A Multi-Sensor Error Detection and Functional Redundancy Algorithm for Dynamic Systems

A Multi-Sensor Error Detection and Functional Redundancy Algorithm for Dynamic Systems

Jianyuan Feng ⁽¹⁾, Iman Hajizadeh⁽¹⁾, Sediqeh Samadi⁽¹⁾, Mert Sevil⁽²⁾, Nicole Frantz⁽²⁾, Rachel Brandt⁽²⁾, Caterina Lazaro⁽³⁾, Zacharie Maloney⁽³⁾, Xia Yu⁽⁴⁾, Elizabeth Littlejohn⁽⁵⁾, Laurie Quinn⁽⁶⁾, Ali Cinar^(1,2)

⁽¹⁾ Department of Chemical and Biological Engineering, Illinois Institute of Technology, Chicago, IL, USA

⁽²⁾ Department of Biomedical Engineering, Illinois Institute of Technology, Chicago, IL, USA

⁽³⁾ Department of Computer Engineering, Illinois Institute of Technology, Chicago, IL, USA

⁽⁴⁾ Department of Control Theory and Control Engineering, Northeastern University, Shenyang, Liaoning China

⁽⁵⁾ Department of Pediatrics, University of Chicago, Chicago, IL, USA

     ⁽⁶⁾ College of Nursing, University of Illinois at Chicago, Chicago, IL, USA

Abstract:

Sensor errors degrade the performance of supervision and control systems. Sensor accuracy can be affected by many factors such as extreme working conditions, sensor deteriorations and interferences from other devices. Sensor error such as missing signals may disrupt system operation and signal bias may cause inaccurate manipulate variable calculation by the controller that may threaten the productivity and safety of the system. For a dynamic system, it may be difficult to distinguish between some sensor errors and real dynamic changes in the system. Many systems may have multiple sensors and the signals from those sensors may be correlated with each other.  This offers the opportunity for using the redundancy to detect sensor errors and reconcile erroneous signals.

A hybrid online multi-sensor error detection and functional redundancy (SED&FR) system is developed to monitor the performance of multiple sensors and reconcile the erroneous signal values. The SED&FR system relies on two technologies, outlier-robust Kalman filter (ORKF) and a locally-weighted partial least squares (LW-PLS) regression model. The two methods have their unique way of using data, ORKF is comparing current signal samples with signal trace indicated by previous samples and LW-PLS is comparing previous samples with the samples from a database and use the most similar samples to build a model to predict current signal values. Sensor errors such as outliers can be detected more easily by ORKF. Long-duration sensor error detection and reconciliation can be handled by LW-PLS which relies on the prediction ability based on historical data sets. SED&FR leverages the advantages of automatic measurement error elimination with ORKF and data-driven prediction with LW-PLS.

The performance of this system is illustrated with a clinical case study which collects data from a continuous glucose monitoring (CGM) system and a wearable device that provides energy expenditure (EE) and Galvanic Skin Response (GSR) information from people with type 1 diabetes. More than 10,000 signal sets are used for testing. The results indicate that the proposed system can successfully detect most of the erroneous signals and substitute them with reasonable estimated values computed by functional redundancy system.