began to quantify the relationship between resonance frequency and material quality. Later studies extended the application to ductile iron castings, crankshafts, and other automotive components. The development of digital signal processing and personal computers allowed the implementation of automated audio inspection systems. My work is built upon these earlier achievements and focuses on the practical implementation of a low-cost, high-accuracy audio inspection instrument for casting defects.
Compared with other NDT methods, audio frequency inspection has several distinct advantages. First, it is a global inspection method, meaning that the entire volume of the casting is tested at once, including internal regions that are inaccessible to surface techniques. Second, it does not require surface preparation or coupling agents, unlike ultrasonic testing. Third, it is highly amenable to automation, making it suitable for online quality control in foundries. Fourth, the hardware cost is relatively low because a standard PC sound card can serve as a high-resolution data acquisition device. The main limitation is that the method is primarily comparative; it requires a known good reference for establishing acceptance criteria. Nevertheless, for mass-produced components of identical geometry, audio inspection provides an excellent screening tool for casting defects.
2. Theoretical Fundamentals of Audio Inspection
2.1 Vibration of a Free Elastic Body
When a solid structure is excited by an external impulse, it undergoes free vibration. The equations of motion for a damped single-degree-of-freedom system can be written as
$$ m \frac{d^2 x(t)}{dt^2} + R \frac{dx(t)}{dt} + K x(t) = 0 \tag{1} $$
where \( m \) is the mass, \( R \) is the damping coefficient, \( K \) is the stiffness, and \( x(t) \) is the displacement as a function of time. The general solution is given by
$$ x(t) = A e^{-\delta t} \cos(\omega_d t + \phi) \tag{2} $$
with the decay rate
$$ \delta = \frac{R}{2m} \tag{3} $$
and the damped natural angular frequency
$$ \omega_d = \sqrt{\frac{K}{m} – \left(\frac{R}{2m}\right)^2} \tag{4} $$
For a long slender bar of uniform cross-section, the longitudinal natural frequencies can be derived from the one-dimensional wave equation. Consider a bar of length \( L \), cross-sectional area \( S \), density \( \rho \), and Young’s modulus \( E \). The longitudinal displacement \( u(x,t) \) satisfies
$$ \frac{\partial^2 u}{\partial t^2} = a^2 \frac{\partial^2 u}{\partial x^2} \tag{5} $$
where \( a = \sqrt{E/\rho} \). For a bar with free-free boundary conditions, the natural frequencies are
$$ f_n = \frac{n}{2L} \sqrt{\frac{E}{\rho}} \qquad (n = 1, 2, 3, \dots) \tag{6} $$
For the fundamental longitudinal mode (\( n = 1 \)), the resonance frequency becomes
$$ f = \frac{1}{2L} \sqrt{\frac{E}{\rho}} \tag{7} $$
This equation reveals that the resonance frequency depends on the material’s elastic modulus and density as well as the specimen length. In the presence of a surface crack or an internal flaw, the effective stiffness of the bar is reduced while the mass remains nearly unchanged, causing a measurable decrease in the resonance frequency. Similarly, discontinuities increase the internal friction, leading to faster decay of the vibration amplitude.
2.2 Internal Friction and Energy Dissipation
The internal friction of a material quantifies its ability to dissipate mechanical energy during cyclic deformation. In free vibration experiments, the logarithmic decrement is commonly used to characterize the damping. The logarithmic decrement \( \delta \) is defined as
$$ \delta = \frac{1}{n} \ln\left( \frac{A_1}{A_{n+1}} \right) \tag{8} $$
where \( A_1 \) is the amplitude of the first cycle and \( A_{n+1} \) is the amplitude after \( n \) complete cycles. The relationship between internal friction \( Q^{-1} \) and the logarithmic decrement is
$$ Q^{-1} = \frac{\delta}{\pi} \tag{9} $$
Alternatively, the energy dissipated per cycle relative to the total vibrational energy can be expressed as
$$ \frac{\Delta W}{W} = 2\delta \tag{10} $$
Casting defects, such as micro-cracks and porosity, cause additional frictional losses at the defect surfaces, thus increasing the internal friction value. Therefore, both the resonance frequency and the internal friction serve as complementary indicators for casting defects. My audio inspection system is designed to extract both parameters from the transient acoustic signal.
