Multivariable adaptive identification, fault detection, diagnosis and control of artificial pancreas systems

An artificial pancreas control system automates insulin pumps by using a closed-loop controller that receives information from sensors, computes the optimal insulin amount to be infused and manipulates the infusion rate of the pump for continuous blood glucose regulation in patients with type 1 diabetes. An integrated multivariable adaptive artificial pancreas control system is developed. Multivariable recursive time-series models are developed by using additional physiological measurements from a sport armband. The multivariable models are obtained with a proposed constrained weighted recursive identification algorithm to guarantee the stability conditions and satisfy the physiological properties of human body and glucose-insulin dynamics. Hypoglycemia early alarm system is developed based on the multivariable time-series models. By use of the physiological measurements, different thresholds are defined for different conditions such as meal, exercise and sleep for prevention of hypoglycemia. Generalized predictive control based adaptive control algorithm is proposed for blood glucose regulation in patients with type 1 diabetes. The control algorithm is completely adaptive and does not require any manual announcements. A meal detection algorithm is implemented into the control algorithm. Meals are detected based on the estimation of rate of appearance of glucose by use of Unscented Kalman filter. A novel framework is developed for meal bolus when a meal is detected. In addition to all, a fault detection and diagnosis algorithm is also developed. Multiway principal component analysis is used for detection of system failures. All proposed algorithms are tested with both simulation and clinical experiments. The result indicates that the proposed integrated artificial pancreas system provide significant improvements. The prosed system is able to deal with blood glucose regulation problem under various challenging conditions. Being fully automated and adaptive, it provides many advantages to patients with type 1 diabetes.