Model Predictive Control of Artifical Pancreas for Type -1 Diabetes Management

Abstract

During the past few decades, bio-medical sciences have experienced growing interest from mathematicians and researchers in developing mathematical models able of mimicking the physiological characteristics of human body and its processes. Different bio-medical modeling techniques have been developed for numerous medical problems which require monitoring and control, which have significantly improved the lives of people living with those medical problems. Out of many diseases which need and deserve such attention from bio-medical researchers, Diabetes is considered to be one of the most important and critical medical problems, needing attention from the bio-medical research community. This thesis focuses on the design of an advance control technique called Linear Model Predictive Control for improving glucose regulation in type I diabetic patients along with other simple linear controllers like PID, State Feedback, State Observer, Linear Quadratic Regulator for linearized versions of the nonlinear 3 state, 4 state and 11 state glucose-insulin dynamic model of diabetic patients. All of the nonlinear models are linearized before applying linear control techniques. Nonlinear control techniques like Feedback linearization and Backstepping Control to the 3 state and 4 state nonlinear glucose-insulin models. All linear and nonlinear control techniques applied on 3, 4 and 11 state glucose-insulin models are simulated in MATLAB and Simulink environment and their performance is evaluated after applying different linear controllers. The novelty of the research work is the design of the Linear Model Predictive Control for the linearized version of the 4 state glucose-insulin diabetic model. Its results are compared with the other simpler linear controllers and the response time, settling time, constraints satisfaction shows significant improvement. The rest of both linear and nonlinear controllers violates the constraints causing episodes of both hyperglycemia and hypoglycemia which is the crust of the research work being done while addressing the problem of diabetes management. Hypoglycemic attacks are more concerning than that of hyperglycemia. They can be fatal and should be addressed right on time and this novel control approach achieve this goal. The extension of the research work will be implementation of both Linear and nonlinear MPC on the higher order models and some nonlinear control application to the artificial pancreas problem.