Dokument-ID Dokumenttyp Verfasser/Autoren Herausgeber Haupttitel Auflage Verlagsort Verlag Erscheinungsjahr Seitenzahl Schriftenreihe Titel Schriftenreihe Bandzahl Fachbereich Konferenzname Quelle:Titel Quelle:Jahrgang Quelle:Heftnummer Quelle:Erste Seite Quelle:Letzte Seite ISBN ISSN URN URL DOI Sprache Abstract OPUS4-14088 Konferenzveröffentlichung Gómez Zuluaga, Mauricio Andrés; Ordikhani, Ahmad; Bauer, Christoph; Glösekötter, Peter Development and implementation of a neural network-based abnormal state prediction system for a piston pump Springer Nature Switzerland AG 2021 2021 IWANN2021 Elektrotechnik und Informatik (ETI) 978-3-030-85029-6 10.1007/978-3-030-85030-2_24 deu Piston pumps play a key role in factory automation and their availability is very critical for the smooth running of production processes. Modern installations, such as production plants and machines, are becoming increasingly complex. Therefore, the probability of a complete system failure due to a single critical component also increases. Maintenance processes with intelligent devices are therefore very important to achieve maximum economic efficiency and safety. Periodic or continuous monitoring of system components provides key information about the current physical state of the system, enabling early detection of emerging failures. Knowledge of future failures makes it possible to move from the concept of preventive maintenance to intelligent predictive maintenance. In this way, consequential damage and complete system failure can be avoided, maximizing system availability and safety. This paper reflects the development and implementation of a neural network system for abnormal state prediction of piston pumps. After a short introduction into piston pumps and their potential abnormal states, statistical and periodical analysis are presented. Then the design and implementation of suitable neural networks are discussed. Finally, a conclusion is drawn and the observed accuracies as well as potential next steps are discussed.