3. Design of the Audio Inspection System
3.1 System Architecture
The complete audio inspection system consists of a pulse excitation unit, an acoustic sensor, a signal conditioning circuit, a data acquisition device, and a personal computer for processing and display. The schematic architecture is presented in the block diagram below, which illustrates the flow from the mechanical excitation to the final decision on the presence of casting defects.
The excitation is provided by a hardened steel ball of known mass and height, ensuring a repeatable impact energy. The ball falls through a vertical guide tube onto the casting surface. The resulting acoustic waves are picked up by an electret condenser microphone located at a fixed distance from the excitation point. The microphone converts the acoustic pressure oscillations into an electrical voltage signal. This signal is amplified and filtered by a low-noise analog conditioning circuit before being fed into the line-in port of a personal computer sound card. The computer sound card functions as a high-resolution analog-to-digital converter, sampling the signal at a rate of 44.1 kHz with 16-bit resolution. MATLAB controls the data acquisition, stores the time-domain record, and performs all subsequent digital signal processing.
3.2 Excitation Method Selection
Several excitation methods were considered, including electromagnetic shakers, mechanical vibrators, instrumented hammers, and falling ball impact. Each method has advantages and disadvantages. Electromagnetic shakers can provide continuous sinusoidal excitation, but they are expensive and require a power amplifier. Impulse hammers are simple but are difficult to automate for online inspection. The falling steel ball was chosen because it provides a repeatable impulsive force with a broad frequency spectrum. The hardness and mass of the ball determine the effective frequency range and the force amplitude. For my experiments, a steel ball of 20 mm diameter was released from a fixed height, producing a short-duration impact that excites the fundamental and lower-order resonance modes of the test specimens.
3.3 Support Structure
To obtain reliable resonance measurements, the test object must be supported at its vibration nodes, where the displacement is zero. For a free-free bar in longitudinal vibration, the nodes occur at distances \( L/4 \) and \( 3L/4 \) from one end for the second harmonic. The support system used in my experiments consists of two thin rubber pads positioned at these nodal points. This minimizes the energy leakage through the supports and ensures that the measured decay is primarily attributable to internal friction of the material rather than external losses. For irregularly shaped castings, finite element analysis can be used to determine the optimal support locations, but for simple cylindrical bars, the analytical solution is sufficient.
3.4 Acoustic Sensor Selection
The microphone must have high sensitivity, wide flat frequency response, and long-term stability. The electret condenser microphone satisfies these requirements and is inexpensive. The selected model (OB-D22) has a frequency response of 0.04–16 kHz, a sensitivity of −38 dB ± 3 dB, and an output impedance of 2.2 kΩ. It is a back-electret type with an omnidirectional polar pattern. The microphone is positioned about 30 mm from the excited surface, ensuring a good signal-to-noise ratio without affecting the vibration of the specimen. The performance parameters of the microphone are summarized in Table 1.
| Parameter | Value |
|---|---|
| Sensitivity | −38 dB ± 3 dB |
| Output impedance | 2.2 kΩ |
| Operating voltage | 2.0 V |
| Frequency response | 0.04–16 kHz |
| Signal-to-noise ratio | > 68 dB |
| Polar pattern | Omnidirectional back-electret |
Table 1: Performance parameters of the electret condenser microphone used in the audio inspection system.
3.5 Signal Conditioning Circuit
The output voltage of the microphone is very small (millivolts), while the line-in input of a computer sound card expects a signal in the range of about ±1 V. Therefore, a low-noise amplifier is required. The conditioning circuit I designed is based on a three-op-amp instrumentation amplifier followed by a proportional amplifier stage. The total voltage gain can be expressed as
$$ A_u = \left( 1 + \frac{2R_2}{R_W} \right) \cdot \frac{R_9}{R_8} \tag{11} $$
With the component values used, the gain can be adjusted from 500 to 2100, which is sufficient to amplify the microphone output to the optimum level for the sound card. A transient voltage suppressor diode (1.5 V) is connected at the output of the amplifier to protect the sound card from accidental overvoltage.
