Study On Variable Complexity Optimization Based On Approximate Models

Simulation-based design optimization is playing an increasingly prominent role in the designof everything from spacecraft to consumer products. Applying nonlinear optimization techniquesto simulation-based design becomes prohibitively expensive as computer models become morecomplex and increase in fidelity.Recently, variable complexity optimization have been developed to address this problem byincorporation low complexity models and high complexity models into one optimization frame-work.The focus of this thesis in on improving the efficiency and applicability of variable complex-ity optimization based on approximate models. Highlights of original contributions made in thisresearch include:(1) A variable complexity optimization framework based on approximate models.(2) A multiple scaling method that relieves designers from having to choose a priori whichscaling method, multiplicative or additive or hybrid, is most suitable to their problem with limitedinformation.(3) A deflective trust-region method based on gradient information, which make the methodmore efficient.(4) Kriging modeling method based on adaptive Latin Hypercube Sampling, which reducedthe sampling size.(5) Using benchmark problems and a real complexity simulation problem demonstrates theefficiency and applicability of the framework proposed in this thesis.