Predicting features in complex 3D surfaces using a point series representation: A case study in sheet metal forming

This paper presents an integrated framework for learning to predict geometry related features with respect to 3D surfaces. The idea is to use a training set of known prediction values to create a model founded on local 3D geometries associated with a given surfaces so that predictions with respect to a new “unseen” surfaces can be made. The local geometries are represented using point series curves. Two variations are proposed: (i) discretised and (ii) real number. To act as a focus for the work a sheet metal forming application is considered where we wish to predict the errors that are introduced as a result of applying a forming process. Given a desired surface T, the surface T ′ actually produced as a result of the sheet metal forming process is affected by a phenomena called Springback (the feature we wish to predict). The proposed process has been evaluated using two flat-topped pyramid shapes and by considering a variety of parameter settings. Excellent results have been obtained in terms of accuracy and Area Under ROC Curve (AUC).