Hybrid adaptive sequential sampling for reliability-based design optimization

To reduce the cost for reliability analysis and optimization of complex engineering systems, surrogate models can be used to replace expensive physical models. One of the critical tasks of employing surrogate models for system design optimization is to develop an accurate surrogate model cost-effectively. In this thesis, a new hybrid adaptive sequential sampling strategy has been developed to substantially improve the efficiency of the surrogate model development process. The developed sampling strategy combines local sampling that focuses on regional model fidelity improvements with global sampling that ensures effective design updates in the design optimization process. Specifically, a confidence-guided sequential sampling scheme is developed for local sampling, which identifies most useful sample points along the descending direction of the objective function as well as the constraints to improve the regional model fidelity. Similarly, a constraint boundary sampling scheme is adopted for the global sampling purpose, which efficiently locates the constraint boundaries and balances the efforts devoted to global sampling and local sampling processes. The efficacy of the developed hybrid adaptive sequential sampling technique for reliability-based design optimization using surrogate models is assessed with several numerical case studies, through comparisons with existing approaches that have been reported in the literature. The case study results have demonstrated that the developed new sampling strategy can significantly reduce the number of sample points required in updating the surrogate model along with the design optimization process. By using the developed adaptive sequential sampling strategy for surrogate modeling, the design processes become more efficient and cost-effective.