ML-analysis on miniature sensors

A lot of new exciting sensor applications can be enabled by distributed processing. Moving the analysis and decision closer to the point where a parameter is measured may in some situations improve the energy efficiency. Data processing, including machine learning, can be done in the CUPs inside the sensor itself. This may minimize the energy used for communication without impacting the response time of the system as a whole, and can open up existing new applications.

Constraints on size and battery capacity of sensors delivered by Disruptive Technologies makes this an interesting problem, with many possible solutions and tradeoffs. The current size of the sensor is 19x19x2.5mm with up to 15 years of battery life.

The assignment may involve:

• Literature study of state of the art techniques machine learning techniques like TinyML for processing sensor measurements on very constrained devices
• Gather requirements for a specific use case for on sensor processing and decisions making
• Build machine learning model for the selected use case
• Implement on sensor, possibly using development boards for the selected
• Evaluate the performance of the selected solution and compare with other potential solutions including performing the calculations in the cloud or on gateways
• Advanced aspects such as differential over the air model updates, collaborative learning where sensors send aggregated model updates to the cloud are also possible

Possible sensor types used in the assignment are vibration, temperature, moisture or electrical current, but other sensor types may be possible depending on the selected use case.

The candidate will be given access to Disruptive Technologies’ state of the art sensor hardware and ultra low power sensor communication infrastructure as needed.

The assignment is well suited for continuation into a Master Project.

Feel free to make direct contact with the Disruptive Technologies external supervisor under it this sounds interesting.

Supervisor: Sigve Tjora, Disruptive Technologies, sigve@disruptive-technologies.com