?
Application of ML methods to predict residual stresses and strains after wire drawing process
It is well known that residual stresses and accumulated deformations during drawing processes can influence mechanical properties of the resulting products. This paper proposes the use of machine learning methods, such as artificial neural networks (ANN) and polynomial regression, to gain insight into the nature of these distributions across the cross-section of round wires. The necessary data sets were generated using finite element simulations (FEM), and several calculations were performed to select the optimal model configuration for these methods. To select the best quality metric, metrics such as mean absolute percentage error (MAPE), mean square percentage error (MSPE), and R-squared (R2) were evaluated. A statistical analysis was also conducted using Friedman’s and Nemenyi’s tests to compare the two methods. It was found that both ANNs and polynomial regression can be used to predict residual stress and strain distributions. However, it is more preferable to use ANNs for the former and polynomial regression for the latter because; in this case, the smallest error in the obtained predictions is achieved.