FRACTAL ANALYSIS OF METAL TRANSFER IN MIG/MAG WELDING.

We apply techniques of fractal analysis in order to classify metal-transfer mode in MIG/MAG (metal inert/active gas) welding, which are among the most commonly employed arc-fusion processes for industrial applications. We work with voltage and current time series obtained during welding, and evaluate statistical fluctuations present in those series by Hurst, detrended-fluctuation, and detrended-cross-correlation analyses, for each of three different metal-transfer modes: short-circuiting, globular, and spray. For a given total timespan of each series, curves of fluctuation as a funtion of the time-window size are processed by using pattern-classification techniques, such as principal-component analysis and Karhunen-Loève expansions. We obtain near 100% success rate for the classification, with timespans as small as 100 miliseconds, with a processing time of the same order. This suggests that our set of tools can be incorporated into an industrial welding apparatus in order to guarantee automatic correction of a process requiring a single metal-transfer mode. [ABSTRACT FROM AUTHOR]