Impedance matching between the amplifier output and the sound card input is crucial to avoid signal degradation. The output impedance of the amplifier is kept lower than the sound card’s line-in input impedance (generally around 10 kΩ). No additional buffer is required because the instrumentation amplifier provides a low output impedance.
3.6 Data Acquisition Using a Computer Sound Card
A standard computer sound card is a cost-effective data acquisition device for audio frequency signals. Most modern sound cards use high-quality sigma-delta A/D converters with 16-bit or 24-bit resolution and sampling rates up to 96 kHz. For the purpose of casting defects detection, the maximum analysis frequency is set to 16 kHz. According to the Nyquist–Shannon sampling theorem, the sampling frequency must be at least twice the highest frequency of interest. I selected a sampling rate of 44.1 kHz, which provides a comfortable margin while remaining compatible with all standard sound cards. The sampling resolution of 16 bits gives a dynamic range of about 96 dB, which is adequate for resolving the weak early part of the decaying signal. The sound card also provides anti-aliasing filters, simplifying the analog front end.
The MATLAB Data Acquisition Toolbox (R2007a) provides a convenient interface for controlling the sound card. The function analoginput('winsound') establishes a connection to the Windows sound card driver. The input channel type, sampling rate, trigger mode, and samples per trigger are configured through properties of the analog input object. The acquisition software waits for a rising edge trigger on the microphone channel when the signal exceeds a predefined threshold (typically 0.1 V). After triggering, 1 second of data (44,100 samples) is captured and stored in the computer’s memory. The data is then saved in a CSV file for offline analysis. The flowchart of the acquisition program is shown in the following steps:
- Initialize the sound card device and display its hardware properties.
- Add a single analog input channel.
- Set sampling rate to 44,100 Hz.
- Set software trigger with rising edge condition and threshold 0.1 V.
- Set samples per trigger to 220,500 (i.e., 5 seconds), but wait for only the first 1 second after the trigger.
- Start the acquisition and wait until the specified number of samples is acquired.
- Retrieve the data and time vector from the data acquisition object.
- Plot the time-domain signal.
- Save the data to a .csv file.
- Delete the analog input object to release the sound card.
4. Acoustic Signal Processing
4.1 Preprocessing
The raw acquired signal contains not only the desired vibration response but also the ambient noise and possible DC offset from the microphone preamplifier. Preprocessing is necessary before extracting features. The following steps are applied:
- Trend removal: The signal may contain a slow-varying baseline due to amplifier drift. A polynomial of degree \( m \) (typically \( m = 1 \) or \( 2 \)) is fitted to the data using the least-squares method, and the fitted polynomial is subtracted from the original signal. If the measured signal is \( x_k \) and the fitted polynomial is \( p_k \), the corrected signal is
$$ y_k = x_k – p_k \tag{12} $$
- Smoothing: To reduce high-frequency noise, a five-point three-degree smoothing algorithm is applied. The smoothed value \( y_i \) is calculated from the neighboring raw data points using the following coefficients:
$$ y_i = \frac{1}{70}(69x_i + 4(x_{i+1} + x_{i-1}) – 6(x_{i+2} + x_{i-2})) \tag{13} $$
for \( i = 3, 4, \dots, n-2 \). This algorithm effectively attenuates random noise while preserving the shape of the resonant peaks.
- Digital filtering: A band-pass digital filter is applied to isolate the frequency band containing the fundamental resonance. Since the resonance frequencies of the test bars lie between 7 kHz and 10 kHz, a fourth-order Butterworth band-pass filter with lower and upper cutoff frequencies of 6 kHz and 12 kHz is used. The filter is implemented using the zero-phase filtering function
filtfiltin MATLAB, which avoids phase distortion.
4.2 Frequency Domain Analysis
The resonance frequency is determined by computing the Fast Fourier Transform (FFT) of the preprocessed time-domain signal. The power spectrum is obtained as the squared magnitude of the FFT. The frequency corresponding to the maximum spectral peak is taken as the resonance frequency. In my experiments, I normalized the power spectrum to the maximum amplitude for the purpose of comparison. The frequency resolution is given by
$$ \Delta f = \frac{f_s}{N} \tag{14} $$
With \( f_s = 44,\!100 \) Hz and \( N = 44,\!100 \), the resolution is exactly 1 Hz, which is sufficient for accurate resonance frequency measurement.
