1. Introduction and Research Significance
In recent years, the rapid development of metallurgy, aviation, shipbuilding, architecture, and heavy machinery has placed increasingly severe requirements on cast components that must withstand complex loading and harsh service conditions. To ensure product quality, reduce manufacturing cost, and guarantee engineering safety, online detection and evaluation of castings in the foundry has become an essential step. Among the many defects encountered in foundry production, cracks are among the most dangerous because they can propagate under cyclic loading and lead to sudden failure. The detection of a sand foundry defect such as a crack, shrinkage cavity, or inclusion at the earliest possible stage of production is therefore of great practical importance.
Nondestructive testing (NDT) is defined as a family of techniques used to inspect the internal and external quality of materials without damaging their serviceability. The principle is to study the changes in thermal, acoustic, optical, electrical, or magnetic responses caused by internal discontinuities. Numerous NDT methods have been applied to castings, including penetrant testing, magnetic particle testing, eddy current testing, radiography, ultrasonic testing, acoustic emission testing, and infrared thermography. Table 1 compares these conventional methods from the viewpoint of casting inspection.
| Method | Physical Principle | Advantages | Limitations |
|---|---|---|---|
| Penetrant testing | Capillary action draws liquid penetrant into surface-open discontinuities | Simple, low cost, direct indication | Detects only open-to-surface defects; surface preparation is critical |
| Magnetic particle testing | Leakage magnetic fields attract magnetic particles around surface/near-surface defects | Fast, sensitive to cracks, simple equipment | Only ferromagnetic materials; post-demagnetization may be required |
| Eddy current testing | Electromagnetic induction changes coil impedance near conductive material | High speed, no couplant, useful for tubes/bars | Limited depth; lift-off effects; only conductive materials |
| Radiography | Differential absorption of penetrating radiation reveals internal discontinuities | Direct volumetric image, useful for porosity and inclusions | Expensive, radiation hazard, poor sensitivity to tight cracks |
| Ultrasonic testing | Reflection/transmission of high-frequency mechanical waves | High penetration, good sensitivity, portable | Requires couplant; operator-dependent; complex geometry can be difficult |
| Acoustic emission | Transient elastic waves generated by sudden internal energy release | Dynamic monitoring of growing defects | Needs service load; multiple sensors for location |
| Infrared testing | Surface temperature variations caused by defect-induced thermal property changes | Non-contact, full-field imaging | Limited by emissivity, geometry, and heating conditions |
The conventional NDT methods mentioned above are mostly aimed at detecting and visualizing a specific discontinuity. For many foundries, however, the more important question is whether the casting is acceptable or not, rather than the exact morphology of the defect. Audio frequency inspection differs from conventional scanning methods because it measures the global structural response of the workpiece. The mechanical resonance of a casting is determined by its geometry, elastic modulus, density, and damping. When a sand foundry defect is present, the local stiffness and damping change, which in turn changes the resonance frequencies and the decay rate of free vibration. Therefore, by exciting the casting and analyzing the emitted acoustic signal, I can evaluate the overall quality of the casting in a single measurement.
The audio frequency method is particularly attractive for online casting inspection because it does not require surface preparation, does not need a couplant, and can detect defects located both on the surface and in the interior of the casting. It provides a statistical evaluation of structural integrity through parameters such as resonance frequency and internal friction. Compared with traditional radiography and ultrasonic testing, the hardware is simpler and the test speed is much higher. In this work, I built a complete audio frequency inspection system consisting of an impulse excitation device, an acoustic sensor, a signal conditioning circuit, a data acquisition unit, and a computer. I used the system to detect cracks, shrinkage cavities, and inclusions in cast specimens, and the results confirm that the method is capable of identifying a sand foundry defect accurately and rapidly.
2. Fundamental Principles of Audio Frequency Inspection
When a casting is struck by an external impulse, it begins to vibrate. The vibration is transmitted to the surrounding air as an audible acoustic wave. The frequency content and amplitude decay of that wave carry information about the casting’s mechanical properties and structural integrity. I used two main physical principles in the design: acoustic resonance measurement and acoustic energy attenuation measurement.
