Fraunhofer Institute for Electronic Microsystems and Solid State Technologies EMFTThe aim of predictive maintenance is to maintain equipment proactively and with foresight. Thus, downtimes and maintenance effort are to be reduced to a minimum. Researchers at Fraunhofer EMFT are testing new concepts for using machine learning methods to efficiently process even extremely heterogeneous data and make exact maintenance predictions for production equipment.
Project KIWA: AI-based predictive maintenance for manufacturing equipment
Basic technologies for predictive maintenance are networked sensor nodes and a central data repository, so the topic is closely linked to the technology trends Internet of Things (IIoT), Industry 4.0. A model trained by machine learning with the centrally stored data can later also be used at the decentralized nodes for data evaluation.
In the KIWA project, a Fraunhofer EMFT research team combines sensor data from a manufacturing plant with other, external sensors to generate additional relevant characteristic values such as the vibration of a drive. Critical plant components are then monitored during operation and correlated to the process flow. An application example are plants of the project partner Mühlbauer: There, additional, external sensors such as vibration sensors or sensor systems for the exact measurement of further parameters are installed for an exact recording of the time sequences in order to predict maintenance or the replacement of drive systems and plant components of the production plant. The data collected is to be used to draw conclusions about the current operating condition and to make predictions about the further service life or necessary maintenance requirements.
Development of a machine learning-based solution for predictive equipment maintenance in high-throughput RFID production.
While classical questions of condition monitoring and predictive maintenance usually operate on unimodal data, often even on one-dimensional data of constant clock frequency, the challenge in this project lies in the selection and processing of extremely heterogeneous data from a wide variety of sources with clock rates ranging from a few hertz to about 50 kHz using machine learning methods. The currently prevailing preventive maintenance could be replaced by predictive maintenance management in the future as a result of the expected results.
In addition to Fraunhofer EMFT and the Mühlbauer Group, Procon IT and Munich University of Applied Sciences are also involved in the project. The project is funded by the Bavarian State Ministry for Economic Affairs, Regional Development and Energy under the grant number DIE0147.