AN ADVANCED EVOLUTIONARY PROGRAMMING METHOD FOR MECHANICAL SYSTEM DESIGN: FEASIBILITY ENHANCED PARTICLE SWARM OPTIMIZATION


Tezin Türü: Doktora

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, Makina Mühendisliği Bölümü, Türkiye

Tezin Onay Tarihi: 2019

Öğrenci: MEHMET SİNAN HASANOĞLU

Danışman: MELİK DÖLEN

Özet:

Constrained optimization problems constitute an important fraction of optimization problems in mechanical engineering domain. It is not rare for these problems to be highly-constrained where a specialized approach that aims to improve constraint satisfaction level of the whole population as well as finding the optimum is deemed useful especially when the objective functions are very costly. This dissertation introduces a new algorithm titled Feasibility Enhanced Particle Swarm Optimization (FEPSO) to handle highly-constrained optimization problems. FEPSO, which is based on particle swarm optimization technique, treats feasible and infeasible particles differently. Infeasible particles do not need to evaluate objective functions and fly only based on social attraction depending on a single violated constraint, called the activated constraint (AC), which is selected in each iteration based on constraint priorities and flight occurs only along dimensions of the search space to which the AC is sensitive. To ensure progressive improvement of constraint satisfaction, particles are not allowed to violate a satisfied constraint in FEPSO. Unlike its counterparts, FEPSO does not require any feasible solutions in the initialized swarm. A modified version of the new method called the multi-objective FEPSO (MOFEPSO) is also introduced. MOFEPSO, which is capable of handling highly-constrained multi-objective optimization problems, employs repositories of non-dominated and feasible positions (or solutions) to guide feasible particle flight. In this study, several constrained optimization problems are described. For the given problems, the performance of FEPSO- and MOFEPSO are comparatively evaluated against a number of popular optimization algorithms found in the literature. The results suggest that FEPSO- and MOFEPSO are effective and consistent in obtaining feasible points, finding good solutions, and improving the constraint satisfaction level of the swarm as a whole..