| Summary: | <p> Artificial pancreas (AP) systems are implemented as a treatment for type 1 diabetes (T1D) patients to regulate blood glucose concentration (BGC). With continuous glucose monitoring (CGM), information related to BGC can be measured at a high frequency. It is widely known that besides meals, BGC is also influenced by many other factors such as exercise, sleep, and stress. In order to get information about these factors, different kinds of measurements such as heart rate, acceleration and derived variables such as energy expenditure (EE) should also be collected using equipment like armband and chest band devices to be used as inputs for AP systems. With adequate information about patients, BGC, and other related factors, the controller in AP systems is able to calculate insulin infusion rate for patients based on the model and control algorithm. The insulin pump will deliver the calculated amount of insulin to patient's body to close the loop of BGC regulating. </p><p> For AP systems, the performance of model-based control systems depends on the accuracy of the models and may be affected when large dynamic changes in the human body occur or when the equipment performance varies. And those factors may move the operating conditions away from those used in developing the models and designing the control system. Sensor errors such as signal bias and missing data can mislead or stop the calculation of insulin infusion rate. All of these possible performance failures can make AP systems unreliable and endanger the safety of patients. </p><p> This project aims to develop additional modules focused on fault detection and diagnosis of the controller and the sensors of the AP system. A controller performance assessment module (CPAM) is developed to generate several indexes to monitor different aspects of controller performance and retune the controller parameters according to different types of controller performance deterioration. A sensor error detection and reconciliation module (SED&RM) is developed to detect sensor error in CGM measurements. The SED&RM is based on two model estimation technologies, outlier-robust Kalman filter (ORKF) and locally weighted partial least squares (LW-PLS) to replace the erroneous sensor signal with the model estimated value. A novel method, the nominal angle analysis (NAA) is introduced to solve problems of false positive and candidate selection for signal reconciliation. SED&RM is extended to multi-sensor error detection and reconciliation module (MSED&RM), which also includes error detection and reconciliation for other sensor signals such as galvanic skin response (GSR) and values derived from original sensor signals such as EE. A multi-level supervision and controller modification (ML-SCM) module integrates CPAM and MSED&RM together and extends the controller modification into different time scales including sample level, period level, and day level. </p><p> CPAM is tested with a single input and single output (SISO) version of AP system in UVa/Padova simulator. The results indicate that a generalized predictive control (GPC) with the proposed CPAM has a safer range of glucose concentration variation and more reasonable insulin suggestions than a GPC without controller retuning guided by the proposed CPAM. The performance of SED&RM and MSED&RM is tested with data from clinical experiments. The results indicate that the proposed system can successfully detect most of the erroneous signals and substitute them with reasonable model estimation values. The ML-SCM is tested with both simulation and clinical experiments. The results indicate that the AP system with ML-SCM module has a safer range of glucose concentration distribution and more reasonable insulin infusion rate suggestions than an AP system without the ML-SCM module.</p><p>
|
|---|
