How can audio-based anomaly detection be used to optimise maintenance work in commercial properties?
Commercial properties contain numerous intricate systems for heating, cooling and ventilation. The maintenance of these systems requires extensive resources, and is often conducted through a combination of planned, periodical inspections and sporadic dispatches.
There are two notable drawbacks to this approach. On the one hand, it entails a significant amount of time being wasted inspecting well-functioning systems. On the other hand, by the time maintenance is required, undesirable consequences for the tenant have often already occurred.
Maintenance of commercial buildings
Our objective is to optimise maintenance work of commerical property systems by developing a model for automatic anomaly detection based on real-time-data from microphones and vibration sensors in mechanical rooms.
Malfunctions often manifest as audio anomalies and vibrations in pumps, motors and other machinery. The model is designed to detect significant anomalies in standard operational patterns, while simultaneously disregarding irrelevant sounds, such as transient noise and seasonal variations.
Furthermore, the model will provide enough information for the janitor or other technical personel to identify the type of problem and assess its urgency. The tool will not require detailed information about the monitored equipment or the positioning of the sensors. As a result, the model should be applicable in various technical settings.
Short distance between development and practical application
EarOnEdge is a collaborative project involving NR, SoundSensing, SINTEF Digital, and the project’s client partner, Malling & Co. Users gain access to notifications as models are developed and quality assured. The close proximity between research, development and practial application is a significant advantage, making it easier to target areas that are crucial for our end users.
Project: EarOnEdge: On-Edge Anomaly Detection in Machinery Using Sound as a Data Source
Partners: SoundSensing, Malling & Co. and SINTEF Digital
Funding: The Research Council of Norway