wbk Institut für Produktionstechnik

Tobias Schlagenhauf, M.Sc.

  • 76131 Karlsruhe
    Kaiserstraße 12

Tobias Schlagenhauf, M.Sc.

Forschungs- und Arbeitsgebiete:

  • Maschinen-, Anlagen-, Prozessautomatisierung unter Verwendung von Verfahren des maschinellen Lernens
  • Data Science im industriellen Kontext
  • Themenbereich Industrie 4.0

 

Allgemeine Aufgaben:

  • Vorlesungsbetreuer der Vorlesung Umformtechnik

 

Lebenslauf:

seit 08/2018 Wissenschaftlicher Mitarbeiter am Institut für Produktionstechnik (wbk) des Karlsruher Instituts für Technologie (KIT)

 

Veröffentlichungen

[ 1 ] Schlagenhauf, T.; Hillenbrand, J.; Klee, B. & Fleischer, J. (2019), „Integration von Machine Vision in Kugelgewindespindeln“, wt Werkstattstechnik online, S. 605-610.
Abstract
Unvorhergesehene Maschinenausfälle von Werkzeugmaschinen durch natürlichen Verschleiß sind häufig auf den Kugelgewindetrieb zurückzuführen. Für eine frühzeitige Erkennung der auftretenden Schäden, präsentiert dieser Beitrag einen Ansatz für die Überwachung von Spindeln von Kugelgewindetrieben mittels integriertem Kamera - system. Ziel ist die frühzeitige Detektion von Schäden, die auf der Spindeloberfläche erscheinen, um entsprechende Wartungsmaßnahmen abzuleiten.

[ 2 ] Schlagenhauf, T.; Feuring, C.; Hillenbrand, J. & Fleischer, J. (2019), „Camera Based Ball Screw Spindle Defect Classification System“. Production at the leading edge of technology, Hrsg. Springer, S. 503-512.
Abstract
This paper shows how to detect Pitting on a ball screw drive (BSD) with the help of a Convolutional Neural Network (CNN). Building on a previous approach where we presented an integrated camera system for the BSD by applying a camera to the nut which is able to monitor the ball screw spindle´s surface, this paper deals with the Condition Monitoring of the component by feeding spindle surface images into a CNN to identify Pitting defects. The authors develop a CNN that is able to distinguish between images showing Pitting and images without. For training purposes a balanced dataset of images with and without Pitting is used. With a number of 200 images, a four-fold Cross Validation approach is used to maximize the amount of data used for training and testing. The trained model performs with a mean accuracy of 91,50% on the new test data. Further, the model performs with a mean precision of 93,68% at a recall of 89,00%.

[ 3 ] Schlagenhauf, T.; Ruppelt, P. & Fleischer, J. (2020), „Detektion von frühzeitigen Oberflächenzerrüttungen“, wt Werkstattstechnik online, Band 7, S. 501-506. https://e-paper.vdi-fachmedien.de/webreader-v3/index.html#/2657/49 [30.11.-1].
Abstract
Condition monitoring of plants, machines and their components is a central topic of Industry 4.0. Unforeseeable failures of machine tools are often caused by wear and the resulting failure of ball screws for which surface disruptions represent an important characteristic. This article describes how image-based monitoring of ball screws by an electronical camera system combined with deep learning based models is able to detect premature surface disruptions to derive appropriate and preventive maintenance measures.

[ 4 ] Schlagenhauf, T.; Heinzler, M. & Fleischer, J. (2020), „Extraction of surface image features for wear detection on ball screw drive spindles“. Forum Bildverarbeitung 2020, Hrsg. KIT, KIT Scientific Publishing, Karlsruhe, S. 305-314.
Abstract
Failures of production machines are often caused by wear and the resulting failure of components. Therefore, condition-based monitoring of machines and their components is becoming an increasingly important factor in industry. Due to the simple conversion of the motion of electric rotary drives into precision feed motion, the ball screw is an inherent element of many production machines. Thus, a failure of the ball screw often leads to costly production stops. This paper shows the determination and extraction of wear-describing image features, allowing an image-based condition monitoring of ball screws using hyperparameteroptimized machine learning classifiers. The features to train the algorithms are derived and extracted based on the deep domain knowledge of ball screw drive failures in combination with further developed state of the art feature extraction algorithms.

[ 5 ] Kein Eintrag gefunden.
Abstract
Using Machine Learning Techniques in general and Deep Learning techniques in specific needs a certain amount of data often not available in large quantities in some technical domains. The manual inspection of Machine Tool Components, as well as the manual end of line check of products, are labour intensive tasks in industrial applications that often want to be automated by companies. To automate the classification processes and to develop reliable and robust Machine Learning based classification and wear prognostics models there is a need for real-world datasets to train and test models on.