A Data-Driven Framework for Axis Prediction and Springback Compensation in Spatial Metal Tube Bending

Spatial metal tubes with free-form curved axes are critical components in advanced industrial applications, such as aerospace and nuclear power systems, but their manufacturing is challenged by springback, which compromises axial precision. Traditional bending methods and theoretical models fall short in addressing the complex plastic deformation in spatial tube bending. This study introduces an innovative data-driven framework that integrates a Multi-layer Perceptron (MLP) for axis prediction with an enhanced Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) for springback compensation in Multi-roller bending (MRB). A finite element simulation sample library trains the MLP to capture the nonlinear relationship between geometric/process parameters and springback responses. The MOEA/D, enhanced with reference points and hybrid evolution strategies, optimizes forming parameters to minimize axis deviation. Simulation results demonstrate high prediction accuracy and significant springback reduction, with base circle radius errors reduced by 90% and pitch errors by 33%, offering a robust and efficient solution for precision manufacturing of spatial metal tubes.