Reliability-Based Design Optimization Considering Model Uncertainty

Abstract Reliability-based design optimization (RBDO) ensures the optimal design satisfying both the performance improvement and the target reliability under uncertainties. Current RBDO methods often assume that all the input uncertainties are explicitly known (e.g., given random variables or random processes), and the performance function is deterministic under a given realization of the input random variable/process. However, in many realistic engineering problems, performance function is intrinsically uncertain due to that the performance model itself is uncertain. Many existing researches neglect such model uncertainty, which results in significant deviations between theoretical prediction and actual structural performance. To overcome this limitation, this paper proposes a novel RBDO method considering model uncertainty. The performance function with model uncertainty is mathematically characterized using spectral expansion, and a spectral-expansion-based stochastic surrogate model is constructed to approximate the performance function as a random field. Based on the output distribution and the mean response of the spectral-expansion-based stochastic surrogate model, a shift scalar of the output distribution is calculated to decouple the RBDO process into the performance response space rather than the input design space, which enables an efficient optimization process through deterministic design optimization. The proposed RBDO method is applied in two numerical examples and one engineering example that designs an 8 megawatt wind turbine tower under dynamic wind loads.