Energy consumption reduction for an ultra-low-cost artificial pancreas using an event-trigger MPC strategy

This work aims to evaluate the performance of an impulsive Model Predictive Control (MPC) strategy with an event-triggered (ET) approach implemented in an artificial pancreas (AP) system to determine if insulin injection is necessary, avoiding the execution of the control algorithm and the use of hardware associated to that in each sampling period. The primary focus is addressing practical challenges such as energy autonomy and glucose regulation using real hardware, actuators, batteries, and virtual patients instead of relying solely on simulations. The proposed system integrates an ET-based MPC strategy with a custom-built ultra-low-cost insulin infusion mechanism driven by a stepper motor. The system is tested on two embedded systems, one with greater computing capacity, power consumption, and a 5000 mA battery (Raspberry Pi 3B) and another control unit with lower computing capacity and power consumption using a 1200 mA battery (Raspberry Pi Zero 2 W). Both are equipped with a power manager and integrated with a smart sensor that allows the control algorithm to not only not be executed during periods when injection is not required but also the device to remain off during periods of inactivity. Hardware-in-the-loop (HIL) testing is conducted to analyze energy autonomy and blood glucose regulation (BG). The evaluation includes scenarios with parametric variations, unannounced meals, and noise measurement, comparing the ET approach to the standard time-trigger (TT) strategy. The ET-based approach achieved significant energy savings compared with the TT strategy, extending battery life by 45%–60%, corresponding to an additional 100–186 operational hours while maintaining BG levels within an average range of 112–126 mg/dL and a time-in-range percentage of approximately 80%. The number of controller calculations was reduced by up to 60% for model 3B. For the Zero 2 W, emulation time increased from 17 to 53 h under the ET strategy, maintaining the BG regulation obtained by the TT base strategy. The ET approach avoids calculating and dosing unnecessary insulin in each sampling period to achieve the control objective. The strategy compensated the reduced controller executions with higher insulin doses in extended time intervals, where peaks increased from approximately 1.3 units with TT strategy to 5.3 units with ET approach. Unlike other works, which show energy savings based on assumptions made by running simulations only, this work presents actual measurements of the energy consumed by the device during BG regulation. By strategically managing device shutdown periods, the ET controller activation effectively reduces computational demand and enhances energy autonomy. This approach significantly extends the operational duration of the AP system as a portable device while maintaining effective BG regulation. The results demonstrate the potential of ET-based MPC strategies to address the energy and performance trade-offs in practical AP applications.