Reliability analysis of tensile membrane structures using active learning-aided metamodeling

The design and construction of tensile membrane structures (TMS) are undergoing standardization currently. Modern structural design is generally based on reliability analysis which assesses the structural performance under various limit states in a probabilistic sense. Traditional reliability methods work well for linear or mildly nonlinear systems, and are therefore unsuitable for the highly nonlinear TMS behavior. Probabilistic simulation methods provide a more robust framework for complex limit states, but are computationally prohibitive, besides the already computation-heavy form-finding and load analysis of TMS. To alleviate this computational burden, metamodeling techniques offer a surrogate for full-scale simulations. However, accurate metamodels often require a large number of training points, increasing the computational cost. This paper proposes the integration of active learning techniques into metamodeling by strategically choosing training points within the region of interest near the limit state, enabling a more accurate estimation of the reliability index, with fewer training samples compared to standard metamodeling methods. The proposed methodology is tested on diverse TMS shapes to demonstrate its effectiveness in evaluating their reliability for different (and complex) limit states. The results clearly demonstrate how the proposed approach achieves this with reduced computational costs and higher accuracy, compared to “conventional” approaches.