Derivation of cause-effect relationships for effective surface design on the basis of data-driven process modelling for fineblanking

Project Overview

Principle Investigators	Prof. Dr.-Ing. Thomas Bergs Prof. Dr. Peer Kröger Prof. Dr. Sebastian Trimpe
Project Team:	Daria Gelbich, M. Sc. / Frank Schweinshaupt, M. Sc. Andreas Lohrer, M. Sc. / Dr. Daniyal Kazempour Antonia Holzapfel San Martin, M. Sc.
Research Institutions:	Manufacturing Technology Institute MTI, RWTH Aachen University Institute for Computer Science, Information Systems and Data Mining (ISDM), Christian-Albrechts-Universität zu Kiel Institute for Data Science in Mechanical Engineering (DSME), RWTH Aachen University
Semi-finished Material(s):	Sheet metal strip 58CrV4
Manufacturing Processes:	Fine blanking

Motivation

Design of active surfaces in fineblanking to optimise the quality of cut parts by linking process data with empirical and physical knowledge

Utilisation of time- and location-dependent process data for predictive evaluation of cut part quality using AI-supported models

Development of methods that can explain and interpret AI models for fineblanking processes in order to improve reliability and adaptability

Aims

The aim of the project is the development of a data-to-knowledge-to-tool pipeline, which represents a methodical procedure with which knowledge about the interdependencies of tool surfaces and cut part quality in fineblanking can be obtained from process data and process noise that cannot be explained or predicted so far. The input for this pipeline is empirical and formalised knowledge, process signals and quality measurements of relevant target variables in the form of the resulting cutting surface quality. Modelling is carried out using data-centric AI (unsupervised learning) for pattern recognition and physics-oriented AI (supervised learning) for explainable predictions. The output of these models are efficient, digital representations and hypotheses about the interactions between the effective surfaces of the cutting gap and the bevel and ring serration on the cutting plate and their influence on the quality of the cut part during fineblanking.

Schematic representation of the data-to-knowledge-to-tool pipeline to be developed

Working Program

Work Package	Description
WP1	Design and implementation of varied fineblanking tests to specifically influence the cut part quality
WP2	Domain-specific data pre-processing and training of supervised learning models for quality prediction and condition monitoring
WP3	Data-centred AI - knowledge about interdependencies, influencing factors, representations between sub-processes and superordinate for the overall process
WP4	Physics-orientated AI: Derivation and verification of causal relationships of physics-related processes from explainability approaches (XAI) and causal inference

Expected Results

The result is a linking of process data with empirical and formalised physical knowledge of process engineering laws in fineblanking, so that previously unknown cause-and-effect relationships and non-linear effects can be included in the design of active surfaces in fineblanking. A combination of forming process knowledge with data-centred and physics-oriented AI approaches enables the derivation of cause-and-effect relationships.