Zhuojin Pipe Fitting Co., LtdMIG Welding Penetration Monitoring via Arc Sound Wavelet Energy Feature Extraction
Literature Overview
This paper by Liu Lijun, Lan Hu, Wen Jianli, and Yu Zhongwei, published in the Transactions of the China Welding Institution (2010, Vol. 31, Issue 1, pp. 45-49), presents a signal processing methodology for real-time monitoring of weld penetration status using arc sound signals during MIG welding. The work is supported by multiple funding sources including the Ningbo Natural Science Foundation (2008A610031) and the Heilongjiang Provincial Natural Science Foundation (E2007-01). The research addresses one of the most persistent challenges in automated welding: non-destructive, in-process verification of weld penetration without interrupting the welding cycle.
Technical Methodology
The authors developed a complete signal processing pipeline for arc sound-based penetration diagnosis. The methodology proceeds through several distinct stages:
Stage 1: Arc Sound Acquisition and Noise Reduction
Arc sound signals were captured during MIG welding using a microphone-based sensing system. The raw signal contains both useful penetration-related information and significant noise from ambient welding environments. Wavelet denoising was applied as the first processing step to extract the meaningful signal components from the noise floor.
Stage 2: Frequency Band Extraction via Wavelet Packet Shift Algorithm
A critical innovation in this work is the use of a wavelet packet shift algorithm for frequency band extraction. The authors identified that classical wavelet packet decomposition algorithms suffer from frequency aliasing caused by decimation sampling during the decomposition process. The shift algorithm eliminates this aliasing, providing more accurate frequency band energy measurements. This is a significant technical contribution because frequency aliasing in classical methods can lead to incorrect feature extraction and consequently erroneous penetration diagnosis.
Stage 3: Feature Vector Construction
After frequency band extraction, the energy in each frequency band was computed and assembled into a feature vector. These feature vectors serve as input for subsequent classification algorithms (not detailed in this paper but implied as future work) to distinguish between different penetration states.
Signal Processing Parameters
| Processing Step | Method | Purpose |
|---|---|---|
| Noise reduction | Wavelet denoising | Remove high-frequency noise |
| Frequency extraction | Wavelet packet shift algorithm | Eliminate aliasing |
| Feature construction | Band energy calculation | Build diagnostic feature vector |
| Signal type | Arc sound (acoustic) | Non-contact penetration monitoring |
Engineering Practice Integration
In pipe manufacturing, particularly for API 5L line pipe and ASME B31.3 process piping, weld penetration verification is typically performed through post-weld non-destructive testing (NDT) methods such as radiographic testing (RT) or ultrasonic testing (UT). However, these methods are inherently post-hoc and cannot prevent defects from occurring. The arc sound monitoring approach presented in this paper offers the potential for real-time, in-process penetration feedback, which could be integrated into automated pipe welding cells.
The practical implementation would require:
- Robust microphone mounting that can withstand the thermal and electromagnetic environment near the arc
- Signal conditioning hardware capable of real-time wavelet processing
- A classification algorithm trained on sufficient samples of penetrated and non-penetrated welds
Key Technical Challenges
The arc sound signal is inherently noisy due to the stochastic nature of metal transfer in MIG welding. The distinction between spray transfer and short-circuiting transfer produces fundamentally different acoustic signatures, which must be accounted for in the feature extraction process. The authors specifically analyzed the spectral characteristics of both spray transfer and short-circuiting transition arc sounds, demonstrating that the methodology is applicable across different transfer modes.
Another challenge is the spatial relationship between the microphone position and the weld pool. The acoustic signature varies with distance and angle from the arc, requiring either fixed sensor geometry or adaptive signal processing to account for variations.
Study Insights and Implications
This work represents an important step toward closed-loop weld quality control in automated welding systems. The wavelet packet shift algorithm is a particularly valuable contribution because it addresses a fundamental limitation of classical wavelet methods that has been overlooked in many prior studies. For pipe welding automation, the ability to detect penetration status in real-time could significantly reduce rework rates and improve first-pass quality. However, the transition from laboratory signal processing to production-grade monitoring systems requires substantial engineering effort in sensor design, signal conditioning, and algorithm robustness under varying welding conditions.