Quality assessment for laser plastic welding using virtual temperature sensors, AI and computer vision – LAIserWeld

Laser transmission welding (LDS) is a well-established process for joining thermoplastics, particularly in series production. Currently, quality testing is often carried out randomly and destructively, leading to high material and energy costs in the event of errors. A non-destructive, inline-capable 100% inspection would significantly reduce these disadvantages. There is a correspondingly high need for early defect detection and intelligent process control.

The LAIserWeld project aims to develop an AI-based virtual temperature sensor to optimise process control, as well as a system for online quality assessment of the weld seam. The virtual sensor will precisely record thermal power conversion and serve as an input variable for the control system. At the same time, a camera-based test procedure is being developed that provides comprehensible visual feedback on weld seam quality. Both systems will be validated in a real production environment.

TMDT's research focuses on developing multimodal deep learning architectures for intelligent sensor data fusion. Different data sources, such as thermal imaging, CCD cameras, and process parameters, are combined to create robust models for temperature reconstruction and quality prediction. Particular attention is paid to processing temporally and spatially resolved sensor data in deep learning architectures.