Elman and feedforward neural network based models for predicting mechanical properties of flow formed AA6082 tubes

The measurement of the mechanical properties of flow-formed products typically requires destructive testing, which may not always be feasible. To address this, the present study proposes a parametric predictive model for H30 aluminium tubes produced via flow forming, enabling the estimation of the final mechanical properties without additional physical trials. This approach offers designers the advantage of reducing the need for extensive experimentation. This study also facilitates the selection of optimal flow-forming parameters to achieve the desired mechanical properties. The key input parameters-feed speed (FS) ratio, axial stagger (AS), and infeed (IF)-were systematically varied, and the corresponding outputs-yield strength, ultimate tensile strength (UTS), and percentage elongation-were measured. Three predictive models were developed and evaluated: multivariate regression (MR), feedforward neural network (FNN), and Elman neural network (ENN). Among these, the FNN demonstrated superior predictive accuracy when validated against experimental data with maximum average prediction error of 7.45%, outperforming ENN and MR having 7.64% and 12.4%, respectively.