Abstract
Penetration estimation is a prerequisite for the automation of backing welding based on vision sensing technology. However, arc interference in the welding process complicates the extraction of weld pool characteristic information. This paper investigates the relationship between weld pool geometry parameters and penetration during backing welding. A passive vision sensing system was established to extract parameters such as area (AWP), maximum width (MWWP), and length (MLWP). Using multivariable linear regression, mathematical models were developed to estimate back-bead width, providing a robust solution for real-time penetration monitoring in Gas Metal Arc (GMA) welding.
Introduction
In automated manufacturing, "backing welding"—the initial pass in a joint—is a critical step. Proper penetration is essential: insufficient penetration leads to weak joints, while excessive penetration causes "burn-through." The primary challenge in automating this process is the intense brightness and chaotic nature of the welding arc, which obscures the molten pool. While complex 3D laser mapping systems exist, they are often too bulky for industrial environments. This research explores a simpler, 2D passive vision-based approach to predict penetration by analyzing the geometry of the weld pool.
Methods and Materials
The researchers developed a passive vision sensing system that utilizes the light from the welding arc itself, filtered through a narrowband 650nm filter to minimize glare. The experimental setup used Q235 steel plates with V-groove butt joints.
Feature Extraction
Using the HALCON image processing library, the system segments the weld pool and extracts six key geometric parameters:
- AWP: Area of the weld pool.
- MWWP & MLWP: Maximum width and length.
- HLWP: Half-length.
- LWR: Length-to-width ratio.
- ACA: Advancing contact angle (α).
Statistical Modeling
A dataset of 532 experimental samples was generated by varying welding current (110A to 160A) and speed (15 to 40 cm/min). Multivariable linear regression was then applied to link these top-side geometric features to the back-bead width (the physical indicator of penetration).
Key Results and Insights
The study revealed that the weld pool does not simply grow in size; its proportions change significantly as it reaches full penetration.
The "Sweet Spot" for Penetration
Optimal penetration occurs when the pool area (AWP) falls within a specific range, the length-width ratio (LWR) is near its maximum, and the advancing contact angle (ACA) is at its minimum. Statistical analysis identified the Area of the Weld Pool (AWP) as the most significant predictor of penetration, with length (MLWP) serving as a secondary indicator.
Model Accuracy
The regression models demonstrated high predictive accuracy. For instance, at a current of 140A, the model estimated a back-bead width of 2.86 mm, closely matching the measured 2.67 mm. The models achieved an Adjusted R-squared of 0.767, confirming that the selected geometric parameters effectively capture the variation in weld penetration.
Conclusion
This research provides a practical solution for real-time penetration estimation in GMA backing welding. By decoupling the effects of current and speed, the authors developed specific equations that accurately predict weld quality from simple 2D images. Future work will focus on addressing the "coupling effect"—where current and speed vary simultaneously—potentially through non-linear models or neural networks, and integrating these models into closed-loop control systems.
Summary updated