Notes on derivation of streamline fields using artificial neural networks for automatic simulation of material forming processes

TL;DR

This paper presents an automatic neural network-based method to predict material flow patterns and force requirements in deformation processes, offering a fast alternative to complex mathematical models and finite element simulations.

Contribution

It introduces a novel ANN approach to derive streamline fields and estimate forces in material forming, enhancing simulation speed and accuracy.

Findings

ANN predictions matched finite element results for force estimates

The method effectively predicts flow patterns in hot deformation processes

Approach enables rapid simulation of steady-state forming processes

Abstract

Here I introduce an automatic approach to determine the material flow patterns during deformation process using artificial neural networks (ANN). Since deriving and calibrating complex mathematical models for prediction of power requirements in each individual deformation process is inconvenient, the generality of using streamline field method has been limited. I propose an automatic approach to build and calibrate streamlines with ANN. The coordinates of specific points within the deformation region were used as input and the stream function values on the points were used as output dataset in ANN training algorithm. A specific neural network architecture was then implemented to predict the flow patterns of the deforming body by an ANN-based streamline equation. At the next step, the upper bound theorem was incorporated to estimate the force and power requirements for equal channel angular extrusion (ECAE) of a composite system under hot deformation. For verification of the results, a finite element software was utilized in parallel to investigate the accuracy of the proposed approach for estimation of force requirements. The predicted force requirements under various temperature ranges with ANN based technique were consistent with that of the finite element predictions which demonstrates the accuracy of the proposed approach. The proposed approach may be appropriate for fast simulation of a wide range of steady state material forming processes.