Karriere am WZL


Studienarbeit oder Masterarbeit oder Bachelorarbeit oder Projektarbeit

am Forschungsbereich Technologie der Fertigungsverfahren, Abteilung Umformen, Research Group Forming Technologies

Neural Networks and Machine Learning in Production Engineering

Fineblanking is a highly economical sheet metal cutting process for the mass production of components with very shearing surface quality. Especially in the automotive and aerospace industries, fine blanked components are widely used.

Within the framework of the Internet of Production, which was approved as one of the two clusters of excellence of the RWTH, the potentials of a fine blanking press as a cyber-physical system are being researched. Various sensor systems were installed for the online recording of process forces, peripheral forces and acoustic emission signals, as well as material properties and environmental influences on the fine blanking process. The objective of this data acquisition is to detect machine faults at an early stage and to avoid highly expensive downtimes. In addition to that, deviations during the forming process are to be detected in order to react to them.

To achieve this goal, a reference state of the material must be identified by means of non-destructive material testing. This allows different material behavior and qualities of the material from the reference state to be quantified and correlated with effects on the process (e.g. forces) or quality characteristics of the fine blanked component. For this purpose, data-driven models, which identify the reference condition of the material and can robustly quantify deviations in real time, must first be designed.

The issues to be addressed in this context include various areas of data and signal processing. The material sensor provides a periodic signal with a very high sampling rate, so basic methods for processing digital signals are essential part of this work. Furthermore, the disciplines Data Science and especially Machine Learning (both 'Supervised Learning' and 'Unsupervised Learning') are important tools for signal processing and reference signal derivation. The scope and type of methods used in data-driven modeling can be individually adjusted depending on your personal preferences and the type of thesis.
• Motivation
• Independent working method
• Interest in data analytical questions and machine learning in an industrial context
• Basic programming skills, e.g. in Python

We offer:
• Comprehensive Support
• Flexible working times
• Introduction to data science and machine learning in an industrial context
• Introduction to machine learning libraries (Python) like Keras / TensorFlow and Scikit-learn
• Introduction to working with a non-destructive material testing sensor
Zeitaufwand: 11,00 Arbeitsstunden

Martin Unterberg
Cluster Produktionstechnik 3A 328
Tel.: +49 241 80-22958
Fax: +49 241 80-22293
Mail: M.Unterberg@wzl.rwth-aachen.de