Real-Time Predictive Maintenance – Artificial Neural Network Based Diagnosis

David Kaufmann, Florian Steffen Klück, Franz Wotawa, Iulia-Dana Nica, Hermann Felbinger, Adil Mukhtar, Petr Blaha, Matus Kozovsky, Zdenek Havranek, Martin Dosedel

Publikation: Beitrag in Buch/Bericht/KonferenzbandBeitrag in Buch/BerichtBegutachtung


In this article, we discuss the use of artificial neural networks for monitoring and diagnosis to be used in the context of real-time predictive maintenance. There are two use cases analysed here. As a first one, we discuss the motor model used for diagnosis in detail. In particular, we introduce a detailed acausal six-phase e-motor model to be used for different stator and inverter faults simulations. The inter-turn short circuit fault is targeted here. Simulation data and data measured on a real custom-made six-phase motor with the ability to emulate this fault are pre-processed based on the mathematical analysis of the fault. Such data are then used for modular neural network training. The trained modular neural network is optimized and deployed into the NVIDIA Jetson platform. The second ANN presented in this article is designed for bearing fault detection based on vibration measurements. The vibration data taken from publicly available datasets are transformed into suitable condition indicators which are analysed by the multilayer perceptron network running on a PC in MATLAB with the possibility to implement the resulting network into a small edge device. As such, two use cases are shown how artificial neural networks can be used on edge devices. Obtained results show that the approaches can be used in real setups.
TitelArtificial Intelligence for Digitising Industry
Redakteure/-innenOvidiu Vermessen, Reiner John, Cristina De Luca, Marcello Coppola
Herausgeber (Verlag)River Publishers
ISBN (elektronisch)9788770226639
ISBN (Print)9788770226646
PublikationsstatusVeröffentlicht - Sep. 2021


Untersuchen Sie die Forschungsthemen von „Real-Time Predictive Maintenance – Artificial Neural Network Based Diagnosis“. Zusammen bilden sie einen einzigartigen Fingerprint.

Dieses zitieren