As the use of laser welding processes increases in production, visual, random inspection processes increasingly have to be replaced by 100% quality control. To implement real process control, the parameters for high dynamic processes, such as laser welding, have to be recorded and evaluated in real-time. This is accomplished by an innovative camera technology: optical Cellular Neural Networks (CNN). ((teaser finish))

In laser welding processes the material to be welded is evaporated by the high intensity of the concentrated laser beam. This high dynamic process often causes seam faults. Many of these seam faults can be dealt with by changing process parameters, such as the feed rate, laser output, or the focus.

For 100% quality control of laser-based welding processes in the long term a process control system is unavoidable. Because of the complex high dynamic processes a control system such as this must be based on very fast optical image processing. The required speed is several kilohertz. This is not possible with conventional image processing. For this reason, work is being carried out on an innovative camera architecture in which the intensity values of neighboring pixels are coupled into so-called "Cellular Neural/Nonlinear Networks" (CNN). Through intelligent neural networking CNN camera systems achieve a massive data reduction and thus enable image processing in the multi-kilohertz range.

For this purpose calculation and storage elements are integrated into the electronic circuits of the pixels. This enables fast, pixel-parallel image processing. Shifting image processing steps to the camera chip enables up to 10,000 images per second to be recorded and evaluated. To monitor the welding process the camera is integrated coaxially into the processing optics of the welding robot via a beam splitter. This allows the interaction zone to be observed from the top. "Complete penetration" has become established as a quality feature for the strength of the welded seam—this is suitable as a measure for inadequate and also too strong penetration. Initial results show that for this criterion images with an image rate of 10 kHz can be evaluated and corresponding control signals can be generated. This technology has been available only since the end of 2007 and, for the first time, enables laser welding processes to be controlled on the basis of quality features.

The new measuring system is a development by Fraunhofer IPM, Frankfurt University, and the Stuttgart-based Research Association for Laser Tools (FGSW). Based on the measurements of the welding process from the diagnostics center of the Research Association for Laser Tools (FGSW) and the CNN expertise of the Fraunhofer IPM and Frankfurt University a process control system for laser welding processes is to be developed and tested within the scope of this project. The project will be funded by Landesstiftung Baden-Württemberg.