Optimum parameters design for friction stir spot welding using a genetically optimized neural network system

Abstract

A method based on a genetically optimized neural network system (GONNS) is introduced to enhance the selection of the optimum parameters for the friction stir spot welding (FSSW) process. For a given FSSW setup, an artificial neural network (ANN) is designed with three process parameters as inputs and three process variables as outputs. The outputs of the ANN are selected as the weld's tensile force, plunging load, and process duration. Preliminary experimental results are utilized in order to train the ANN. After verifying the accuracy of the trained ANN, an optimization method based on the genetic algorithm heuristic search method is used to optimize the evaluation functions that are normalized functions of the ANN outputs. Eventually, the minimization of the evaluation functions yields the optimum ANN inputs (FSSW parameters) that are verified by additional experiments. Results affirm that the analytically obtained optimums of the FSSW parameters are valid and that, by utilizing these parameters, higher weld strength, lower plunging load, and shorter process duration are obtained.