Current trends in laser material processing towards higher efficiency, improved quality as well as increased processing of challenging materials and components lead to the objective of integrated process monitoring and control. For industrial laser welding it is thus desirable to obtain data on process stability and to generate information about processing quality or even to build up inline process control.
The technical approach of a multi-sensor concept is presented, based on optical, acoustic and thermal sensors for the observation of laser welding processes for metallic materials. Basically, high-speed camera recordings and laser acoustic signals are used to classify laser welds of transmission components regarding typical defects. For this purpose, acquired process data were processed in a combined machine learning model, so that promising prediction accuracy can be achieved.
In current work, a more complex sensor setup is being developed to further improve the prediction quality and enable transferability to a wide range of applications.
The development of a cyber-physical system for AI-assisted data processing is presented, which will enable real-time response to dynamic process variations in the future. For this purpose, an optical system for 3D spatial dynamic beam shaping for laser welding and cutting is linked to the multi-sensor system. The data are processed via fast logic circuits (FPGA) into AI models using a cloud-based database and fed into a process control loop.
The technical setup, AI model and database structures are presented as well as first results of the sensor data evaluation.