基于Kriging的不确定性量化及鲁棒性优化的研究

Research on Kriging-Based Uncertainty Quantification and Robust Design Optimization

基于Kriging代理模型,提出了适用于多维、不同概率分布的不确定性量化方法。耦合自适应多目标差分进化算法和RANS方程求解技术,建立了鲁棒性优化设计框架。在数值验证基础上,考虑进口雷诺数和倒角半径的不确定性,以综合换热性能的均值最大、方差最小为目标,完成了内冷通道三角形涡发生器强化换热鲁棒性优化设计。最优设计的平均综合换热性能提升超过11.5%,且性能对不确定性变量的敏感度显著降低。最后,通过流场分析揭示了最优设计换热性能提升的主要原因。

An uncertainty quantification method was proposed based on Kriging surrogate model.Due to the advantage that all probability distributions can be employed in the method without limitations, the method can also be easily applied to various design cases. Based on the uncertainty quantification method, a Kriging-based robust design optimization framework is built by coupling Self-adaptive Multi-Objective Differential Evolution algorithm(SMODE) and 3D Reynolds-Averaged Navier-Stokes(RANS) Solver technique. Upon numerical validation, the robust design optimization is carried out, with the aim of simultaneously maximizing averaged thermal performance and minimizing variance. In the optimization process, the fillet radius and the inlet Reynolds number are taken as uncertainty parameters. After optimization, the averaged thermal performance of the optimal solution is greatly improved by 11.5% with reduced sensitivity to uncertainty parameters. At last,mechanism behind thermal performance improve of optimal design is explored by detailed flow analysis.