In Line Process Monitoring

One system that monitors every laser weld, including the defects that never reach the surface.

The Challenge

Battery and e-mobility components such as busbars and motor hairpins require large numbers of high-quality laser welds. At these volumes, inspecting only a sample of welds can leave critical defects undetected.

Surface defects such as spatter and blowout can indicate process problems, while subsurface issues such as insufficient penetration, incomplete fusion, insufficient bonding width, or porosity may remain hidden until downstream inspection or even product failure. Manufacturers need a scalable way to inspect every weld without adding significant inspection effort or cost.

Our Approach

Fraunhofer USA developed a multimodal AI-based quality monitoring system for in-line laser welding inspection. A high-speed camera, microphone, and machine process data provide complementary information that AI models use to assess weld quality in real time.

The system can identify a broad range of surface, subsurface, and process-related anomalies and provides a Good, Bad, or Repairable quality decision for every weld. The modular architecture can be adapted to different welding applications, as demonstrated with both battery busbars and electric motor hairpins. Running the AI on edge hardware enables real-time decisions while keeping production data within the manufacturing environment.

The Impact

The system enables 100% in-line weld inspection, providing rapid quality decisions that allow potential issues to be identified before defective parts move further through production or reach the customer. By combining multiple sensing modalities, it can identify both visible defects and hidden weld quality issues that conventional visual inspection may miss.

This creates the potential to reduce quality escapes, scrap, rework, and inspection effort, while providing a scalable foundation for AI-based quality assurance across different welding processes and production lines.

This system tells you which welds are good. Our work on closed-loop process control goes a step further by using AI to correct the welding process while the weld is still being made.