Prediction of tensile strength of weldments in gas metal arc welding process through new intelligent approaches

Mehran Seifollahi¹ Mohammadreza Ghavi²

1) MSc, Department of Civil Engineering, University of Tabriz, Tabriz, Iran.
2) MSc, Department of Mechanical Engineering, University of Tabriz, Tabriz, Iran.

IOI Code : XDAC-EFBAD
Publication : 2nd- Scientific Research Conference Mechanics, Electrical, Computer and Engineering Sciences(mececonf.com)
Abstract :
    The prediction of tensile strength of weldment in gas metal arc welding process is very important. Due to the progressive industrial developments and the requirement of having high quality weldments, the estimation of tensile strength is necessary because of its complicacy. Therefore, in this research 53 samples of experimental data were implemented to predict the tensile strength of weldment using new intelligent approaches. In fact, the applied approaches (i.e. Artificial Neural Network (ANN), Wavelet-ANN (W-ANN), ANN which optimize with Particle Swarm Optimization (ANN-PSO) and Principal Component Analysis-ANN (PCA-ANN) and Adaptive Neuro-Fuzzy Interface System (ANFIS)) are a hybrid model of practical mathematical theory for decompose and assemble signals during portray their properties and features to ANN. Several parameters such as frequency, duration time and voltage of pulse, the rate of consumption of filler metal, welding speed, main voltage, the effective value of current and voltage were considered as the input variables. The obtained results showed that the combination of wavelet and ANN has better performance and higher precision in the prediction of tensile strength of weldment than the other approaches. The W-ANN model achieved 97.7% and 4.03% for correlation coefficient and mean absolute errors, respectively. Namely, the W-ANN method is more capable than ANN, ANN-PSO, PCA-ANN and ANFIS for prediction of tensile strength of weldments.

Keywords : weld strength new intelligent approaches adaptive neuro fuzzy interface system wavelet theory principal component analysis