Increasing security in an artificial pancreas: diagnosis of actuator faults

The current and incoming technology for the treatment of type 1 diabetes mellitus (T1DM) suggests that the concept of an artificial pancreas is a feasible goal [1]. Many engineering areas have key roles in the development of such system since they are involved in every part of the process (monitoring, dosing algorithms and delivering). Control engineering has presented many advances in the dosing strategies and some research groups are developing their own prototypes and plan to test them in a small-scale clinical set of trials [2]. As any other automated system, an artificial pancreas is susceptible to faults, hence the study of their possible effects and corrective solutions becomes an important topic in health care systems. In this sense, this paper addresses model-based fault diagnosis for an artificial pancreas, which consists of a continuous glucose monitor, a portable insulin pump and a glucose control strategy. Specifically, this work is focused on sub- and over-dosing scenarios caused by the malfunction of the insulin delivering device. The T1DM mathematical model developed in [3] is employed to design the fault diagnosis algorithm and its performance is evaluated in a virtual environment, where the patient inter-variability and the meal carbohydrate content variations can be simulated.