UP-KI-Data: AI-Assisted Process Monitoring for Multi-Wire Submerged Arc Welding

To enable the early detection of production defects in multi-wire submerged arc welding, the UP-KI-Data research project is developing an AI-based assistance system for real-time process monitoring. By intelligently combining electrical, vibroacoustic and thermographic process data, process instabilities can be detected at an early stage and weld quality can be assessed during production. The project thus lays the foundation for the next generation of digital, resource-efficient and reproducible welding production.

AI-based analysis of process data

Submerged arc welding (SAW) is one of the most important welding processes for the cost-effective production of thick-walled steel structures. In multi-wire SAW in particular, the use of several arcs enables high deposition rates and high productivity. At the same time, however, the requirements for stable process control and reproducible weld quality increase.

Currently, the quality of submerged arc welds is predominantly assessed by downstream non-destructive testing methods, such as ultrasonic or radiographic testing. Consequently, defects are often detected only after completion of the welding process.

The project combines different types of process information for the first time. These include:

• welding current and arc voltage
• wire feed speed
• vibroacoustic signals (structure-borne sound)
• thermographic process data
• technological welding parameters

Characteristic features are extracted from the synchronously acquired sensor data and evaluated using machine-learning methods. The objective is to develop an AI model capable of identifying relationships between process signals and the resulting weld quality and automatically assessing the welding process.

Digital quality monitoring in real time

The developed system is intended to support welding system operators during production and enable continuous assessment of process stability. Process disturbances, such as unstable arcs or unfavorable parameter combinations, can be detected at an early stage before they result in cost-intensive welding defects.

In the long term, the generated database will provide the basis for:

• AI-assisted quality prediction
• adaptive process control
• digital quality documentation
• complete traceability of welding processes 

Research focus

The project particularly addresses the following research topics:

• multi-channel data acquisition in multi-wire submerged arc welding
• correlation between process signals and weld quality
• development of AI models for quality assessment
• identification of process instabilities and welding defects
• integration of intelligent process monitoring into industrial welding systems 

Benefits for industry and SMEs

The project results will contribute to the digitalization of industrial welding production. Small and medium-sized enterprises in particular can benefit from automated process monitoring, as it can reduce scrap, rework and inspection efforts.

Potential fields of application include:

• large-diameter pipe manufacturing
• offshore wind energy structures
• shipbuilding
• pressure vessel manufacturing
• hydropower systems
• heavy steel and mechanical engineering 

Benefits for users

• Increased process reliability through continuous monitoring of relevant welding process data
• Early detection of process deviations and potential welding defects during welding
• Reduction of scrap and rework through timely correction of process disturbances
• Potential reduction in non-destructive testing effort through AI-assisted quality assessment based on process data
• Digital quality documentation with complete traceability of welding parameters
• Improved cost efficiency through stable processes and reproducible weld quality
• Straightforward integration into existing automated submerged arc welding systems
• Basis for Industry 4.0 applications through data-driven process optimization and intelligent assistance systems