SSV Software Systems
Embedded TinyML LPWAN Wireless Modules
To solve decentralised monitoring tasks with embedded machine learning concepts and low-power IoT wireless data transmission without having to worry about the new cybersecurity requirements of RED or EN 18031: the new Embedded TinyML LPWAN Wireless Module from SSV Software Systems makes it possible.
The wireless transmission of sensor data for monitoring the status of machines and systems to a cloud service has numerous disadvantages: On the one hand, the transmission processes require a relatively large amount of electrical energy. On the other hand, large amounts of data accumulate over time, some of which incur considerable operating costs. In addition, there are often problems with data and privacy protection, as well as the normative requirements of the Radio Equipment Directive (RED) and corresponding EN 18031 standards, including the associated additional costs.
In view of these multidimensional challenges, SSV now offers an alternative – the application-specific wireless module concept (ASWM) as a conceptually and functionally highly integrated solution, including RED compliance. Instead of streaming sensor data to the cloud, data analysis is performed directly in the sensor using TinyML methods. The results are sent via an LPWAN radio connection depending on their relevance, for example in the event of certain status changes or a detected anomaly. ASWM soft sensor assemblies use either LTE-M, LTE450 or LoRaWAN for this purpose.
Machine learning (ML) consists of two subtasks: the learning phase with ML model creation and the inference phase for ML model use, for example with sensor data. The first part requires large amounts of data and computing capacity. For inference, a sensor data source and a ‘tiny’ microcontroller with minimal resources are sufficient – this is why it is also referred to as ‘TinyML’. TinyML inference enables, for example, the processing of telemetry data directly in the microcomputer of a wireless module without the need to send the measured values of individual sensors together with other status data to external servers or a cloud. If necessary, no data relating to individuals or privacy is transmitted via LPWAN at all. Instead, only event-driven abstract status classification parameters are sent, such as a categorical variable for a detected anomaly in the status of a machine or plant. This security-by-design approach offers a significantly higher level of security and considerable cost reductions compared to products that continuously transmit raw telemetry data to a central cloud in order to perform the necessary data analyses there.
‘In a networked factory environment with high-performance LAN concepts, a central AI or ML solution with data streaming can certainly work,’ says Henrike Gerbothe, the SSV manager responsible for the ASWM product area. ‘For a decentralised infrastructure application, however, this is unsuitable for functional and economic reasons. The sensor data must be analysed directly on site and radio data transmissions reduced to the necessary minimum.’
