The KoMoM project develops real-time monitoring, fault detection and predictive maintenance for power electronics systems used for electrical energy distribution. Due to the nonlinearities and parameter uncertainties in complex power electronics systems, it is very hard to derive a precise mathematical model, which would satisfyingly describe system behavior and yet be useful for model based fault detection. To overcome these constraints, we are using hybrid neural/statistical methods for classification and time series analysis on data recorded from the sensors. By combining knowledge of system dynamics and parameter values with knowledge extracted from recorded signal analysis we are developing techniques for achieving reliable online system condition monitoring. We have implemented the approach on a small-scale real-time system, an Arduino Nano 33 BLE Sense with a 32-Bit-ARM-Cortex-M4-CPU @ 64 MHz (nRF52840 von Nordic Semiconductors) and 1MB memory.

Partners:
Isabellenhütte Heusler GmbH & Co. KG
Avasition GmbH
Ruhr Universität Bochum & Energiesystemtechnick und Leistungsmechatronik
Ruhr Universität Bochum & Kognitive Signalverarbeitung
Siemens AG

Project start date: October 2017
PhD student working on this project: M.Sc. Nikola Markovic