Image-based condition monitoring for modelling the temporal development of the wear condition of ball screws by means of machine learning
Final Report Abstract
The aim of the research project was to investigate image-based condition monitoring for modeling the temporal development of the wear condition of ball screw drives using machine learning. The research project was divided into four phases. In the first phase, extensive wear tests were carried out to create a data set that depicts the entire wear development on the surface of the ball screw drive (BSD). This data set served as the basis for further investigations. Based on this data set, powerful approaches for classifying image data in technical domains were initially selected from the state of the art in research and technology. A preselection was made from these models on the basis of preliminary tests. In a subsequent step, this preselection was supplemented by further models using a test plan-based approach and their performance was examined. The influence of data augmentation and the influence of data generated synthetically using generative adversarial neural networks were also examined. In a subsequent step, an empirical investigation of the damage development was carried out on the basis of the image data. The aim of the investigations was to describe the development of damage on the spindle. In order to ensure industrial operation, the next step was to define and formalize industrial influencing variables based on domain knowledge, which must be taken into account when the system is later used industrially in order to ensure reliable data generation during operation. In order to move from a system that can reliably identify defects in image data to a system that quantifies the wear condition and thus the service life of the BSD on the basis of the image data, this aspect was considered in a final work package. In summary, all work packages were successfully completed. Specifically, an image data set was generated that depicts the entire wear history on the BSD surface. Based on this data set, it was possible to identify models that reliably indicate defects on the BSD surface. The wear development could be formalized on the basis of the image data and thus theses that were expressed in previous work in the state of research could be empirically proven. Furthermore, influencing factors for industrial image acquisition were successfully identified. Finally, a method was investigated that is capable of quantifying the extent of damage. The project can therefore be described as successfully completed. In follow-up studies, the aspect of data efficiency in particular should be addressed, as data-efficient and generalizing methods are essential for the successful and scalable integration of machine learning methods in production.
Publications
-
Extraction of surface image features for wear detection on ball screw drive spindles. Tagungsband "Forum Bildverarbeitung 2020, 305-316. KIT Scientific Publishing.
Schlagenhauf, Tobias; Heinzler, Max & Fleischer, Jürgen
-
Frühzeitige Detektion von Oberflächenzerrüttungen/Premature detection of surface disruption – Deep Learning-based method for classification of damage on ball screw drives. wt Werkstattstechnik online, 110(07-08), 501-506.
Ruppelt, Peter; Schlagenhauf, Tobias & Fleischer, Jürgen
-
A stitching algorithm for automated surface inspection of rotationally symmetric components. CIRP Journal of Manufacturing Science and Technology, 35, 169-177.
Schlagenhauf, Tobias; Brander, Tim & Fleischer, Jürgen
-
Industrial machine tool component surface defect dataset. Data in Brief, 39, 107643.
Schlagenhauf, Tobias & Landwehr, Magnus
-
Intelligent vision based wear forecasting on surfaces of machine tool elements. SN Applied Sciences, 3(12).
Schlagenhauf, Tobias & Burghardt, Niklas
-
Online Learning für die präventive Verschleißdetektion/Online Learning for preventive wear detection – Online Retraining of Deep Learning models for unknown wear patterns. wt Werkstattstechnik online, 111(07-08), 475-480.
Schlagenhauf, Tobias; Ammann, Nicholas & Fleischer, Jürgen
-
Bildbasierte Quantifizierung und Prognose des Verschleißes an Kugelgewindetriebspindeln, Dissertation, Karlsruher Institut für Technologie, Fakultät für Maschinenbau, Karlsruhe
Schlagenhauf, T.
-
Analysis of the visually detectable wear progress on ball screws. CIRP Journal of Manufacturing Science and Technology, 40, 1-9.
Schlagenhauf, Tobias; Scheurenbrand, Tim; Hofmann, Dennis & Krasnikow, Oleg
