Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Failure recognition in tailored blank welding by image processing

: Abels, P.; Kaierle, S.; Kreutz, E.W.; Poprawe, R.

Beyer, E. ; Laser Institute of America -LIA-:
Laser Materials Processing Conference 1998. Proceedings. Vol. 1
Orlando, Fla.: LIA, 1998 (LIA 85A)
ISBN: 0-912035-58-7
International Congress on Applications of Lasers and Electro Optics (ICALEO) <1998, Orlando/Fla.>
Laser Materials Processing Conference <1998, Orlando/Fla.>
Fraunhofer ILT ()
automatic optical inspection; automobile industry; feature extraction; flaw detection; image classification; image segmentation; laser beam welding

Laser beam welding of tailored blanks is growing rapidly and the forecast for the following years predict a further development. The automotive industry as main customer for tailored blanks requires 100% failure free machined parts. A big amount of work and costs in tailored blank welding results from the visual inspection of the welded seams to detect weld failures. Typical failures to be detected are holes, surface voids, undercut, and welding gaps. Usually, the inspection of the welded blanks is done visually including mainly manual work without any technical support. This means, the results of the optical inspection depend strongly on the operator and his varying ability with the consequence that failures can be overlooked or faulty classified. In order to relieve and to support the operator in his quality inspection work, the goal is to develop an optical seam inspection system which is able to detect and classify the mentioned defects in the welded seams. This is carried out by the use of image processing methods. Basic aspects of image processing methods applied to laser beam welding are described. The experimental setup and the different processing steps are shown. Processing steps are image pre-processing, extraction of characteristic features, and classification of failures. The results show that typical failures which have to be recognized by manual inspection can be detected reliably by the use of image processing methods.