Project duration from 2017-10-01 to 2020-03-31
Funded by: Land OÖ

Industrial fans and filter systems, in particular radial fans, are essential components of dedusting and exhaust gas cleaning plants. In the course of increasing performance and features, these fans are facing exceptional challenges concerning their reliability. Damages to the bearing, engine or impeller are likely to happen at some point that might cause an outage of the entire plant.

While preventive maintenance and fixed service intervals can lead to an acceptable level of reliability, unexpected damages and downtimes remain potential issues, as well as the unnecessary replacement of components, which have not reached the end of their lifetime.

Today different kinds of sensors for rolling bearings and engines are available that record temperature and vibrations and monitor the adherence to specified thresholds. However, there is no system monitoring radial fans in their entirety, including all relevant parameters, such as volume flows and pressure ratios. Moreover, no state of the art sensor-based system is capable of measuring deposit build-ups on high-speed impellers yet.

Within this concrete project, the research focuses on investigating proactive maintenance strategies for radial fans. Therefore, the fans will be equipped with additional sensors that record vibration, acceleration, air pressure, temperature, flux, power consumption, rotation, abrasive wear, and deposit building, in order to monitor the system at any given moment. This installation enables Predictive and Condition-Based Maintenance, which considers the usage history, the current condition as well as the prospective operational load.

Data recorded in an experimental setup, combined with empirical knowledge is analyzed with machine learning methods in order to provide forecasts regarding the prospective system’s behavior and potential breakdowns. This allows the early detection of maintenance needs and hence prevents total outages. In this course, novel algorithms for anomaly detection will be developed that trigger maintenance actions before a system starts degenerating.

All lessons learned during this research project regarding Predictive Maintenance, the suitable sensor technology, or Machine Learning, can be mapped to other application domains. Especially the combination of sensors for monitoring industrial plants, data based modeling for system identification and novel algorithms for anomaly detection will be a valuable contribution for further joining industry and information technology.