The quest for artificial pancreas can be structured in three tasks: glucose sensor,
insulin pump and control algorithm. The latter is a key point of the diabetes “closing
the loop” problem and its primary prerequisite is a valid model able to describe the
blood glucose system. Among the many models appeared in the literature, the model
of Magni et al [9] is widely used and represents a relatively complex nonlinear model
with glucose absorption as well as subcutaneous glucose and insulin dynamics incorporated
into its structure. Our aim is to hide the nonlinearity of this model by transforming
the signal coming from a linear controller so that the response of the model would
mimic the behavior of a linear system, desirably the one acquired through steady-state
linearization; hence the validity of linear controllers could be extended. The nonlinear
method known as asymptotic tracking of a linear system and presented in [13], needs
the values of the state variables; hence a Kalman-filter extended for nonlinear systems
is used. The capabilities of this approach are examined through simple control algorithms
and realistic input scenarios.
