Tube geometry prediction in rotary draw bending process using Random Forest regression

Abstract

Abstract In this study, we introduce a data-driven learning model for predicting the arc geometry of bent steel tubes in rotary draw bending processes, through the use of data from finite element simulations. In 162 simulations, machine tool forces, movements, and the resulting tube geometry data were collected based on pre-defined machine setups. To predict the geometry, we trained a model using Random-Forest regression which could predict the geometry with RMS errors below 0.19 mm for a 22 mm tube diameter. The random Forest model also allows to investigate data features according to their predictive power, highlighting promising features such as the mandrel extraction and the collet boost. We argue that such prediction models could assist in finding better mould designs.