Robust active surface design for multi-stage sheet metal forming processes based on data- and calculation-based equivalent modelling of component springback

Project Overview

Principle Investigators:	Prof. Dr.-Ing. Dr. h. c. Michael Weyrich Prof. Dr.-Ing. Dr. h. c. Mathias Liewald MBA
Project Team:	Pascal Heinzelmann M.Sc. Sebastian Baum M.Sc.
Research Institutions:	Institute for Forming Technology (IFU), University of Stuttgart Institute for Automation Engineering and Software Systems (IAS), University of Stuttgart
Semi-finished Material(s):	Sheet metal (DP600)
Manufacturing Processes:	Deep drawing and trimming

Motivation

Development of the virtual process environment for deep drawing
Determination of the influences of material and process parameters on the springback and dimensional deviations of complex drawn parts
Optimisation of the effective surface design of 2-stage forming processes

Aims

The aim of this project is the development of a novel design method for active surfaces of deep-drawing moulds based on a data- and calculation-based surrogate model. The surrogate model is built using Generative Adversarial Networks or Denoising Diffusion Probabilistic Models and is intended to learn and map the relationships between component geometry, the existing process parameters and the component springback. In addition, a delta modelling is set up based on the surrogate modelling, which is also intended to learn the deviations between the FE simulation and the real forming process. The data of the real forming process is obtained in systematic endurance tests and thus takes into account many other process parameters and disturbance effects that are only considered to a limited extent in the simulation. The FE simulation thus first learns known and modellable relationships by machine and then expands these with experimental process data. The surrogate and delta modelling ultimately forms the starting point for an active surface generator, which can generate an improved modelling of the active surfaces on the basis of the spring-back component geometries, e.g. using a gradient method, which already takes into account the spring-back to be expected in reality.

Figure 1: Problem definition and working hypothesis.

Working Program

Work Package	Description
WP1	Component definition and generation of associated active surface geometries
WP2	Generation of synthetic springback geometries and process data for training the surrogate model
WP3	Conception of the surrogate and delta model
WP4	Development of the endurance test process
WP5	Carrying out the endurance tests to calculate the delta model
WP6	Evaluation of the delta modelling and explanation of the analysis results

Expected Results

The expected result of the research project is the development of a novel data-driven method for optimising the springback problem of a two-stage production process (deep drawing & trimming). The basic idea of this method is that machine learning processes can be taught in such a way that improved springback compensation is possible. Both numerical and experimental data are used for the machine learning process.