4.3 Time Domain Feature Extraction
The internal friction (or logarithmic decrement) is obtained from the envelope of the decaying time-domain signal. After band-pass filtering, the signal is approximately a single-frequency damped sinusoid. The Hilbert transform is used to extract the analytic amplitude envelope. Then a linear regression is performed on the natural logarithm of the envelope with respect to time. The slope of the fitted line equals \( -\delta \), where \( \delta \) is the logarithmic decrement. From \( \delta \), the internal friction value is computed as
$$ Q^{-1} = \frac{\delta}{\pi} \tag{15} $$
To improve robustness, multiple measurements are averaged for each specimen. The coefficient of variation is typically below 2%, indicating good repeatability.
5. Experimental Setup and Specimens
5.1 Artificially Notched Steel Bars
To investigate the influence of defect depth and location, I prepared 16 identical steel bars with dimensions of 20 mm diameter and 250 mm length. The bars were free from detectable defects. Artificial notches were machined at different positions along the bar length: at \( L/6 \), \( L/4 \), \( L/3 \), and \( L/2 \) from one end. The notch width was fixed at 1 mm, while the notch depth varied from 1 mm to 4 mm. A schematic of the notch geometry is shown in the form of parameters in Table 2.
| Parameter | Symbol | Value / Description |
|---|---|---|
| Specimen length | \( L \) | 250 mm |
| Specimen diameter | \( d \) | 20 mm |
| Notch width | \( b \) | 1 mm |
| Notch depth | \( h \) | 1, 2, 3, 4 mm |
| Notch location | \( L_1 \) | L/6, L/4, L/3, L/2 |
Table 2: Geometry of artificial notches on steel test bars.
5.2 Cast Ductile Iron Bars with Real Defects
A second set of specimens was cast from ductile iron (QT200). The cylindrical bars had the same dimensions as the steel bars. The casting process was intentionally altered to produce different types of defects, including surface cracks, slag inclusions, and shrinkage cavities. A reference group of sound bars was also cast under normal conditions. All specimens were subjected to audio inspection to record the resonance frequency and internal friction values.
6. Experimental Results and Discussion
6.1 Resonance Frequency of Notched Bars
The measured resonance frequencies of the notched steel bars are summarized in Table 3. The baseline (no notch) resonance frequency was approximately 8537–8540 Hz for all bars, confirming good consistency. When a notch is present, the resonance frequency decreases. For a fixed notch location, increasing the notch depth causes a monotonic decrease in frequency. For example, at the L/2 location, the frequency drops from 8517 Hz for a 1 mm notch to 8199 Hz for a 4 mm notch. The results also show that the notch location has only a minor effect; the depth of the notch is the dominant factor.
| Notch depth h (mm) | Resonance frequency (Hz) at notch location L1 | |||
|---|---|---|---|---|
| L/6 | L/4 | L/3 | L/2 | |
| 0 (no notch) | 8537 | 8540 | 8538 | 8537 |
| 1 | 8512 | 8518 | 8520 | 8517 |
| 2 | 8486 | 8480 | 8473 | 8468 |
| 3 | 8357 | 8344 | 8345 | 8339 |
| 4 | 8219 | 8205 | 8224 | 8199 |
Table 3: Resonance frequencies of steel bars with artificial notches at different depths and locations.
This behavior can be explained by the bending and longitudinal vibration characteristics of the bar. A notch creates a local reduction in cross-section, thereby lowering the effective stiffness. According to Equation (7), a lower stiffness reduces the resonance frequency. The effect is more pronounced when the notch is deep because the remaining load-bearing area is smaller.
6.2 Power Spectrum Analysis
To visualize the spectral changes, the normalized power spectra of a sound bar and notched bars are compared. The fundamental resonance peak shifts to lower frequencies as the notch depth increases. The peak amplitude also decreases and the spectral width broadens slightly due to increased damping. Although I cannot include the actual images, the trend is evident from the numerical frequency data and from the corresponding internal friction values presented in the next section.
