AKoS – Acoustic inspection of weld seams of safety-critical components as part of quality assurance

Initial situation

For companies that manufacture safety-critical components, the quality of welds is paramount. Conventional inspection methods are often time-consuming and costly because they are carried out once the welding process is complete. There is therefore a clear need for a reliable, non-contact and non-destructive method of detecting welding defects early on, directly during the process.

Innovation for Industry: Acoustic Inspection of Welds and Welding Processes

In the "AKoS" project, which was funded by the BMBF, the Fraunhofer IDMT developed acoustic methods for inspecting welds in various welding processes. These methods are based on analyzing airborne sound in the audible frequency range (up to 12.5 kHz) using machine learning algorithms.

In feasibility studies, changes were made to the welding processes to simulate defects and wear for experimental purposes. These changes included reducing the shielding gas coverage, introducing contaminants by applying oil, altering the welding speed and distance, and using different tools. The process was intentionally designed to produce specific irregularities and quality defects. These defects included porosity clusters, cavities, and variations in gap dimensions. Other defects were due to insufficient penetration, tool wear, or shielding gas anomalies.

Successful Feasibility: AI Can Detect Welding Defects

After the experiments in the “AKoS” project were completed, it was evident that analyzing process sounds in TIG, MSG, and FSW processes could reveal information about weld quality and process stability.

Following the project’s conclusion, the ML algorithms developed by the consortium will be further refined and adapted to detect additional types of irregularities. The AKoS project paved the way for implementing AI in arc-based additive manufacturing processes and friction stir welding.

Benefits of Acoustic Monitoring

Acoustic testing methods enable companies to monitor welding processes in real time, detect defects early, and automate quality assurance. Additional advantages include:

• Avoiding costly rework and reducing scrap
• Early detection of tool wear, which reduces unplanned downtime
• Greater process reliability with reduced testing effort
• Seamless integration into the ongoing manufacturing process without additional cycle time

This technology increases efficiency, reduces costs, and improves quality in industrial manufacturing overall.

Responsibilities of Fraunhofer IDMT

• Development of energy-efficient machine learning algorithms for acoustic event detection for single case analysis
• Development of a dynamic framework, with transferability to various joining processes