Improving geometric accuracy in incremental sheet metal forming using convolutional neural networks

Abstract

Single point incremental forming (SPIF) is a flexible sheet metal forming process. Unlike sheet metal stamping, SPIF does away with costly forming dies but instead uses a tool to incrementally form the sheet into the desired geometry. However, a key weakness of SPIF is its poor geometric accuracy, which is largely caused by material spring-back throughout the forming process. This paper presents a framework which minimises SPIF geometric error through optimisation of the forming toolpath. The approach utilises a trained convolutional neural network (CNN) to model the forming process, which provides greater flexibility and compatibility with a wide range of geometry. A geometric compensation algorithm was developed to compensate for the predicted spring-back. Experimental validation of the proposed framework demonstrated consistent accuracy improvements in both trained and untrained geometry. This paper highlights the viability of using CNNs in improving SPIF accuracy.