Modelling and optimisation of welding parameters for multiple objectives in pre-heated gas metal arc welding process using nature instigated algorithms

Abstract

A back propagation neural network model was developed for the prediction of bead geometry (depth of penetration, height and width) in gas metal arc welded (GMAW) stainless steel plates. The plates are preheated before welding which enhances the weldment properties by reducing the cooling rates in the weld metal. The penetration of the weld has been taken into consideration for accessing the performance of the weld. The neural model developed was based on the experimental data. The experiments were conducted by intriguing different combination of five levels of welding parameters such as welding gun angle, welding velocity, welding current, shielding gas flow rate and wire feed rate. The welding experiments were conducted on AISI 202 stainless steel plates. The output of the developed model is compared with the experimental data; it is observed that the percentage error is less than 4% and correlation coefficient is found to be 0.985. Regression equation was developed between the input process parameters and bead geometry for doing optimisation. Multiple objectives constituted with maximum depth of penetration, minimum bead width and height within the bound of range of parameters and the optimisation was done based on the genetic algorithm using ‘C’ language.