Multi-sensor based prediction of metal deposition in pulsed gas metal arc welding using various soft computing models

The deposition efficiency is an important economic factor in welding. A multitude of uncontrollable factors influence the metal deposition, which indicates the necessity of robust sensors with an intelligent system to monitor the process in real time. This paper attempts to develop artificial neural network (ANN) models to predict the weld deposition efficiency using the welding sound signal along with the welding current and the arc voltage signals in pulsed metal inert gas welding. Three different implementations of ANNs have been used: gradient descent error back-propagation, neuro-genetic algorithm and neuro-differential evolution. The results indicate that the sound signal kurtosis, used in conjunction with the current and the voltage signals, is a reliable indicator of deposition efficiency.

Figure optionsDownload as PowerPoint slideHighlights
► The metal deposition efficiency in arc welding is an economic factor affected by arc stability.
► Effective monitoring of arc stability in P-GMAW is possible using various sensor based features.
► RMS value of voltage and current sensors’ signals along with arc sound kurtosis are highly correlated with weld deposition.
► Neuro-DE was found to be the best predictor among BPNN, neuro-GA, and neuro-DE.