2.1 Acoustic Resonance Measurement
The vibration caused by an impact is a transient process, but it can be analyzed by expanding the excitation into a Fourier series. Any periodic function satisfying the Dirichlet condition can be expressed as an infinite sum of harmonic components:
$$ f(t) = \frac{a_0}{2} + \sum_{n=1}^{\infty}\left[a_n \cos(n\omega_0 t) + b_n \sin(n\omega_0 t)\right] $$
where the Fourier coefficients are obtained from:
$$ a_n = \frac{2}{T_0}\int_{t_0}^{t_0+T_0} f(t)\cos(n\omega_0 t)\,dt $$
$$ b_n = \frac{2}{T_0}\int_{t_0}^{t_0+T_0} f(t)\sin(n\omega_0 t)\,dt $$
and the fundamental circular frequency is:
$$ \omega_0 = \frac{2\pi}{T_0} $$
For a single-degree-of-freedom vibrating system with mass \(m\), damping coefficient \(R\), and stiffness \(K\), the equation of free vibration is:
$$ m\frac{d^2x(t)}{dt^2} + R\frac{dx(t)}{dt} + Kx(t) = 0 $$
Assuming a solution of the form:
$$ x(t) = A e^{\lambda t} $$
I obtained the characteristic equation:
$$ m\lambda^2 + R\lambda + K = 0 $$
Defining the undamped natural circular frequency and the damping factor:
$$ \omega = \sqrt{\frac{K}{m}} $$
$$ \delta = \frac{R}{2m} $$
the roots of the characteristic equation become:
$$ \lambda_{1,2} = -\delta \pm \sqrt{\delta^2 – \omega^2} $$
For a uniform rod of length \(L\), diameter \(d\), density \(\rho\), and dynamic elastic modulus \(E\), the fundamental longitudinal resonance frequency can be expressed as:
$$ f_1 = \frac{1}{2L}\sqrt{\frac{E}{\rho}} $$
This shows that the resonance frequency depends on both the geometry and the material properties of the casting. If a sand foundry defect changes the effective stiffness or mass distribution, the resonance frequency will shift. Therefore, by measuring the resonance frequency, I can identify whether the casting is internally sound. The measurement can be performed in three ways: counting the number of vibration cycles per unit time, computing the resonance frequency from mechanical properties, or indirectly deriving it from the measured sound velocity.
2.2 Acoustic Energy Attenuation and Internal Friction
When the external excitation is removed, the casting undergoes damped free vibration. The amplitude of vibration decreases exponentially because internal friction dissipates the mechanical energy. The general solution of the damped free vibration equation is:
$$ x(t) = A_0 \exp\left(-\frac{R}{2m}t\right)\cos(\omega t + \phi) $$
Let \(A_n\) be the amplitude of the \(n\)-th cycle and \(A_{n+1}\) be the amplitude of the next cycle. The ratio of two successive amplitudes is:
$$ \frac{A_n}{A_{n+1}} = \exp\left(\frac{RT}{2m}\right) $$
Taking the natural logarithm and using \(n\) cycles, I define the logarithmic decrement:
$$ \delta = \frac{1}{n}\ln\frac{A_n}{A_{n+1}} $$
The logarithmic decrement is directly related to the energy loss per cycle. If \(W\) is the total vibration energy and \(\Delta W\) is the energy dissipated in one cycle, then:
$$ \delta = \frac{\Delta W}{2W} $$
or equivalently:
$$ \frac{\Delta W}{W} = 2\delta $$
The phase lag angle \(\varphi\) between stress and strain is related to the energy loss by:
$$ \varphi = \frac{\Delta W}{2\pi W} $$
Therefore, the internal friction \(Q^{-1}\) can be written as:
$$ Q^{-1} = \frac{\varphi}{\pi} = \frac{\delta}{\pi} $$
In practice, I used the logarithmic decrement as the internal friction parameter because it is easy to extract from the decaying time-domain signal. Defects such as cracks, inclusions, and shrinkage cavities increase energy dissipation during vibration. Consequently, a casting containing a sand foundry defect usually exhibits a higher internal friction value than a sound casting of the same geometry and material.
