Trends in Welding Home      Event Overview      ASM Homepage
Back to "Session 7 - Sensing, Control and Automation I" Search
    Back to Main Search

Tuesday, June 3, 2008 - 9:45 AM

Multi-sensor data fusion for real-time monitoring of GMAW

P. C. Boulware, Edison Welding Institute, Columbus, OH

In the industries which utilize robotic Gas Metal Arc Welding a desire has been long held for accurate and robust weld process monitoring.  This attraction to process monitoring stems from the ever-present need to reduce costly scrap and rework, decrease or eliminate expensive weld inspectors, and to rapidly solve day-to-day welding process problems.  Many methodologies and systems, from electro-optical sensory, to thermographic imaging, to light spectrometry, have been developed to comply with the want for better process control, but none have displayed the ability to robustly detect the wide range of defects that can be generated in the complex process of arc welding.  Each developed system has displayed its own advantages, but also has been exposed due to its incapacities.

Two examples of this occurrence are arc data monitoring and laser vision sensing.  Arc data monitoring has evinced the ability to predict such defects as gross porosity and lack of fusion, but struggles with identifying bead surface related issues such as excessive reinforcement or insufficient leg length.  The complete opposite tendencies are exhibited through laser vision sensing.  This technique provides no information of subsurface characteristics of the weld, but its inherent characteristic as a direct quality indicating process has resulted in a strong ability to detect weld surface quality issues.

This work looked to study the feasibility and benefit of fusing the arc data monitoring and laser vision techniques into one integrated system, with the hope taking advantage of the core competencies of each technique while negating their deficiencies.  A monitoring system was built which synchronized the measurements from both monitors.  Quantitative correlations were developed between each sensor reading with respect to measured quality through means of regression analysis.  These correlations were then built into quality differentiating algorithms which exhibited the capability to identify a multitude of defective occurrences in the real-time.


Summary: The fusion of multiple real-time quality monitoring sensors and a study of their effectiveness as an integrated system was examined. Arc data monitoring and laser vision sensors were synchronized, and correlations between each measurement were developed. These correlations were accommodated in algorithms which displayed the capability to identify poor quality.