Event-Triggered Model Predictive Control for Embedded Artificial Pancreas Systems.

Area of Science:

• Biomedical Engineering
• Control Systems Engineering
• Diabetes Technology

Background:

• Artificial pancreas (AP) systems require efficient control algorithms for low-energy, embedded devices to manage type 1 diabetes mellitus.
• Reducing controller updates is key to lowering AP energy consumption and enabling practical deployment.

Purpose of the Study:

• To develop an energy-efficient, event-triggered model predictive controller (MPC) for artificial pancreas systems.
• To reduce processor computations and controller updates without compromising glycemic regulation safety.

Main Methods:

• Designed an event-triggered MPC leveraging physiological insights for glucose management.
• Deployed the controller on a wearable platform and tested its robustness using the UVA/Padova metabolic simulator.
• Evaluated performance with and without meal announcements, assessing hypoglycemia, model mismatch, and meal misinformation.

Main Results:

• The event-based controller operated for 18 hours out of 41 with unannounced meals, maintaining target glucose levels.
• With meal announcements, the controller operated only 25.88% of the time, achieving similar glycemic control to standard MPC.
• Demonstrated robustness to various challenging conditions within the metabolic simulator.

Conclusions:

• A novel control architecture for AP systems was developed, enabling safe glycemic regulation with reduced computational demands.
• The framework integrates with existing MPC variants and customizes the trade-off between regulation and efficacy.
• Eliminates the need for pre-selecting controller sampling times, offering greater flexibility.