An Insulin Infusion Fuzzy Controller with State Estimation for Artificial Pancreas Systems

Abstract

A fuzzy controller for a minimal states model is proposed to achieve a continuous and effcient insulin infusion in patients with Type 1 Diabetes. An Extended Kalman Filter is also applied to supply the deficiencies of the current glucose sensor technologies and estimate residual insulin in the system to predict future behavior. The controller is tuned manually and iteratively, and achieves closed-loop responses of glycemia constrained between [80,140] (View the MathML source), with a mean of 117, 6 (View the MathML source) and a standard deviation of 11, 3 (View the MathML source) over a whole year ensemble of 24-hour system responses with 4 meal intakes per day. These results show that is possible to design low complexity controllers that are easily tunable by experienced users or physicians focusing on a closed-loop system response analysis. Also, the combination of heuristic and model-based techniques allows to synthesize robust controllers for real application situations and, furthermore, effciently manage the insulin usage. Nevertheless, the actual application of a closed-loop system on a human being should require higher dimension models to fit different situations, a proven robust controller and wide adaptability to different patients and their meal intake routine.