Online capability is key to uncertainty propagation
Automation in the sense of the automated application of fundamental methods to IoT systems is a key building block of metrology for sensor networks. In a study conducted as part of the FunSNM project, methods for in situ calibration of sensors at their location of deployment as well as co-calibration using reference sensors already present in the network were found to be key focus areas in this regard.
Furthermore, machine learning methods as well as methods for data aggregation are necessary given the large amounts of data generated and processed in a sensor network. Methods for online and federated machine learning using distributed systems will play a significant role. The necessary tools to achieve this end are those for automated uncertainty propagation, for agent-based simulations and a middleware framework as the core communication and data-management layer.
The key requirements of the uncertainty propagation software are online capability and the ability to evaluate uncertainties for time-varying quantities. Furthermore, scalability and interoperability are important to both the middleware framework and the agent-based simulation software. The ability to incorporate mobile sensor nodes will be of particular importance to the agent-based simulations.
The following software tools were identified as being of particular relevance: PyDynamic for dynamic uncertainty evaluation, the agentMET4FOF package for agent-based simulations and the FIWARE middleware framework. The results of this study were presented at the IMEKO World Congress on the 27th of August 2024.