Banca de QUALIFICAÇÃO: JOÃO LUCAS CORREIA BARBOSA DE FARIAS

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : JOÃO LUCAS CORREIA BARBOSA DE FARIAS
DATE: 02/09/2024
TIME: 09:00
LOCAL: Remoto - Online
TITLE:

Control of an Automated Insulin Delivery System using Artificial Intelligence

KEY WORDS:

automated insulin delivery system, nonlinear control, feedback linearization, machine learning, reinforcement learning, recurrent neural networks.

PAGES: 40
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

Type 1 Diabetes Mellitus is a disease that affects millions of people worldwide. Recently, thanks to the advancements in the field of embedded devices, proposals for devices that inject insulin subcutaneously have emerged, aiming at automatically regulating blood glucose concentration in diabetic patients. The Automated Insulin Delivery (AID) system can provide a life with higher quality, autonomy, and comfort for patients. This work aims to design an intelligent nonlinear controller with an estimator developed with Artificial Intelligence (AI) algorithms for an AID system. An extensive review on diabetes is conducted introducing its definition, physiological context, diagnosis, global impact and available state of the art therapies. Furthermore, a literature review is carried out presenting the types of AID systems, current challenges, virtual patient models, control strategies and AI algorithms that shape modern designs. The Bergman, Hovorka and OHSU models are introduced along with their mathematical formulation and specific uses cases. The OHSU model, which incorporates physical activity to the insulin-glucose dynamics, is chosen to simulate the virtual patient population of this work. Then, Feedback Linearization controller is presented and mathematically described. The nonlinear controller proposed in this work is combined with an intelligent estimator to compensate for model uncertainties, external disturbances and time delays inherent of the biological system in question. The intelligent estimator consists of a combination of Machine Learning and Reinforcement Learning algorithms. The goal is to use a Recurrent Neural Network as a universal estimator to assist the control law in tracking the blood glucose concentration of the virtual patient to the desired range. Then, use a Reinforcement Learning algorithm to tune the hyperparameters of the neural network. The proposed control strategy will be trained offline to encompass the system dynamics and tune the neural network’s parameters. As the in-silico simulation takes place, the parameters are fine tuned with a small learning rate to keep the controller up to date with physiological changes. A Projection Algorithm is used to constraint the neural network’s weights in a confined region and avoid overdosage of insulin. The closed loop stability of the proposed control strategy is proven using Lyapunov stability theory. The overall goal of this work is to develop a controller for an AID system that will properly regulate blood glucose concentration without meal input from the patient and under physical activity. Finally, a detailed timeline is presented describing the steps to be taken in order to implement the aforementioned planning until the Thesis is completed.

COMMITTEE MEMBERS:
Presidente - 1445637 - WALLACE MOREIRA BESSA
Interno - 2885532 - IVANOVITCH MEDEIROS DANTAS DA SILVA
Externo ao Programa - 1338796 - PHILIPPE EDUARDO DE MEDEIROS - UFRN