Date of Award
5-2025
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Mechanical Engineering
Committee Chair/Advisor
Dr. Cameron Turner
Committee Member
Dr. John Wagner
Committee Member
Dr. Laura Redmond
Abstract
Engineering design optimization often requires balancing multiple competing objectives while navigating complex, high-dimensional tradespaces. Despite significant advances in tradespace exploration techniques, current methods face critical challenges in efficiently navigating high-dimensional design spaces. As the number of design variables and competing objectives increases, traditional multi-criteria decision-making and optimization methods struggle to explore the vast solution space comprehensively. The curse of dimensionality exacerbates this issue, making it impractical to analyze all possible solutions and effectively identify trade-offs between objectives. Additionally, the computational complexity of high-dimensional spaces further complicates optimization, making exhaustive solution analysis infeasible. Traditional multi-objective optimization methods rely on static Pareto-front calculations, which are computationally expensive and inflexible in handling evolving constraints. This research introduces a reinforcement learning (RL)-based decomposition and coordination (DC) framework to systematically explore and optimize multi-objective tradespaces. The study applies this RL-driven methodology to a four-dimensional tradespace, where the agent learns optimal relaxation strategies to balance competing objectives. Through iterative relaxation, the RL agent discovers efficient designs and evaluates their Pareto efficiency in the feasible region. Unlike conventional optimization approaches, this method enables adaptive exploration by refining constraints and improving design efficiency through continuous learning. The findings demonstrate that RL can effectively optimize engineering trade-offs, reducing reliance on heuristic-based decision-making and manual iterations. By automating the exploration of high-dimensional tradespaces, this study provides a foundation for Machine Learning (ML)-driven design optimization, offering a structured and scalable approach to multi-objective decision-making in engineering design.
Recommended Citation
Kothavade, Aaditya, "Reinforcement Learning for High-Dimensional Tradespace Exploration in Multi-Objective Optimization" (2025). All Theses. 4458.
https://open.clemson.edu/all_theses/4458