In this paper, we present the basic approach to quality classification and process control of laser spot-welded connections. The process is observed with several sensors, which acquire data during the process. One of the most crucial aspects in the system is the signal processing, through the extraction of signal features. Laser welding experts have co-operated in selecting a dedicated set of time-domain based features.
The "К-Nearest Neighbour" classifier has been used in the project, because of its simple training strategy. It operates basically on a large set of reference patterns of features with associated manual classification. Evolutionary algorithms are used to select the appropriate features for successful classification. The same set of features was used to build a process model based adaptive controller, allowing off-line process parameter check.
A real-time control system is developed using fast feature extraction to evaluate the initial process situation and change the laser power within the pulse for optimum result.
Experiments indicate that the goal of reliability improvement by a factor of ten can be achieved for a lot of laser welding processes. Intelligent control paves the way to introduce laser welding on products which could not be laser welded up to now.