Structural Analysis for the Traceability of Formed Parts

Identify and Trace Components

Supplied on huge rolls, steel sheets are cut to the required size in the press shop. They then pass through several presses to obtain the desired shape. Traceability of the component with its individual characteristics at every stage of processing is a prerequisite for production decisions and helps to optimize quality, productivity and costs.

In the INSITU project, we are working with two other Fraunhofer Institutes to develop tools and methods for the seamless traceability of formed parts throughout the entire production chain.

Identifying and tracing a component with its individual characteristics at any time during processing is a prerequisite for many process-influencing decisions in production. The optimization of quality, productivity and costs depends on it. This traceability is indispensable, especially in the keyword Industry 4.0 – in the self-organizing value-added networks of the future. The product to be manufactured must be clearly identifiable and localizable at all times so that the production process can be automated.

Non-Destructive Testing of Components

Previously used object markings – such as labels, barcodes, IC chips, hologram stickers or other identification tags – are only suitable to a limited extent because they do not remain permanently and undamaged on the object. If the surface changes during processing, identification is no longer possible. Until now, there has been a lack of sensory methods that allow continuous traceability even if the components have been changed during processing, e.g. by forming, machining or coating.

Identification Through Material Properties

In the INSITU project, we therefore use non-destructive testing (NDT) methods to identify the component on the basis of features from the inside of the component – such as material properties or tolerable »defects«. Such sensory methods have so far mainly been used for quality testing. Through large series of tests, we were able to determine that certain material properties provide unambiguous sensor signals, which in turn enable clear identification.

By combining several methods, this approach to component identification is optimal for different metallic materials (e.g. steel, aluminum) and different component types (sheet metal, castings, etc.). The identification data obtained by sensors is stored in a feature space and represents the component as a unique individual.

Idealisiertes Presswerk — © Fraunhofer IWU
Forscher am Fraunhofer-Institut für Werkzeugmaschinen und Umformtechnik IWU in Chemnitz haben sich Industrie 4.0 im Bereich des Maschinenbaus auf die Fahnen geschrieben – und widmen sich unter anderem dem Presswerk der Zukunft. Ein solches idealisiertes Presswerk hat viele potentiell mehrwerterzeugende Positionen für Identifikationssensoren.

Machine Learning and Digital Object Memory.

We then classify this feature space using suitable machine learning methods. The identification is supported by a digital object memory. A DOMe (for Digital Object Memory) is a data memory in which all relevant information about a concrete physical object is continuously collected.

Aims of the Project

Development of an industrial sensor system for component identification in the field of sheet metal forming, including a cloud-based data management system for administration and tracking of product-related data
Extension of existing sensor systems by the additional functionality of component identification, e.g. for quality inspection of cast components
Realization of a complete component identification along the entire production chain based on a multi-sensory approach.

Our Tasks and Expertise in the Project

Machine Learning has been an integral part of many research activities at our institute for years. In numerous projects we develop mathematical models and image analysis algorithms and implement them in industrial software. We also work with hybrids of «classical» parameterizable methods and learning approaches.

Our focus at INSITU lies in the following fields:

Development of the characteristics database
Generation of stochastic structural models and the training data
Model-based machine learning

Learning methods such as Deep Learning require a large amount of reliable data, which is often not available. Therefore we develop model-based learning approaches. Assumptions about objects are modelled and this modelling is used as input for automated procedures. We also use such a »Model-Based-Machine-Learning« approach in the project.

Project Participants

Coordination: Fraunhofer Institute for Machine Tools and Forming Technology
Fraunhofer Institute for Nondestructive Testing IZFP
Fraunhofer Institute for Industrial Mathematics ITWM

Duration of the Project:

01.02.2018 – 31.01.2021