6.3 Internal Friction of Notched Bars
The internal friction values (\( Q^{-1} \times 10^{-3} \)) for the notched bars are listed in Table 4. The baseline internal friction of sound bars is about \( 1.50 \times 10^{-3} \). The presence of a notch increases the internal friction. For instance, at the L/2 location, a 4 mm deep notch raises the internal friction to \( 2.386 \times 10^{-3} \), an increase of approximately 58% relative to the baseline. The internal friction increases with notch depth at all locations. The location has a relatively weak influence on internal friction, similar to its effect on resonance frequency.
| Notch depth h (mm) | Internal friction \( Q^{-1} \times 10^{-3} \) at notch location L1 | |||
|---|---|---|---|---|
| L/6 | L/4 | L/3 | L/2 | |
| 0 (no notch) | 1.505 | 1.501 | 1.509 | 1.506 |
| 1 | 1.809 | 1.786 | 1.796 | 1.824 |
| 2 | 2.031 | 1.998 | 2.050 | 2.067 |
| 3 | 2.152 | 2.147 | 2.202 | 2.178 |
| 4 | 2.376 | 2.352 | 2.408 | 2.386 |
Table 4: Internal friction values of steel bars with artificial notches.
The increase in internal friction is attributed to the rubbing of the two crack faces during vibration, which dissipates energy as heat. Additionally, stress concentration at the notch tip converts some of the vibrational energy into localized plastic deformation, further increasing damping. The measured internal friction therefore serves as a sensitive indicator for casting defects even when the resonance frequency shift is relatively small.
6.4 Real Casting Defects
Table 5 presents the average resonance frequencies and internal friction values for ductile iron bars with different types of casting defects. The sound bars have a resonance frequency of 7630 Hz and an internal friction of \( 2.436 \times 10^{-3} \). All defective bars exhibit lower resonance frequencies and higher internal friction. Surface cracks caused the largest frequency drop (down to 7498 Hz) and a significant increase in internal friction to \( 3.242 \times 10^{-3} \). Shrinkage cavities produced the highest internal friction of \( 3.568 \times 10^{-3} \), while inclusions led to an intermediate response.
| Casting defect type | Resonance frequency (Hz) | Internal friction \( Q^{-1} \times 10^{-3} \) |
|---|---|---|
| None (sound) | 7630 | 2.436 |
| Surface crack | 7498 | 3.242 |
| Inclusion | 7545 | 2.753 |
| Shrinkage cavity | 7537 | 3.568 |
Table 5: Resonance frequencies and internal friction values of ductile iron bars with different casting defects.
The results demonstrate that any form of casting defect causes measurable changes in the acoustic response. The combination of resonance frequency and internal friction provides better discrimination than either parameter alone. For example, an inclusion and a shrinkage cavity may produce similar frequency shifts, but their internal friction values differ significantly, allowing them to be distinguished. Thus, my audio inspection system can be used not only for detecting the presence of casting defects but also for making preliminary assessments of defect type.
7. Statistical Evaluation and Acceptance Criteria
To apply the method in a production environment, it is necessary to define statistical acceptance criteria. For a given population of castings with the same geometry and material, the distribution of resonance frequencies and internal friction values of sound parts is approximately Gaussian. The mean and standard deviation are calculated from a reference set. A casting is classified as defective if its resonance frequency is lower than the mean minus three standard deviations (\( \mu_f – 3\sigma_f \)) or if its internal friction is higher than the mean plus three standard deviations (\( \mu_Q + 3\sigma_Q \)). This 3-sigma rule provides a high confidence level with a false rejection rate of approximately 0.3% for normally distributed data.
In my experiments, the sound steel bars had a mean frequency of 8538 Hz with a standard deviation of 1.5 Hz. Thus, the acceptance lower limit was about 8533 Hz. All bars with notches deeper than 1 mm exceeded this limit, while the 1 mm notch bars had frequencies near the boundary in some locations. The internal friction criterion was even more sensitive because the baseline standard deviation was only about 0.004 × 10⁻³. Therefore, even a 1 mm notch would be reliably detected by the internal friction measurement. In practice, a combination of both parameters in a multivariate control chart offers the best robustness.
