Adaptive Blood Glucose Control Control: for Type 1 Diabetes
No Thumbnail Available
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Type 1 Diabetes Mellitus (T1DM) results from the autoimmune destruction of pancreatic
β-cells, eliminating endogenous insulin secretion and requiring permanent exogenous insulin
administration. The Artificial Pancreas System (APS) automates this therapy through a
closed-loop algorithm that continuously regulates insulin delivery from glucose sensor measurements.
Designing a reliable APS controller is challenging due to the inherent nonlinearity
of the glucose–insulin system, significant inter- and intra-patient physiological variability, and
continuous meal and activity disturbances. Existing approaches either lack formal stability
guarantees, require explicit knowledge of patient parameters, or are computationally prohibitive
for embedded implementation.
This thesis proposes the application of Simple Adaptive Control (SAC) to blood glucose
regulation in T1DM. SAC is a direct Model Reference Adaptive Control strategy that adapts
a gain matrix in real time from the measurable output tracking error alone, requiring no plant
identifier, no state observer, and no knowledge of the patient’s physiological parameters. The
plant is described by the extended four-state Bergman Minimal Model, cast in the nonlinear
state-space form x˙ p = Ap(xp)xp + Bpu. Closed-loop stability is established via the Almost
Strict Passivity (ASP) framework: a dynamic Parallel Feedforward Compensator renders
the augmented plant ASP, and a Lyapunov-based theorem proves uniform boundedness of
all closed-loop signals and asymptotic convergence of the tracking error. The controller is
validated through three simulation studies covering ten virtual patients with ±30% parameter
uncertainty, ten clinical scenarios from normal fasting to critical hyperglycaemia, and a fullday
three-meal protocol. Blood glucose is maintained within the safe zone [70, 180] mg/dL in
all cases with no hypoglycaemic events, confirming the clinical reliability and computational
efficiency of the proposed approach.