Optimizing Sensor Calibration in Open Environments: A Bayesian Approach for Non-Specific Multisensory Systems

This is the authors' abstract. We don't add key points for this paper.

Air and water pollution present significant threats to public health, highlighting the need for precise environmental monitoring methods. Current solutions rely on multisensory systems with limited specificity. Their calibrations often struggle in real-world conditions, resulting in imprecise air and water quality measurements. This paper aim to address the challenge of calibrating non-specific multisensory systems deployed in open periodic environments. A data-driven calibration method is proposed within a Bayesian framework, which considers several sources of uncertainties that are often overlooked in sensor calibration. The method combines a non-parametric approach, capturing correlations between pollutants and environmental variables, with a parametric method, that maximizes sensor-provided information. Unlike conventional sensor calibration, our method prioritizes the inclusion of input uncertainties and model errors during calibration, providing a comprehensive framework for robust sensor performance. The theoretical foundations of the non-parametric approach are presented, and the coupling between non-parametric and parametric methods is detailed. The evaluation using synthetic data demonstrates the method’s efficiency and limitations. Then the approach is validated in an experimental use case where a sensor array based on carbon nanotube is calibrated for monitoring ozone and carbon monoxide in an outdoor deployment.