An efficient and multi-fidelity reliability-based design optimization method based on a novel surrogate model local update strategy

Reliability-based design optimization (RBDO) constitutes a crucial methodology for batch product quality improvement, attracting widespread attention in industrial engineering. The efficiency in calculating the products' performance parameters and the precision in determining the most probable target point (MPTP) are considered two pivotal factors influencing the implementation effects of RBDO. State-of-the-art RBDO methods typically overlook the risks associated with sample aggregation and misclassification of constraint functions, resulting in inefficiency, imprecision, and a pronounced deviation between the optimization scheme and engineering reality. Here, we propose an efficient and multi-fidelity reliability-based design optimization method. Initially, an innovative strategy for locally updating the surrogate model based on a curvature learning function is introduced to address the misclassification of constraint functions and balance the increasing samples' effectiveness and computational accuracy. Subsequently, a novel adaptive MPTP-solving strategy is applied to search for approximate MPTP. Based on this, an efficient and multi-fidelity RBDO framework is formulated, synchronously enhancing the efficiency, precision, and applicability of RBDO. Finally, the effectiveness of the proposed approach is verified by applying the proposed method to three numerical examples and a practical engineering instance involving an electromagnetic relay.