Modeling and prediction of weld strength in ultrasonic metal welding process using artificial neural network and multiple regression method

Abstract

In this paper, multiple regression analysis (MRA) and artificial neural network (ANN) models are used to predict the weld strength of copper to copper joints produced by ultrasonic metal welding process. The process parameters of the models include weld pressure, weld time and amplitude of vibration; whereas, the output parameter is weld strength. Experiments are conducted as per Taguchi design of experiments. The results obtained from experiments are used in the multiple regression analysis and artificial neural network to model the ultrasonic metal welding process. Correlation coefficient is used to find out the adequacy of these models for predicting the weld strength. The performances of multiple regression analysis and back propagation artificial neural network (BP–ANN) models are compared in terms of Mean Prediction Error. The results of this study revealed that ANN model predicts more accurate results than the conventional regression models.