A Bayesian framework for the automated online assessment of sensor data quality

Area of Science:

• Sensor technology
• Data science
• Environmental monitoring

Background:

• Real-time sensor data quality assessment is crucial for reliable applications.
• Existing methods may not adequately capture the uncertainty in sequentially correlated sensor readings.
• Automated quality control is essential for operational efficiency and data integrity.

Purpose of the Study:

• To propose a novel Dynamic Bayesian Network (DBN) framework for probabilistic sensor quality assessment.
• To represent and quantify the uncertainty associated with sensor errors in real-time.
• To develop a flexible framework applicable to various sensor deployments and phenomena.

Main Methods:

• Implementation of a Dynamic Bayesian Network (DBN) to model causal relationships between sensor errors and quality states.
• Probabilistic assessment of sensor data quality using the DBN framework.
• Comparison of DBN performance against a fuzzy logic approach using real-world marine sensor data.

Main Results:

• The DBN framework successfully generated probabilistic quality assessments and uncertainty estimates for sensor readings.
• A 34% average improvement was observed in replicating expert-generated error bars compared to fuzzy logic.
• The DBN approach demonstrated robustness without constraints on physical deployment or measured phenomena.

Conclusions:

• Dynamic Bayesian Networks offer a powerful and flexible tool for automated, real-time sensor quality assessment.
• The proposed DBN framework significantly enhances the accuracy of uncertainty estimation in sensor data.
• This method provides a superior alternative to existing approaches like fuzzy logic for critical sensor applications.