Regression-based compensation of part inaccuracies in incremental sheet forming at elevated temperatures

Incremental sheet forming is a sheet forming process for small lot sizes due to its dieless principle. One of its process variants includes local heating of the sheet to counteract some of the process restrictions (formable materials, forming forces, achievable deformations). Although forming at elevated temperatures provides various advantages, the geometric accuracy of the formed part remains low due to shrinking effects caused by local heating and cooling. This publication presents a data-driven approach where process data is gathered and used in regression learning to predict the geometric accuracy resulting from the shrinking effects. To successfully apply regression learning, a big amount of process data is needed covering a wide range of possible process states. Therefore, a specific experimental series, consisting of 54 individual forming experiments, is designed and carried out. Based on the 3D digitization of the formed parts, a process database is built up comprising 408,296 records, each representing a toolpath point. This process database is used to train 19 different regression models. The performance of their ability to predict the geometric deviations is investigated. A compensation approach is presented that improves the geometric accuracy through a prediction-based modification of the toolpath. Validation experiments demonstrate the improvement of the geometric accuracy of the formed part and the generalizability of the approach.