smart sensors predict freezing in real time
The blender processes meat with additives. Nitrogen
cools the mixture, this should always remain between -2.4 °C and -1.9 °C, unfortunately sometimes freezing of the mixture takes place, making it hard and damaging the machine. To prevent this, student Kasper Tiebe developed a proof-of-concept predictive maintenance solution using TinyML.
Using sensors that record vibration, temperature and current, real-time data was collected via a low-cost microcontroller. The system detects anomalies based on unsupervised machine learning (for those interested: a special type of neural network, Variational Autoencoder). Thus, team members in the factory know when the mixture is in danger of freezing, even before damage to the mixing arms occurs.
This provided tinyml in the production process
Thanks to the smart sensor setup and algorithms, the system can detect anomalies based on normal processes, without historical error data. This allows the system to provide timely warning and intervention. The model runs locally on an energy-efficient chip, making cloud costs and latency low.
The impact? Less damage to parts, less downtime, less production loss, less food waste and higher predictability and safety in the process.
more info or register?
Want to know how to deploy smart technology for your machines? Contact the Fieldlab Edge AI for Smart Industry at Andre Gerver.
CONTACT
Andre Gerver
📧 a.gerver@techport.nl