3. Construction of the Audio Inspection System
I designed the audio inspection system with the following main components: an impulse excitation device, an audio sensor, a signal conditioning circuit, a data acquisition card, and a microcomputer. The system uses a small hardened steel ball with a fixed gravitational potential energy to excite the casting. The emitted acoustic signal is converted to an electrical signal by the sensor, conditioned by the amplification circuit, sampled by a computer sound card, and finally processed by MATLAB. The overall architecture is illustrated by the following component scheme:
| Component | Function | Selected Implementation |
|---|---|---|
| Impulse excitation device | Provides a repeatable broadband mechanical impulse | Steel ball falling from a fixed height |
| Audio sensor | Converts acoustic pressure into an electrical signal | Electret condenser microphone |
| Signal conditioning circuit | Amplifies and filters the weak microphone signal | Instrumentation amplifier with gain control |
| Data acquisition card | Samples the analog signal into digital data | Computer sound card |
| Microcomputer | Stores, processes, and displays the signal features | PC with MATLAB |
3.1 Excitation Method
Common excitation methods for vibration testing include electromagnetic shakers, hydraulic shakers, mechanical eccentric-mass exciters, and impact hammers. Shakers and exciters can provide precise sine sweeps, but they are expensive and difficult to integrate into an automatic production line. An impact hammer is simple but requires an experienced operator and the impact force varies from one strike to another. For industrial online detection, I needed a repeatable and economical excitation source.
I selected a small hardened steel ball released from a fixed height. The ball is guided by an inclined slide or a vertical tube so that it always strikes the same location with approximately the same velocity. The potential energy of the ball is:
$$ U = mgh $$
and the velocity just before impact is:
$$ v = \sqrt{2gh} $$
The hardness of the steel ball controls the frequency content of the impulse. A hard ball produces a very short contact time, which excites a broad frequency range. The mass and the falling height determine the magnitude of the impact force. This excitation method is highly repeatable, free of operator bias, and suitable for automated inspection.
3.2 Support and Boundary Conditions
To obtain accurate resonance measurements, the casting must be supported at points where the vibration displacement is minimal. These points are called vibration nodes. For a free-free uniform bar vibrating longitudinally, the mode shape is:
$$ \phi_n(x) = C\cos\left(\frac{n\pi x}{L}\right) $$
The nodes occur when:
$$ \frac{n\pi x}{L} = \frac{(2k+1)\pi}{2}, \quad k=0,1,2,\ldots $$
For the second longitudinal mode, \(n=2\), the nodes are located at:
$$ x = \frac{L}{4} \quad \text{and} \quad x = \frac{3L}{4} $$
I therefore placed the supports at these two positions. This minimizes the energy loss from the casting to the support frame and ensures that the measured damping mainly reflects the internal friction of the material. In the experiments, I used soft rubber strips or foam pads at the nodal positions so that the boundary conditions remained close to free-free.
3.3 Audio Sensor Selection
The audio sensor, or microphone, is the key element that converts the acoustic pressure radiated by the vibrating casting into an electrical voltage. Several types of microphones are available for acoustic measurement. Table 3 compares the important microphone families.
| Microphone Type | Advantages | Disadvantages |
|---|---|---|
| Capacitive microphone | High sensitivity, flat frequency response, low distortion | Expensive, fragile, requires polarization voltage |
| Moving-coil microphone | Robust, stable, no external power required | Lower sensitivity, larger volume |
| Piezoelectric microphone | Simple, inexpensive, high output | Narrow frequency range, lower accuracy |
| Electret condenser microphone | High sensitivity, wide bandwidth, small size, low cost | Internal FET amplifier needs a small bias voltage |
After comparing the alternatives, I chose an electret condenser microphone. Its performance parameters are listed in Table 4.
| Parameter | Value |
|---|---|
| Sensitivity | −38 dB ± 3 dB |
| Output impedance | 2.2 kΩ |
| Operating voltage | 2.0 V |
| Frequency response range | 0.04 – 16 kHz |
| Signal-to-noise ratio | Greater than 68 dB |
| Directional pattern | Omnidirectional back-electret |
The electret condenser microphone satisfies the two most important requirements for the system: high sensitivity and good long-term stability. It is also non-contact, so it does not add mass or stiffness to the vibrating casting. This is critical because any additional mass would alter the resonance frequency and produce false indications.
3.4 Data Acquisition Hardware
The signal produced by the microphone is analog. I needed an analog-to-digital converter to digitize the signal before processing. A dedicated data acquisition card can perform this function, but it is relatively expensive. Modern computer sound cards contain high-quality A/D converters with 16-bit or even 24-bit resolution, and their maximum sampling rates are more than sufficient for acoustic signals up to 20 kHz.
The key specifications of A/D conversion are resolution, conversion accuracy, conversion time, and full-scale range. The resolution of an n-bit A/D converter is:
$$ \text{Resolution} = \frac{V_{FS}}{2^n} $$
where \(V_{FS}\) is the full-scale input voltage. A 16-bit sound card with a 1 V full-scale input can resolve voltage variations of about:
$$ \frac{1}{65536} \approx 1.53 \times 10^{-5} \text{ V} $$
Because the microphone signal is relatively weak, the analog signal must be amplified before it enters the sound card. I used a computer sound card as the data acquisition hardware because it is inexpensive, widely available, and completely adequate for the frequency range of interest. The sound card is controlled through MATLAB’s Data Acquisition Toolbox, which provides functions for configuring the sampling rate, number of channels, trigger conditions, and the number of samples to acquire.
