An application of spectral analysis to weld penetration detection

High quality welded seams require the material to be joined to be fully molten from the front surface to the back. Most quality control systems require access to the back surface of the join. In many situations, for example pipe joining, only the front surface is accessible. This paper describes a front surface penetration detection technique without the introduction of transducers into the harsh environment of the arc. The arc itself is used to both excite oscillations of the penetrated pool and to monitor the surface displacement. A least squares Prony model is developed for this surface oscillation in which the existence of decaying sinusoids within the motion signify a fully penetrated pool. Arc noise contributions are separated from signal components by Singular Value Decomposition (SVD) and use of both forward and backward prediction in the modelling. Performance of the technique is assessed by application to real arc-pool data