A comparative analysis of forecasting surface hardness in various aluminum friction stir welded joints: FEM-ANN hybrid versus ANN-PSO-integrated approaches

Sara Bocchi,Mariangela Quarto,Gianluca D’Urso,Claudio Giardini

DOI: https://doi.org/10.1007/s00170-024-13770-x

IF: 3.563

2024-06-19

The International Journal of Advanced Manufacturing Technology

Abstract:This paper compares two predictive methodologies for optimizing friction stir welding process and predicting mechanical properties in the welding zone. The first approach combines finite element method (FEM) with artificial neural network (ANN) techniques to forecast hardness variations, reducing reliance on extensive experiments. The second approach integrates ANN and particle swarm optimization (PSO) to predict properties and optimal FSW parameters, particularly emphasizing hardness maximization. Results indicate the superior predictive accuracy of the combined ANN-PSO approach, offering streamlined implementation suitable for practical applications. Indeed, in the FEM-ANN approach, the average error is around 16.50%, while the ANN-PSO approach predicts hardness with an average error of about 4.80%. Moreover, error increases closer to the center line in both approaches, more prominently in the FEM-ANN approach. However, it should be noted that successful implementation of the ANN-PSO model requires a substantial historical dataset for ANN training. On the other hand, the FEM-ANN approach demands relatively less experimental effort but entails longer computational times. Overall, these comparative insights guide the selection of the most suitable approach based on resource availability and project-specific requirements, providing valuable guidance for optimizing FSW process and predicting properties while considering resource constraints.

engineering, manufacturing,automation & control systems