3.5 Signal Conditioning Circuit
The microphone output is usually only a few millivolts. The sound card’s line input expects a signal of about 1 volt. Therefore, I designed a signal conditioning circuit based on instrumentation amplifiers to amplify the weak audio signal while rejecting common-mode interference. The total gain of the amplifier circuit is:
$$ A_u = \left(1 + \frac{2R_2}{R_W}\right)\frac{R_6}{R_5} $$
With the component values used in my design, the gain could be adjusted in the range from approximately 500 to 2100. I also added a transient voltage suppressor diode in parallel with the input to protect the sound card from over-voltage. In addition, I paid attention to impedance matching. The output impedance of the amplifier should be slightly lower than the input impedance of the sound card line input. For this reason, I connected a resistor in parallel with the line input to match the impedance and reduce signal reflection.
3.6 Acquisition Parameters
The resonance frequency of the tested castings was around 7.5 to 8.6 kHz. To obey the Nyquist sampling theorem, the sampling frequency must be at least twice the highest frequency component of interest. I set the sampling frequency to 44.1 kHz and the sampling resolution to 16 bits. The acquisition parameters are summarized in Table 5.
| Parameter | Setting |
|---|---|
| Sampling frequency | 44.1 kHz |
| Sampling precision | 16 bits |
| Number of channels | 1 |
| Trigger type | Software rising-edge trigger |
| Trigger level | 0.1 V |
| Trigger delay | −1 s |
| Samples per trigger | 220500 |
| Acquisition duration | 1 s |
The frequency resolution in the subsequent FFT analysis is:
$$ \Delta f = \frac{f_s}{N} = \frac{44100}{220500} = 0.2 \ \text{Hz} $$
This high frequency resolution allowed me to distinguish small shifts in the resonance frequency caused by a sand foundry defect.
4. Signal Acquisition, Analysis, and Processing
4.1 Data Acquisition Implementation
I implemented the acquisition program in MATLAB using the Data Acquisition Toolbox. The program first queries the sound card hardware capabilities using the daqhwinfo function, then creates an analog input object, adds a channel, sets the sampling rate, trigger condition, and number of samples, and finally starts the acquisition. When the sound card detects the trigger level, the acoustic signal is stored in memory and then written to a file for offline processing.
The main acquisition steps are:
- Initialize the sound card input object.
- Set the sampling frequency and number of bits.
- Set the sample length and trigger parameters.
- Start waiting for the trigger signal.
- Capture one second of audio data.
- Save the data as a CSV file.
- Release the audio device object after processing.
4.2 Time-Domain Analysis
The acquired acoustic signal is a random-like transient signal. In the time domain, I calculated several statistical parameters. The arithmetic mean value is:
$$ \mu_x = \frac{1}{N}\sum_{i=1}^{N} x(t_i) $$
The mean square value is:
$$ \phi_x^2 = \frac{1}{N}\sum_{i=1}^{N} x^2(t_i) $$
The variance is:
$$ \sigma_x^2 = \phi_x^2 – \mu_x^2 $$
The root-mean-square value, which describes the overall acoustic intensity, is:
$$ x_{rms} = \sqrt{\phi_x^2} $$
4.3 Correlation Analysis
Correlation analysis is useful for detecting periodic components in a noisy signal. The autocorrelation function of a discrete signal \(x(n)\) is defined as:
$$ R_{xx}(m) = \sum_{n=-\infty}^{\infty} x^*(n)x(n+m) $$
The cross-correlation function between two signals \(x(n)\) and \(y(n)\) is:
$$ R_{xy}(m) = \sum_{n=-\infty}^{\infty} x^*(n)y(n+m) $$
These functions help determine the similarity between two signals and are useful for verifying the repeatability of the excitation and the stability of the sensor.
