Tiny Machine Learning

Tiny Machine Learning is wireless, inexpensive, secure and simple. So you can immediately start "sticking sensors" on assets. This allows you to perform predictive maintenance that saves costs and increases efficiency.

SMEs are often made aware of the potential value of data. More measurement and analysis leads to more insight and more grip on the production process, the state of the installation, energy consumption and product quality. Yet the threshold for many SMEs to fully step into this is high. Often, sensor systems, connectivity and cloud resources are seen as pricey and complex, and many companies believe they lack the necessary basic knowledge.

With Tiny Machine Learning (Tiny ML), none of this applies. This is because you scale down your machine learning so that it can be run on tiny edge microcontrollers (tiny chips) powered by a battery.

Completed use cases
By September 2023, the following use cases have been completed:

- Tata Steel - noise localization with the aim of preventing noise pollution
- Hilton Foods Holland - predictive maintenance of a machine, with the aim of preventing failure or downtime and thus food waste.
- Cookies International - monitoring expiration date, barcode and label to verify the correct combination and date. With the goal of preventing food waste.
- VanDerEng - energy monitoring with the aim of energy reduction and preventing peak loads and extra sensors placed on machines to collect data for accurate pre-calculation. Companies are very satisfied with the outcomes of the use cases.

Research
In addition to the technical use cases, students at Biscuits International and Hilton Foods Holland researched the longer-term possibilities of Smart Industry, including its business implications and possible business case. The research has led to a road map of possible use cases spread over a number of years, with indications of the technical and economic feasibility and scaling-up possibilities within the organization.
The outcomes of the above use cases will therefore be further developed during follow-up use cases.

New use cases
In addition to the continuation of the existing use cases, new use cases are/will be started:

- Vezet - predicting backflows in the factory through rinse water and energy monitoring with the goal of reducing energy consumption. Strohm - predictive maintenance, preventing machine downtimeBiscuits International - energy monitoring with the aim of reducing energy consumption
- Hilton Foods Holland - monitoring entire production line with the aim of energy reduction and predictive maintenance
- IJssel - predictive maintenance, predicting slip on rolls to prevent scratches on metal at Tata Steel. This to prevent rejection and then re-manufacturing of material.
- KimPlusDelta - developing a sustainable OEE model and automating, testing and implementing it in one of the use cases with the aim of being able to estimate the reduction of environmental impact such as energy reduction, CO2 emissions and other harmful substances prior to use cases. This allows the degree of impact to be understood).

If you would like to start your own use case or have questions about Tiny ML and its application possibilities please contact Andre Gerver at a.gerver@techport.nl.