Steel Drum Leak Detection Using Acoustic Emission Signals
Chen Yangyang
Abstract: This study aims to develop a method for detecting leaks in steel drums using acoustic emission signals. A system was built using an acoustic emission sensor, preamplifier, data acquisition card, and computer to collect the signals. The leakage diameter was set at 0.2 mm, and the system identified the frequency characteristics of the signal without any leakage. The collected acoustic emission signals were decomposed using wavelet packet analysis, and energy features from three characteristic frequency bands (16–30 kHz, 33–47 kHz, and 95–102 kHz) were extracted as input vectors for a support vector machine (SVM). After training the SVM, it was tested to determine if the drum had a leak. The results showed a 100% accuracy rate, proving that this method can effectively detect leaks with a diameter of 0.2 mm or larger. Keywords: steel drum leak; acoustic emission; wavelet packet energy; support vector machine
Steel drums are widely used in industries such as chemicals, petroleum, light manufacturing, and food due to their high strength, corrosion resistance, and reliability. Real-time leak detection is crucial for ensuring safety and quality. Traditional methods like the soap test and sound method have limitations. In this research, acoustic emission technology was employed to detect leaks. When a leak occurs, fluid exits under pressure, causing stress waves on the drum's wall. These waves carry information about the leak and propagate at the speed of sound. According to simulation and theory, larger leaks produce more distinct signals. An acoustic emission sensor was installed on the drum wall to capture these stress waves. Features were extracted using time-domain, frequency-domain, and time-frequency-domain analysis, and pattern recognition was applied to classify the leak status. Through wavelet packet energy extraction and SVM classification, a 0.2 mm leak was successfully detected.
1 Wavelet Packet Decomposition and Support Vector Machine Theory
1.1 Wavelet Packet Decomposition Energy Extraction Principle
Wavelet packet decomposition is an advanced form of wavelet transform that allows for multi-resolution analysis of signals. It breaks down the signal into independent frequency bands without overlap, enabling detailed analysis of different frequency components. The process involves decomposing the signal into multiple layers, calculating local energy for each band, and constructing a feature vector based on these energies. This provides a robust way to extract meaningful features from complex signals.
1.2 Support Vector Machine Theory
Support Vector Machines (SVMs) are powerful tools for classification and pattern recognition, especially in small sample scenarios. Based on statistical learning theory, SVMs map input data into a high-dimensional space where a hyperplane separates different classes. By selecting appropriate kernel functions—such as linear, polynomial, radial basis, or sigmoid—the model can accurately distinguish between different states. This makes SVM ideal for applications like leak detection, where accurate classification is essential.
2 Steel Drum Leak Test and Analysis
2.1 Test Plan
The experiment involved both leak-free and 0.2 mm leaking steel drums. The setup included a transmitting sensor, preamplifier, data acquisition card, and computer. The drum had a 1 mm wall thickness, and the sensor used was the SR150 with a center frequency of 150 kHz. The preamplifier had a gain of 40 dB, and the acquisition card operated at 1 MS/s. Gas was introduced into the drum, and pressure was maintained at 0.15 MPa. Data was collected every minute after stabilization.
2.2 Leak Feature Extraction
The collected signals covered a wide frequency range (0–500 kHz). By comparing the spectra of leaking and non-leaking drums, three key frequency bands (16–30 kHz, 33–47 kHz, 95–102 kHz) were identified. The db9 wavelet was used for 5-layer decomposition, and specific coefficients were extracted. The energy of these bands was calculated and compared. Results showed that the energy in the leaking state was significantly higher than in the non-leaking state, confirming the effectiveness of this approach.
2.3 Leak Identification
A total of 20 sets of data were analyzed, with energy values from the three frequency bands serving as the feature vector. The SVM classifier was trained using 50 samples and tested on 10 others. The radial basis function kernel was selected, and the model achieved 100% accuracy. This demonstrates that the combination of wavelet packet energy and SVM is highly effective for detecting even small leaks.
3 Conclusion
This study shows that acoustic emission signals from steel drums do not occur at a single frequency but appear across a frequency band. By analyzing the energy in specific bands and using SVM for classification, a 100% accuracy rate was achieved in detecting 0.2 mm leaks. This method offers a reliable and efficient solution for real-time leak detection in industrial settings.
Kraft Paper Bag,Brown Kraft Paper Stand Up Bag,Paper Tote Bag
EVER GREEN(ZHEJIANG)NEW MATERIAL CO., LTD / KUNSHAN GREENPACK CO.,LTD , https://www.changjingpack.com