This study proposes a robust and data-driven framework for truss topology optimization under uncertainty in the external loads, using the buffered failure probability (BFP) as a reliability measure. Traditional reliability-based design optimization (RBDO) methods often rely on the probability of failure (PoF), which only reflects the frequency of constraint violation. PoF neglects the severity of violations in the distribution tail, potentially resulting in unsafe designs sensitive to rare but extreme external loads. In contrast, BFP captures tail behavior and penalizes risk-prone solutions promoting safer and more conservative designs. We formulate the truss topology optimization problem under the compliance constraint with the uncertain external load as a second-order cone programming (SOCP) problem. To better handle uncertainty in input data, we consider variation in sample weights in the BFP constraint. Since accurate estimation of BFP at a high-confidence level generally requires a large number of samples and only a small fraction of these samples contribute to the active constraints, we develop a working-set algorithm that selectively incorporates only some of the most violated constraints. This significantly improves computational efficiency without compromising optimality. Numerical experiments show that the proposed method achieves solutions nearly identical to those obtained using all constraints, while reducing computation time by up to 96.9%. These results demonstrate the practical value of the proposed BFP-based formulation combined with constraint selection strategies for safe and efficient structural optimization under uncertainty.
Fujiyama et al. (Wed,) studied this question.