Multiresponse optimization of friction stir welding by an integrated ANN-PSO approach

The optimization of the process performance by considering each process parameter independently is the simplest approach, but its industrial applications are restricted owing to its very limited validity. To overcome this problem, many techniques such as multi-objective optimization techniques, Artificial Neural Network (ANN), and regression analysis have recently received great attention. In this paper, a multiresponse methodology for predicting the main properties and suggesting the optimal process parameters for Friction Stir Welded joints is presented. The dataset applied in the analysis was collected through experimental FSW tests performed on a CNC machine considering different aluminum alloys, process parameters, and cooling fluids. The integrated methodology involves an Artificial Neural Network and a heuristic algorithm, the Particle Swarm Optimization (PSO) and allows to set both input and output values leaving to the PSO algorithm the identification of the other values able to minimize or maximize a predefined objective function, in this case the maximization of both the UTS and the hardness values of the joints. This means that the ANN is interrogated iteratively until the optimum is reached. For this reason, the proposed methodology can be defined as a double direction method. In particular, the double-direction method refers to the possibility of identifying the optimal values of the process parameters (inputs) starting from the desired specifications (outputs) considering that, in the production reality, processes can be constrained by several factors. The results show a good reliability of the approach, since it has been demonstrated that it is able to generate prevision with an error of less than 5%.