V2I8P37

Using Machine Learning-Based Control Co-Design for Feedback Control of Medication Delivery

Geku Diton^1*

Abstract

This study investigates the highs and lows associated with the formation of addiction and focuses on developing a feedback control loop for an infusion pump that optimizes drug concentration in the bloodstream according to specific criteria. To achieve this, the system’s mathematical model was analyzed to derive an open-loop transfer function, which was then used to apply a PID (Proportional-Integral-Derivative) controller for feedback regulation. In addition to traditional control methods, both machine learning (ML) and deep learning (DL) techniques were explored as classifiers to assist the pump in administering accurate doses. The target drug concentration in the patient’s bloodstream was determined to be 7.55 mg/ml, which served as the steady-state setpoint for the transfer function. By incorporating the PID tuner into the feedback loop, the system was optimized to meet design requirements, including a rise time of less than 25 minutes and a maximum overshoot of 5% above the setpoint concentration. Machine learning classifiers, including Naïve Bayes (NB), Decision Trees, and Support Vector Machines (SVM), achieved an impressive classification accuracy of 100%. Additionally, a deep learning (DL) model was successfully developed to predict patient classification based on their metabolic profile. This research lays the foundation for the development of a feedback-driven infusion pump and an algorithm capable of classifying patients according to their metabolism. The goal is to provide doctors with a tool that customizes the pump’s dosage for each patient, ensuring they receive the correct amount of medication tailored to their individual needs.

Keywords:

Addiction modeling; PID controller; Infusion pump optimization; Machine learning classifiers; Personalized drug dosage