Date of Graduation

12-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Engineering (PhD)

Degree Level

Graduate

Department

Industrial Engineering

Advisor/Mentor

Sullivan, Kelly

Committee Member

Rainwater, Chase

Second Committee Member

Zhang, Shengfan

Third Committee Member

Xiang, Yisha

Keywords

Network optimization; Time-based nodes; Partial survival signature; Incremental network design; Selective maintence maxium flow

Abstract

In this dissertation, we consider three types of network optimization problems. In Chapter 1, we consider a network maintenance problem which focuses on time-based redeployment of multi-class nodes for reliable wireless sensor network coverage. Whereas previous research on time-based node redeployment assumes nodes are identical with respect to time to failure, we use multiple classes of sensor nodes to represent a scenario where nodes’ times to failure are dependent on positioning in the network. We propose a partial survival signature (PSS) approach for estimating area coverage reliability under a given time-based redeployment policy, where the PSS is estimated by Monte Carlo simulation. This PSS representation enables efficient re-evaluation of coverage reliability under different redeployment policies, thus allowing the use of metaheuristics to obtain a set of redeployment policies that are near-efficient with respect to cost and coverage reliability. In chapter 2, we consider an incremental network design problem (IND-MF) in which we determine the sequence of failed edges to repair to maximize the cumulative maximum source- to-sink flow over a planning horizon comprised of a finite number of time periods. We develop a deterministic dynamic program (DP) model and solve IND-MF problem using the maskable proximal policy optimization (M-PPO) RL algorithm in which in which masking restricts the action set at each state to failed edges that appear in a minimum source-sink cut. In Chapter 3, we consider a selective maintenance maximum flow (SM-MF) problem where edge failures occur probabilistically in each time period. Similar to IND-MF, we apply a maskable proximal policy optimization (M-PPO) RL where masking at each time is defined based on the minimum cut of the network. We conduct case studies to compare the performance of the proposed masking strategy, and the results showed significant performance gains relative to the base masking strategy.

Download

Share

COinS