Date of Graduation

5-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Engineering (PhD)

Degree Level

Graduate

Department

Mechanical Engineering

Advisor/Mentor

Jensen, David

Committee Member

Hu, Han

Second Committee Member

Liao, Hatiao

Third Committee Member

Campbell, Jenn

Fourth Committee Member

Millett, Paul

Keywords

graph-theoretic dependency modeling, modular architecture, GPU-accelerated evolutionary optimization, agent-based simulation,

Abstract

This dissertation presents the development of HARNESS, a unified, domain-agnostic framework for the modeling, optimization, and evaluation of complex systems under conditions of degradation and uncertainty. The research addresses a persistent challenge in systems engineering: the lack of a generalizable, behaviorally grounded approach for quantifying resilience and robustness across diverse domains. HARNESS integrates graph-theoretic dependency modeling, GPU-accelerated evolutionary optimization, and agent-based simulation into a modular architecture that enables both system design and diagnostic analysis. The framework models system interdependencies through user-defined components, parameters, and constraints, allowing architectures to evolve dynamically while maintaining structural validity. A mutation-only genetic algorithm with a novel Dynamic Bracket Selection (DBS) method enables scalable, constraint-aware optimization, improving population diversity and convergence speed. Agent-based simulation introduces temporal, rule-driven modeling of degradation, cascading failure, and recovery dynamics, producing data-driven insights into system behavior under stress. Four new metrics: Normalized Performance Loss per Unit Degradation (NPLUD), Robustness Index (RI), Sustained Functionality Index (SFI), and Structural Integrity Index (SII); quantify resilience and robustness as measurable system properties rather than abstract traits. These metrics capture performance degradation, persistence, and structural survivability, enabling comparison across system types and operational contexts. Together, these components form a feedback-capable framework that unifies performance optimization with resilience evaluation. HARNESS demonstrates that resilience can be systematically designed, quantified, and analyzed rather than inferred post hoc. While current limitations include node-centric degradation modeling and deterministic agent behavior, future extensions will incorporate edge degradation, probabilistic agents, and simulation-informed optimization feedback loops. Through its methodological innovations and integrative design, HARNESS establishes a foundation for adaptive, mission-assured system development across engineering, defense, infrastructure, and socio-technical domains.

Download

Share

COinS