Abstract:

Sandwich structures were widely used in aerospace and other fields due to their high stiffness-to-weight ratio characteristics. The calculation costs of compliance simulation analyses in optimization design processes were significantly higher than that of weight constraints. However， the existing homogeneous algorithms assumed that the objectives were equivalent to the costs of constraint evaluation， which led to poor optimization adaptability and low efficiency. Thus， a constraint-objective two-stage optimization framework was designed based on feasibility rate to match adaptive optimization direction for real-time optimization paths. In the first stage， a dual offspring population collaborative optimization strategy of exploratory mutation-constraint relaxation screening and exploitative mutation-uncertainty screening was proposed to simultaneously enhance the level of constraint optimization and the reliability of the surrogate model， and the partial evaluation strategy was designed to save time-consuming objective evaluation. In the second stage， the search type was defined by combining feasible solution clustering analyses and dynamic threshold， and the surrogate model modeling and evolution strategy were adjusted adaptively. Under three classical loads， the proposed algorithm obtains optimal structures comparing with the gradient algorithm and other state-of-the-art algorithms of the same type， which confirms the effectiveness in practical applications.

Key words: sandwich structure, expensive optimization, surrogate model, differential evolution algorithm

摘要：

夹层结构因其高刚度-质量比特性而被广泛应用于航空航天等领域，其优化设计时的柔顺度仿真分析计算成本显著高于重量约束，但现有同类型方法假设目标与约束评估成本等价而导致优化适应性差、效率低下。为此，设计了基于可行率划分的约束-目标两阶段优化架构，为实时优化轨迹匹配适应性优化方向。第一阶段提出探索型变异-约束松弛筛选及开发型变异-不确定度筛选的双子代种群协同优化策略，从而同步提高约束优化水平与代理模型可靠性，并设计了部分评估策略以节省高耗时目标评估；第二阶段结合可行解聚类分析与动态阈值划定搜索类型，自适应调整代理模型建模与进化策略。在3种经典载荷下，该算法相较于梯度算法及其他最先进的同类型算法均获得最优结构，证实了其在实际应用中的有效性。

关键词: 夹层结构, 昂贵优化, 代理模型, 差分进化算法