A statistical approach for monitoring stochastic welding processes

This paper presents the theory of stochastic processes as applied to the analysis of gas metal arc welding data. A theoretical approach is presented and some of the commonly assumed hypothesis of process stationarity and ergodicity are verified for data collected from stable processes adjusted for short circuiting and spray modes of metal transfer. Typical fluctuations of the welding voltage and current were calculated for both modes of metal transfer studied. The results showed a significant difference on the magnitudes of the intensity of fluctuation, pointing out that the short circuiting mode of metal transfer is much less stable than the spray mode, in a statistical sense. The long-time behaviour of the randomly fluctuating process was investigated by using auto-correlation functions and power spectral analysis. These analyses produced correlation times, which could be used for optimising the sampling frequencies used for process monitoring. Also, the spectral analysis showed clearly the characteristic frequencies of the process, giving an indication of frequency of metal transfer. The data analysed for dip mode of metal transfer indicated a characteristic frequency to exist around 60 Hz and the data analysed for spray mode showed a sharp characteristic frequency at 300 Hz and harmonics at 600 and 900 Hz.