8. Implementation Notes and Practical Considerations
8.1 Repeatability and Environmental Noise
During the experiments, I observed that the measured resonance frequencies were highly repeatable, with variations below ±2 Hz for the same specimen across multiple tests. The internal friction values varied by less than 2%. The main sources of variation were random background noise and slight differences in the impact position of the steel ball. To reduce the influence of background noise, the system performed a background sample immediately before each test, and the noise spectrum was subtracted from the measured signal spectrum. Alternatively, the band-pass filter effectively removed out-of-band noise, making the measurements less sensitive to acoustic interference in the foundry environment.
8.2 Automation Potential
The audio inspection system is easily automated. The steel ball can be replaced by a solenoid-driven impactor, and the microphone can be positioned at a fixed distance. A conveyor can carry the castings through the inspection station. The signal processing and classification can be implemented in a real-time environment, with results displayed on a screen and used to trigger a sorting mechanism. The total inspection time per casting is less than one second, making it suitable for high-throughput production lines.
9. Comparative Discussion With Other NDT Methods
It is instructive to compare the proposed audio inspection method with conventional NDT techniques in the context of casting defects detection. Table 6 summarizes the qualitative comparison.
| Method | Surface vs. internal | Typical resolution | Cost | Speed | Automation |
|---|---|---|---|---|---|
| Radiographic testing | Internal | High | High | Slow | Moderate |
| Ultrasonic testing | Internal | High | Medium | Medium | Moderate |
| Magnetic particle testing | Surface/near-surface | Medium | Low | Medium | Limited |
| Eddy current testing | Surface/near-surface | Medium | Medium | Fast | Good |
| Penetrant testing | Surface-open | High | Low | Slow | Limited |
| Audio frequency inspection | Both (global) | Medium | Low | Very fast | Excellent |
Table 6: Qualitative comparison of NDT methods for casting inspection.
The audio frequency inspection method is not capable of precisely locating a defect, nor does it provide the detailed shape and size information that radiography or ultrasonic C-scan can deliver. However, for the purpose of sorting defective castings from acceptable ones, it offers an outstanding combination of low cost, high speed, and ease of automation. It is particularly advantageous for mass-produced castings of identical geometry, such as automotive crankshafts, connecting rods, and pipe fittings. In such applications, the vast majority of parts are sound, and the audio method acts as a rapid filter that rejects the few defective parts.
10. Conclusion
In this study, I designed, built, and tested an audio frequency inspection system for detecting casting defects. The main conclusions from this work are as follows:
- The audio inspection system, consisting of a steel ball impact exciter, an electret condenser microphone, a signal conditioning circuit, a computer sound card, and MATLAB-based processing software, is capable of detecting casting defects with high accuracy and repeatability.
- The resonance frequency of a casting is shifted to lower values in the presence of surface cracks, inclusions, or shrinkage cavities. The frequency shift increases with the severity of the defect. For the steel bars studied, a 4 mm notch caused a frequency decrease of approximately 2–4% relative to the base value.
- The internal friction (or logarithmic decrement) is a sensitive indicator of casting defects. The internal friction increased by up to 58% for deep notches and by 30–46% for real casting defects compared with sound specimens.
- The combination of resonance frequency and internal friction provides a robust two-parameter signature that can discriminate between different types of casting defects in many cases.
- The use of a standard PC sound card as a high-resolution A/D converter dramatically reduces the hardware cost while maintaining satisfactory performance for audio frequency signals.
- The smoothing, trend removal, and digital filtering algorithms implemented in MATLAB were effective in extracting clean resonance parameters from noisy acoustic signals.
- The proposed method is highly suitable for online inspection in foundries because it is non-contact, requires no surface preparation, operates in less than one second, and can be fully automated.
Future work will focus on extending the method to complex-shaped castings using finite element analysis to predict resonant modes and to optimize support and excitation positions. Artificial intelligence techniques, such as neural networks, may also be integrated to classify defect types based on multiple feature parameters. The ultimate goal is to create a standalone, industrial-grade audio inspection instrument that can be deployed on high-speed casting production lines for 100% inspection of casting defects.
In summary, audio frequency inspection provides an elegant and economical solution for casting defects detection. It fulfills the modern demands of quality assurance, cost reduction, and automation. The experimental evidence presented in this paper confirms its practical value, and I believe it will find increasing adoption in the foundry industry as a reliable screening tool for casting defects.