Hadi Tagimalek - Faculty of Mechanical Engineering, Semnan University, Semnan, Iran
Mohammad Reza Maraki - Department of Materials and Metallurgy Engineering, Birjand University of Technology, Birjand, Iran
Masoud Mahmoodi - Faculty of Mechanical Engineering, Semnan University, Semnan, Iran
Majid Azargoman - Faculty of Mechanical Engineering, Semnan University, Semnan, Iran

The friction stir welding method is one of the solid-state welding methods for non-homogeneous metals. In this study, the ۵XXX series aluminum sample and pure copper are subjected to four passes friction welding process and then the mechanical and metallurgical properties of the welded samples are compared with the prototype. For this purpose, the effect of welding parameters including rotational speed, forward speed and pin angle of the tool is tested by the full factorial method. In this process, hardness estimation and tensile testing are based on input process parameters in order to obtain mechanical properties is an important issue. For this purpose, a mathematical model of mechanical properties must be defined based on the input process parameters. Due to the complex nature of the effect of input process parameters on mechanical properties, this modeling is a complex mathematical problem in which the use of supervised learning algorithms is considered as an efficient alternative. In this paper, a new combination of Relevance Vector Machine (RVM) and Support Vector Machine (SVM) is presented which has a higher degree of accuracy.

Friction stir welding, Hardness, Support vector machine, Relevance Vector Machine