Abstract:
The simulation-based design optimization and reliability assessment of complex systems is hampered by several hurdles such as large computational times, the difficulty to evaluate sensitivities, the potentially acute sensitivity of responses to uncertainties, and the presence of multiple failure modes. This considerably limits the blind use of traditional optimization and probabilistic methods.

The seminar will introduce a classification-based strategy that facilitates design optimization and reliability assessment by constructing explicit boundaries of the feasible or failure space in terms of the design and random variables. The boundaries are constructed using a Support Vector Machine (SVM) and are iteratively refined using a dedicated adaptive sampling scheme. This methodology does not require response approximations and is therefore ideal for problems with discontinuities and binary outputs.