ENGINEERING JOURNAL CHIANG MAI UNIVERSITY

Volume 27, No. 02, Month AUGUST, Year 2020, Pages 201 - 215

Application of artificial neural network and response surface methodology for mechanical property prediction and optimization in aluminum hull welding for alloy 5083 grade

Prachya Peasura and Suthipong Sopha

Abstract Download PDF

This research was aimed to determine a mathematic model using artificial neural network (ANN) for predicting the of mechanical property and optimization using response surface methodology (RSM) in the aluminum hull 5083 grade with gas metal arc welding (GMAW) process. The following welding factors were studied: the welding current, voltage and travel speed. The resulting welding samples were examined using tensile strength tests hardness test which were observed microstructure with scanning electron microscopy (SEM) and determine a suitable mathematic model. The research results reveal that using a ANN model with the proposed mathematical model, which tensile strength and hardness represents 3 neurons for the input layer 10 neurons for hidden layer 1 10 neurons for hidden layer 2 and 1 output neurons (3-10-10-1). The Levenberg-Marquart training algorithm was also train for weight and bias network. The neuron of log-sigmoid for input layer, tan-sigmoid for hidden layer1 and 2 purelin for output layer activation function was assigned. The mean square error (MSE) and coefficient of determination (R2) for tensile strength and hardness predict was showed that of 0.454 and 0.386 respectively. The optimization of GMAW parameters were welding current of 220 amp, voltage of 26 V and 10 mm/sec travel speed. The welding conditions which have the optimization condition was showed that metal compounds Al (Fe, Mn) Si type could be that small size with distribute intensity in heat affected zone, which results in increased welding material high tensile strength and hardness.

Keywords

Artificial Neural Network, Response Surface Methodology, Mechanical Property, Aluminum Hull Welding

---