In Situ Self-Calibration in a Sensor Network
Self-calibration and related methods are promising approaches in the context of sensor networks, as they open possibilities to achieve and maintain metrological traceability in situ in otherwise cost-ineffective scenarios (e.g., low-cost sensors, hard-to-reach sensor location). Therefore, such methods have been investigated within the FunSNM project to provide guidelines and practical considerations for in situ self-calibration (or co-calibration). Within the project, the need and current state of in situ calibration in sensor networks are discussed.
Methodology
To formulate guidelines and practical considerations for the application of such methods in real sensor networks, three prototypical sensor network scenarios are proposed. For each scenario, multiple promising and existing co-calibration methods are presented. The advantages of these methods regarding metrologically sound results are briefly discussed for each method and reveal a lack of uncertainty evaluation in many methods. Moreover, general remarks are provided that enhance the data quality of suitable datasets and prepare the automation of in situ calibration methods.
Results and Discussion
Finally, the applicability in real-world use cases were discussed for three generic scenarios corresponding to common sensor network configurations:
- Dense networks with stationary sensors, e.g., consensus-based, on a node-to-node basis.
Sparse networks with stationary sensors, e.g., using Gaussian process interpolation.
Sparse networks with mobile sensors and stationary reference nodes, e.g., following a rendezvous-based approach.
The latter of the three scenarios is further discussed for the specific case of air-quality monitoring networks. It was shown that the methods for co-calibration and in situ calibration must take the mobility of individual sensor nodes into account. The use of low-cost sensors in such networks further increases the need to develop methods for uncertainty-aware sensor fusion, drift detection, dynamic uncertainty estimation, and optimized traceability paths.