Prediction of the quality of pulsed metal inert gas welding using statistical parameters of arc signals in artificial neural network

Abstract

One of the big challenges in welding is the prediction of weld-quality without destructive test. This work introduces an intelligent system for weld-quality prediction in a pulsed metal inert gas welding process based on the statistical parameters of the acquired current and voltage signals. Six process parameters and 10 statistical parameters of arc signals are used to describe various welding conditions. These process features obtained from a set of experiments are employed as input patterns to back propagation neural network and radial basis function network models to predict the corresponding weld qualities. The prediction errors show that the neural network model, which has been trained with the statistical parameters of arc signals along with the process parameters, gives superior prediction of weld quality as compared to that from a model developed with only the process parameters as its inputs.

Keywords:

• arc signal
• back propagation neural network
• bead geometry
• deposition efficiency
• distortion
• PMIGW
• radial basis function network
• weld strength