4.4 Frequency-Domain Analysis
The most important analysis for audio frequency inspection is the frequency-domain analysis. The Fourier transform of a continuous time signal is:
$$ X(\omega) = \int_{-\infty}^{\infty} x(t)e^{-j\omega t}\,dt $$
In the discrete implementation, I used the fast Fourier transform (FFT). The power spectrum is estimated as:
$$ P(k) = \frac{1}{N}\left|X(k)\right|^2 $$
The resonance frequency of the casting appears as a clear peak in the power spectrum. By tracking the location of this peak, I could detect changes caused by defects. A defect such as a crack or a shrinkage cavity reduces the local stiffness and increases damping, which causes the resonance peak to shift to a lower frequency and to decrease in amplitude.
4.5 Signal Preprocessing
The raw acoustic data often contains a slowly varying baseline drift called the trend term. This drift is caused by amplifier zero drift and by environmental interference. I removed the trend term using the polynomial least-squares method. A polynomial trend is expressed as:
$$ \hat{x}_k = a_0 + a_1 k + a_2 k^2 + \cdots + a_m k^m $$
The coefficients \(a_j\) are chosen to minimize the error function:
$$ E = \sum_{k=1}^{N}\left(\hat{x}_k – x_k\right)^2 $$
After solving the linear normal equations for \(a_j\), the detrended signal is:
$$ y_k = x_k – \hat{x}_k $$
I also applied smoothing to the signal to reduce high-frequency random noise. A commonly used method is the five-point Savitzky-Golay smoothing formula:
$$ y_i = \frac{1}{35}\left(-3x_{i-2} + 12x_{i-1} + 17x_i + 12x_{i+1} – 3x_{i+2}\right) $$
for \(i = 3, 4, \ldots, N-2\). This method reduces the small random fluctuations without seriously distorting the main features of the signal.
4.6 Digital Filtering
Filtering is used to remove frequency components outside the region of interest. Digital filtering can be performed in the frequency domain or in the time domain. In the frequency-domain method, I computed the FFT of the signal, zeroed or attenuated the unwanted frequency components, and then used the inverse FFT to reconstruct the filtered signal:
$$ y(m) = \sum_{k=0}^{M-1} H(k)X(k)e^{j2\pi km/M} $$
where \(H(k)\) is the frequency response of the filter. For the audio inspection system, I used a band-pass filter to preserve the frequency band containing the resonance peaks, typically from about 4 kHz to 16 kHz.
In the time domain, an IIR digital filter can be described by the difference equation:
$$ y(n) = \sum_{k=0}^{N} a_k x(n-k) – \sum_{k=1}^{M} b_k y(n-k) $$
A FIR digital filter is described by:
$$ y(n) = \sum_{k=0}^{N-1} h_k x(n-k) $$
I preferred the FIR filter when a linear phase response was required, because a linear phase avoids distortion of the shape of the decaying waveform. The IIR Butterworth filter was used when I needed a narrow pass-band with a small number of coefficients and a shorter computation time.
4.7 Feature Parameter Extraction
Two characteristic parameters were extracted from the processed acoustic signal for defect identification. The first is the resonance frequency. I extracted it by locating the maximum peak in the power spectrum within the selected frequency band.
The second parameter is the internal friction value. I obtained it from the time-domain decay envelope of the filtered signal. The decay of the envelope can be modeled as:
$$ A(t) = A_0 e^{-\beta t} $$
where \(\beta\) is the decay coefficient. The logarithmic decrement is then:
$$ \delta = \frac{\beta}{f_r} $$
or equivalently, using the peak amplitudes \(A_1\) and \(A_{n+1}\):
$$ \delta = \frac{1}{n}\ln\frac{A_1}{A_{n+1}} $$
To reduce the influence of random noise, I used a least-squares exponential fit to the envelope rather than relying on only two individual peaks.
5. Experimental Evaluation
5.1 Artificially Notched Specimens
To verify the sensitivity of the system, I first prepared sixteen cylindrical steel castings with identical dimensions and composition. Each specimen was 20 mm in diameter and 250 mm in length. Artificial notches were cut at different positions along the length: \(L/6\), \(L/4\), \(L/3\), and \(L/2\), where \(L\) is the total length. The notch width was 1 mm, and the notch depth was varied from 1 mm to 4 mm. I measured the resonance frequency of each specimen with the audio inspection system.
| Notch depth h (mm) | L/6 | L/4 | L/3 | L/2 |
|---|---|---|---|---|
| 0 | 8537 | 8540 | 8538 | 8537 |
| 1 | 8512 | 8518 | 8520 | 8517 |
| 2 | 8486 | 8480 | 8473 | 8468 |
| 3 | 8357 | 8344 | 8345 | 8339 |
| 4 | 8219 | 8205 | 8224 | 8199 |
The results show that the resonance frequency decreases as the notch depth increases. For a fixed notch depth, the position of the notch has a relatively small influence on the resonance frequency. The notch depth is therefore the dominant factor. Even a 1 mm deep cut caused a measurable frequency drop of roughly 20 Hz, which is much larger than the frequency resolution of 0.2 Hz. This demonstrates that the audio inspection system is extremely sensitive to a small sand foundry defect-like discontinuity.
