Robust automatic monitoring in industrial environments
"Oops, that doesn't sound good", is often the thought when your car or kitchen appliance makes a strange noise and you know that something has broken down. The same happens with industrial machines and vehicles when certain components are damaged or fail. In order to timely detect anomalies and quality problems in machines, we use microphones and microphone arrays in the ACMON project.
In this ICON project, Flanders Make is joining forces with Imec/UGent (IDLab) to develop robust and cost-efficient monitoring solutions for machines and vehicles in an industrial context. Bekaert, I-Care, Siemens and Stow are the industrial partners that join this project to use the developed solutions in their production and control environment.
A failing component in a moving or rotating machine will start to vibrate. These vibrations can be measured by accelerometers for early problem detection. However, fitting such a sensor is often difficult or even impossible without disassembling (part) of the machine. Moreover, each component under control must be fitted with an accelerometer, which entails the necessary costs. The same vibrations also propagate through the air in the form of sound. This sound can then be measured on the outside of the machine by a single microphone (or microphone array) without using invasive methods.
In such industrial environments, there are also many sources of acoustic disturbance such as other machines making noise, people talking, tools falling on the ground, forklift trucks, etc. To avoid false positives, it is important to measure the noise from the machine itself. To avoid false positive detection, robust methods using noise reduction methods, AI models and data augmentation techniques will be developed.
- Increasing signal-to-noise ratio of sound recordings
- Train AI models for robust detection of problems
- Easy deployment, maintenance and validation of AI models for acoustic monitoring