Title: Data-Driven Prediction and Adaptive Decision-Making for Sequential Monitoring of Complex Systems

Date: Tuesday, April 8th, 2025.

Time: 11am EST

Location: Groseclose 304

Meeting link: https://gatech.zoom.us/j/97279783344

Dongmin Li

Industrial Engineering PhD Student

H. Milton Stewart School of Industrial and Systems Engineering

Georgia Institute of Technology

Committee

Dr. Xiaochen Xian (Advisor), H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology

Dr. Jianjun Shi, H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology

Dr. Roshan Joseph, H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology

Dr. Xiao Liu, H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology

Dr. Hongcheng Liu, Department of Industrial and Systems Engineering, University of Florida

Abstract

Sequential monitoring of complex systems, such as advanced manufacturing and environmental monitoring, is essential for ensuring system reliability and performance. However, challenges arise when evaluating system status from real-time observations and making detection decisions. Resource constraints, such as limited sensors and data storage, restrict observations to a subset of system variables, requiring adaptive sampling decisions on data collection that account for uncertainty in system status evaluation and potential changes. In addition, system variables often exhibit intricate interactions and may evolve over time, requiring the consideration of the interactions and prediction of system evolution for accurate and timely detection. To address the challenges, this thesis proposal leverages available data and explores the incorporation of system knowledge to enhance system status prediction, evaluation, and adaptive decision-making, ensuring effective detection.

In Chapter 2, we propose data-driven sampling strategies for moving vehicle-based sensors (MVSs) to quickly detect abrupt changes in an area. By integrating statistical process control and mathematical optimization, the strategies handle noisy and partial observations and adaptively adjust MVS routes under movement constraints. In Chapter 3, we propose a Bayesian jump model-based pathwise sampling approach that further improves the detection performance using MVSs by incorporating spatial correlation and quantifying uncertainties. Chapter 4 provides a theoretical analysis of the average detection delay under partial observations for the design of key parameters, such as the number of sensors, to optimize statistical and/or economic performance. Chapter 5 presents a physics-informed machine learning framework to predict droplet evolution in the inkjet printing process, integrating the governing Navier-Stokes equations, which describe velocity and pressure evolution, with indirect droplet shape observations.