Nemat Zeynalzadeh - Rumaila Power Plant Manager, MAPNA Group, Basra, Iraq;
Mohammad Heidari Farsani - Head of Rumaila Power Plant Mechanic Group, MAPNA Group, Basra, Iraq;
Masoud Azadi Moghaddam - Ph.D. Graduate, Ferdowsi University of Mashhad, Mashhad, Iran;
Farhad Kolahan - Associate Professor, Ferdowsi University of Mashhad, Mashhad, Iran;

In this study an artificial neural network (ANN) basedmodeling and a heuristic based optimization procedureusing simulated annealing (SA) algorithm for modelingand optimization of flux assisted TIG welding processknown as activated TIG (A-TIG) have been addressed.In this study effect of the most important processvariables (welding current (C), welding speed (S)) andpercentage of activating fluxes (TiO۲ and SiO۲)combination (F) on the most important qualitycharacteristics (depth of penetration (DOP), weld beadwidth (WBW), and consequently aspect ratio (ASR)) inwelding of AISI۳۱۶L austenite stainless steel parts havebeen considered. To gather the required data formodeling and optimization purposes, box-behnkendesign (BBD) in design of experiments (DOE) approachhas been used. In order to establish a relation betweenprocess input variables and output characteristics, backpropagation neural network (BPNN) has been employedresults of which have been compared with regressionmodeling outputs. Particle swarm optimization (PSO)algorithm has been used for determination of BPNNarchitecture (number of hidden layers andneurons/nodes in each hidden layer). Simulatedannealing (SA) and PSO algorithms have beenemployed for process optimization in such a way thatdesired AR, minimum WBW, and maximum DOPachieved simultaneously. Finally, confirmationexperimental tests have been carried out to evaluate theperformance of the proposed method. Based on theresults, the proposed procedure is efficient in modelingand optimization (with less than ۴% error) of A-GTAWprocess.

Activated TIG (A-TIG) welding process,optimization, design of experiments (DOE), and simulatedannealing (SA) algorithm.