h1

% IMPORTANT: The following is UTF-8 encoded.  This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.

@PHDTHESIS{Kpper:1002816,
      author       = {Küpper, Ugur},
      othercontributors = {Bergs, Thomas and Klocke, Fritz},
      title        = {{D}ata-driven model for process evaluation in wire {EDM};
                      1. {A}uflage},
      volume       = {2025,6},
      school       = {RWTH Aachen University},
      type         = {Dissertation},
      address      = {Aachen},
      publisher    = {Apprimus Verlag},
      reportid     = {RWTH-2025-00699},
      isbn         = {978-3-98555-260-3},
      series       = {Ergebnisse aus der Produktionstechnik},
      pages        = {1 Online-Ressource : Illustrationen},
      year         = {2025},
      note         = {Druckausgabe: 2025. - Auch veröffentlicht auf dem
                      Publikationsserver der RWTH Aachen University. - Weitere
                      Reihe: Technologie der Fertigungsverfahren.- Weitere Reihe:
                      Edition Wissenschaft Apprimus; Dissertation, RWTH Aachen
                      University, 2024},
      abstract     = {The main areas of application for wire EDM are in tool and
                      die making, as well as in engine and medical technology. It
                      is mainly used in the production of high-priced products and
                      is often the last decisive manufacturing technology. The
                      process reliability and repeatability of this technology are
                      therefore particularly important and can be guaranteed by
                      correspondingly intelligent control and automation
                      solutions. This, along with the digitalization of
                      manufacturing processes in the context of Industry 4.0,
                      requires the use of data-driven approaches in wire EDM. The
                      objective of the present work was therefore to develop a
                      data-driven model for the evaluation of the wire EDM
                      process. This was to be based primarily on continuously
                      recorded physical respectively electrical process data in
                      order to ensure the transferability and general validity of
                      the model. Machine learning models were trained with process
                      data to evaluate the process solely based on the electrical
                      process signals evaluated in real-time. This goal was to be
                      achieved by developing a regression model to evaluate
                      quality and a classification model to evaluate productivity.
                      The scientific design framework in this work is determined
                      in particular by the methods and techniques used in data
                      analysis and the structure of the work is designed
                      accordingly for the development of data-driven models. To
                      this end, a system was first developed for the real-time
                      recording of time and spatially resolved characterized
                      individual discharges in the continuous process. After data
                      processing, including systematic data reduction and feature
                      extraction, characteristic values were correlated with
                      process productivity and product quality in the following
                      step. Based on the initial findings from the process data, a
                      regression model was developed to evaluate product quality.
                      For this purpose, a neural network was trained that predicts
                      the curvature of the component based on continuously
                      recorded data. The model shows good prediction accuracy and
                      explains a significant part of the data variability. The
                      productivity of the process was evaluated using a
                      classification model. A deep learning approach was used, in
                      which various forms of neural networks were used for the
                      model architecture. The results showed high accuracy,
                      especially considering that all tests were performed with
                      completely unknown data. Finally, it was shown how the
                      findings can be transferred to the development of a Digital
                      Twin in an industrial setting. In cooperation with an AI
                      software manufacturer and a wire EDM user, a Digital Twin
                      was developed that can map the generated curvature of the
                      workpiece in a dashboard using data processed in real-time.},
      cin          = {417410 / 053200 / 417400},
      ddc          = {620},
      cid          = {$I:(DE-82)417410_20140620$ / $I:(DE-82)053200_20140620$ /
                      $I:(DE-82)417400_20240301$},
      pnm          = {BMBF 01IS20096 - KI-Erosion: Einsatz von Methoden der
                      Künstlichen Intelligenz zur datenbasierten Bewertung des
                      Drahtfunkenerosionsprozesses (01IS20096)},
      pid          = {G:(BMBF)01IS20096},
      typ          = {PUB:(DE-HGF)11 / PUB:(DE-HGF)3},
      doi          = {10.18154/RWTH-2025-00699},
      url          = {https://publications.rwth-aachen.de/record/1002816},
}