Hybrid online sensor error detection and functional redundancy for systems with time-varying parameters

Area of Science:

• Engineering
• Biomedical Engineering
• Control Systems

Background:

• Sensor performance is crucial for control systems but is often limited by working conditions and interference.
• Sensor errors like outliers, missing values, drifts, and noise can compromise system monitoring and control objectives.

Purpose of the Study:

• To develop a hybrid online system for detecting sensor errors and providing functional redundancy.
• To replace erroneous or missing sensor data with model-based estimates.

Main Methods:

• A hybrid system combining an outlier-robust Kalman filter (ORKF) for error elimination and locally-weighted partial least squares (LW-PLS) regression for data-driven prediction.
• Incorporation of nominal angle analysis (NAA) to differentiate signal faults from genuine process changes.

Main Results:

• The system successfully detected most erroneous signals in continuous glucose monitoring (CGM) data.
• Erroneous values were effectively substituted with reasonable estimates generated by the functional redundancy system.
• Performance was validated using over 50,000 simulated CGM sensor errors across 25 clinical experiments.

Conclusions:

• The proposed hybrid system demonstrates robust performance in online sensor error detection and functional redundancy.
• This approach enhances the reliability of sensor data in critical applications, such as diabetes management.