Development of algorithms to solve different key challenges facing design optimization

Optimization methods play an indispensable role in today’s competitive environmentand there are plenty of practical examples where such methods have been used toidentify better performing designs (Boeing 787 Dreamliner and NASA ST5 antenna).Increasing complexity of the problems have also led to the development of sophisti-cated mathematical models that can only be solved using computationally expensivenumerical simulations such as finite element methods (FEM), computational fluidsdynamics (CFD), computational electromagnetics (CEM), etc. Repeated use of suchnumerical simulations is necessary in the context of optimization, i.e., to identifyoptimum products and processes with outstanding performance features. In reality,such problems often involve a large number of constraints and often demand multipleperformance considerations.Over the decades, population based metaheuristics have proven to be efficient, robustand versatile methods for numerical optimization as they are more amenable to dealwith such black-box problems. The major downside of any of these population basedmetaheuristics is their extremely long run time. Therefore, it is no surprise that thedevelopment of fast and efficient metaheuristics is an actively pursued research area.In this thesis, an effort is made to address three key challenges facing the adoption ofpopulation based metaheuristics for practical design optimization. The first challengerelates to the development of an efficient and reliable optimization algorithm capableof dealing with constrained optimization problems. In particular, two novel constrainthandling mechanisms are introduced i.e., one with the concept of partial evaluationusing constraint sequencing and the other involving adaptive constraint handling. Thestudy is motivated by the fundamental question should one evaluate all constraintsof a solution even if it has violated one constraint? and what is the difference inthe underlying search process if multiple constraint sequences are used?. The secondcontribution reported in this thesis relates to the development of an algorithm totackle optimization problems involving more than four objectives, i.e., many objectiveoptimizations. In this context, an algorithm based on decomposition is introducedwhich extends the capability of the well-known MOEA/D to deal with many objectiveoptimization problems. The algorithm incorporates a systematic sampling scheme and the balance between convergence and diversity during the course of search is maintained via a simple preemptive distance comparison scheme. The third contribution made inthis thesis is in the area of robust design optimization where the effects of variousformulations are studied in the framework of six-sigma quality. Four different problemformulations of robust design and methods to solve them have been proposed.The performance of these algorithms/schemes is rigorously assessed using well es-tablished benchmark functions and a suite of engineering design optimization problems.The results assessed using various measures clearly indicate that the proposed develop-ments offer competitive advantages over existing schemes.Finally, a summary of the findings of the work is presented. In addition, futureissues and directions which could be pursued with the aim of making the algorithmsmore efficient for handling various types of optimization problems are identified