I also extracted the internal friction values from the same specimens. The measured logarithmic decrement values are listed in Table 7.
| Notch depth h (mm) | L/6 | L/4 | L/3 | L/2 |
|---|---|---|---|---|
| 0 | 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 |
The internal friction values in the table are expressed in units of \(10^{-3}\). They increase consistently with the notch depth. This means that a notch, which simulates a crack, dissipates more mechanical energy during vibration. The combination of decreasing resonance frequency and increasing internal friction provides a reliable criterion for defect identification.
5.2 Real Casting Defects
To confirm that real metallurgical defects behave in the same way as artificial notches, I cast a group of spherical graphite iron specimens with different defects. The specimens had the same nominal dimensions as the steel bars. I tested four types of specimens: a sound casting, a casting with a surface crack, a casting with inclusions, and a casting with a shrinkage cavity. The measured resonance frequencies and internal friction values are summarized in Table 8.
| Defect type | Resonance frequency (Hz) | Internal friction \(Q^{-1} \times 10^{-3}\) | Normalized peak amplitude |
|---|---|---|---|
| Surface crack | 7498 | 3.242 | 0.82 |
| Inclusions | 7545 | 2.753 | 0.85 |
| Shrinkage cavity | 7537 | 3.568 | 0.78 |
| No defect | 7630 | 2.436 | 1.00 |
All three real defects caused the resonance frequency to decrease compared with the sound casting. The shrinkage cavity produced the highest internal friction value, followed by the surface crack and then the inclusions. The normalized amplitude of the resonance peak also decreased when a defect was present. This confirms that the audio frequency method can distinguish defective castings from sound ones by using the resonance frequency and internal friction as feature parameters.
During the experiments, I noticed that the exact resonance frequency of a sound casting can vary from batch to batch because of small changes in chemical composition and cooling conditions. Therefore, it is important to establish a reference baseline using a set of known sound castings. Once the baseline is established, a casting whose resonance frequency falls below the threshold or whose internal friction rises above the threshold can be rejected as containing a sand foundry defect.
The experiments also showed that the audio inspection system is much faster than conventional ultrasonic scanning. The total time from striking the specimen to obtaining the frequency and damping parameters was less than one second. The hardware is simple, robust, and inexpensive. These features make the system suitable for online inspection in a foundry production line.
6. Conclusions
In this work, I designed, constructed, and tested an audio frequency inspection system for the identification of casting defects. The system uses a small steel ball as a repeatable impulse source, an electret condenser microphone as a non-contact acoustic sensor, a computer sound card as the data acquisition hardware, and MATLAB for signal processing and feature extraction.
The main conclusions drawn from this study are as follows:
- The audio frequency inspection method is effective for detecting cracks, inclusions, and shrinkage cavities in castings. A casting with a sand foundry defect exhibits a lower resonance frequency and a higher internal friction than a sound casting of the same geometry and material.
- The resonance frequency is the most reliable feature for defect identification because it is highly sensitive to stiffness changes. Even a 1 mm deep artificial notch produced a measurable frequency shift.
- The internal friction value, derived from the logarithmic decrement of the free-decay signal, increases as the defect severity increases. It is a useful complementary parameter for distinguishing different types of defects.
- Using a computer sound card as the data acquisition device significantly reduces the hardware cost of the inspection system. The sound card, combined with MATLAB’s Data Acquisition Toolbox, provides sufficient sampling rate and resolution for acoustic measurements up to 16 kHz.
- The five-point smoothing, polynomial trend removal, and digital band-pass filtering are necessary preprocessing steps that improve the accuracy of the extracted resonance frequency and internal friction.
- The proposed system has the advantages of simple operation, low hardware cost, high detection speed, and good accuracy. It is therefore very suitable for online quality control of castings in a modern foundry.
In future work, I plan to extend the system to more complex casting shapes and to develop an automated classification algorithm based on the resonance frequency and internal friction parameters. The ultimate goal is a fully automatic, real-time audio inspection station that can screen every casting for the presence of a sand foundry defect before machining and assembly.