Reducing controller updates via event-triggered model predictive control in an embedded artificial pancreas

Abstract

An essential component in the design of a wearable or implantable artificial pancreas is a low-power, embedded control algorithm capable of tight glucose regulation in people with Type 1 Diabetes Mellitus (T1DM). In this paper, physiological insights into glucose management are leveraged to formulate a safety-constrained, event-triggered model predictive control (MPC) algorithm customized to reduce the number of controller updates, leading to reduced processor runtime and lowered energy expenditure. The proposed event-triggered MPC is deployed on a wearable (65 mm × 30 mm × 5 mm) embedded prototyping platform running a single core 1 GHz 32-bit ARM11 processor. The efficiency of the proposed method is verified in a hardware-in-the-loop simulation study with ten subjects simulated in-silico using the United States Food and Drug Administration (US FDA) accepted UVA/Padova Metabolic Simulator. The controller is challenged with large, unannounced meals and sudden hypoglycemia. Empirical studies reveal that the proposed event-triggered MPC remains idle for 50% of the time in closed-loop, while maintaining 32 (out of 53) hours in the clinically accepted safe glucose region (70–180 mg/dL), compared to 35.5 hours for its time-triggered counterpart, with reduced time in hypoglycemia. Thus, the proposed strategy significantly reduces the number of controller updates required, without compromising safety in glycemic regulation.