Date of Award

Fall 2017

Document Type


Degree Name

Doctor of Philosophy (PhD)


Mechanical & Aerospace Engineering

Committee Director

Miltiadis Kotinis

Committee Member

Gene J.-W. Hou

Committee Member

Charles Keating


Particle Swarm Optimization (PSO) has received increasing attention from the evolutionary optimization research community in the last twenty years. PSO is a metaheuristic approach based on collective intelligence obtained by emulating the swarming behavior of bees. A number of multi-objective variants of the original PSO algorithm that extend its applicability to optimization problems with conflicting objectives have also been developed; these multi-objective PSO (MOPSO) algorithms demonstrate comparable performance to other state-of-the-art metaheuristics. The existence of multiple optimal solutions (Pareto-optimal set) in optimization problems with conflicting objectives is not the only challenge posed to an optimizer, as the latter needs to be able to identify and preserve a well-distributed set of solutions during the search of the decision variable space. Recent attempts by evolutionary optimization researchers to incorporate mathematical convergence conditions into genetic algorithm optimizers have led to the derivation of a point-wise proximity measure, which is based on the solution of the achievement scalarizing function (ASF) optimization problem with a complementary slackness condition that quantifies the violation of the Karush-Kuhn-Tucker necessary conditions of optimality. In this work, the aforementioned KKT proximity measure is incorporated into the original Adaptive Coevolutionary Multi-Objective Swarm Optimizer (ACMOPSO) in order to monitor the convergence of the sub-swarms towards the Pareto-optimal front and provide feedback to Mamdani-type fuzzy logic controllers (FLCs) that are utilized for online adaptation of the algorithmic parameters. The proposed Fuzzy-Adaptive Multi-Objective Optimization Algorithm with the KKT proximity measure (FAMOPSOkkt) utilizes a set of reference points to cluster the computed nondominated solutions. These clusters interact with their corresponding sub-swarms to provide the swarm leaders and are also utilized to manage the external archive of nondominated solutions. The performance of the proposed algorithm is evaluated on benchmark problems chosen from the multi-objective optimization literature and compared to the performance of state-of-the-art multi-objective optimization algorithms with similar features.