Data-based tool processing in sheet metal forming

Project Overview

Principle Investigators:	Prof. Dr.-Ing. Steffen Ihlenfeldt Prof. Dr. Oliver Niggemann
Project Team:	Adili Yiming M.Sc. Michael Hohmann M.Sc.
Research Institutions:	Institute of Mechatronic Mechanical Engineering, Chair of Machine Tool Development and Adaptive Controls, Dresden University of Technology, Institute of Automation Technology, Chair of Computer Science in Mechanical Engineering, Helmut Schmidt University
Semi-finished Material(s):	Sheet metal (DC04)
Manufacturing Processes:	Deep drawing

Motivation

Manual, time-consuming and physically demanding tool preparation accounts for 30% of tool development costs in sheet metal forming.
Due to the diverse interactions between sheet metal, tool and machine, previous scientific endeavours have not led to automation of the tool machining process.
Modern machine learning methods make it possible to process large volumes of heterogeneous data and abstract information by recognising complex patterns and relationships.
Neural networks such as Generative Adversarial Networks (GANs) or Variational Autoencoders (VAE) can generate 3-dimensional structures.
The application of machine learning to tool development promises an automatable diagnosis and prognosis, a better understanding of complex relationships, a reduction in the actual training time and a more targeted adaptation of the active surface of future tool generations as early as the design stage.

Aims

The aim of the project is the integration of machine learning methods for the automation and optimisation of tool incorporation in sheet metal forming. Generative neural networks are to be used to model the complex interactions between workpiece, tool and machine and to calculate the material removal during tool incorporation on the basis of simulation, control and optical data. A further aim is to predict tool active surfaces and optimised machine parameters for future tools.

Identification of optimal values through iterative optimization on a tryout press

Working Program

Work Package	Description
WP1	Development of an experimental platform for data generation and testing
WP2	Provision and generation of numerical and experimental data
WP3	Formalisation of deep-drawn components, tools, presses and sheet metal blanks, sensor data and the integration of prior knowledge
WP4	Implementation and initial validation of machine learning methods

Expected Results

A robot-supported experimental platform for the automated generation of data for the training of AI-based incorporation methods in sheet metal forming is being developed. As a result, suitable neural network topologies and trained models for the interpretation of the required material removal based on conventional spotting patterns using the example of a deep-drawing tool are expected. The implementation of the networks in an overall method enables automated path planning for material removal on the mould using milling robots. A second AI-based method enables the prediction of the machined tool active surfaces and the underlying tool structure, taking into account the tool-machine interaction and information from previous machining processes for tools that have not yet been manufactured.