In the field of laser welding, the lack of suited inspection means generall
y leads users to develop their own systems. From this observation, an adapt
able system to all welding conditions, dedicated to on line inspection of p
ulsed laser welding operations, has been devised. Its principle is very clo
se to Quality Control by Artificial Vision systems. It is based on the reco
rding and the treatment of signals resulting from laser-material interactio
n. The use of classification techniques and signal processing tools such as
the wavelet transform, introduce an innovating approach of on line inspect
ion. Thus, defects detection by discriminant parameters determination becom
es automatic and nonsubjective. The presented method allows to determine th
e weld spot position (pulse number) and the type of defects and to estimate
the reliability of the detection. This article presents the principle of t
he system and its validation on some classical defects in laser welding. Th
e final objective is, in particular, to detect penetration variations durin
g welding hollow parts, at the Valduc center of the French Atomic Energy Ag
ency. In most cases penetration can be easily estimated by observing (with
a photodiode or charge-coupled camera for example) the heating of the back
of the welded part. Unfortunately in case of hollow parts, the sensors can
only be placed at the outside! (C) 2000 Laser Institute of America. [S1042-
346X(00)00602-1].