关键词: 关键词：向心涡轮, 昂贵高维多目标优化问题, 代理模型辅助的多目标进化算法, 代理模型优化, 多学科设计优化

Abstract: An efficient multi-objective optimization algorithm, MO-GSGA/D(Multi-objective Generalized Surrogate-assisted Genetic Algorithm based on Decomposition), is proposed to effectively solve high-dimensional (100 dimensions and above) multi-objective design problems within a limited budget of 1,000 performance evaluations. Based on this algorithm, a multidisciplinary design optimization (MDO) study for radial inflow turbines was conducted. First, to address the bottlenecks of expensive, high-dimensional, and multi-objective design in radial inflow turbines, the proposed MO-GSGA/D utilizes a weight-based decomposition strategy to transform complex multi-objective problems into a series of dynamic single-objective subproblems. A Radial Basis Function (RBF) surrogate model is introduced to assist the evolutionary process. By implementing local search and individual pre-screening strategies, the number of computationally expensive high-fidelity CFD evaluations is significantly reduced. Meanwhile, non-dominated sorting and crowding distance mechanisms are incorporated to update the population, ensuring the uniformity of the Pareto front distribution. Convergence tests on 100-dimensional ZDT2 and DTLZ2 benchmark functions demonstrate that MO-GSGA/D achieves faster convergence compared to classical algorithms such as MOEA/D and NSGA-II. Subsequently, taking MO-GSGA/D as the core and integrating it with a 3D blade parametric modeling method, a multi-objective optimization with 16 design variables was performed to maximize isentropic total-to-total efficiency and minimize maximum root stress. The optimized results show an increase of 1.12% in isentropic total-to-total efficiency and a 17% reduction in maximum root stress. This study effectively enhances the structural strength of the radial turbine while significantly improving its aerodynamic performance, thereby validating the effectiveness of the proposed multidisciplinary optimization design method for radial inflow turbines.

Key words: Keywords: Radial inflow turbine, Expensive high-dimensional multi-objective optimization problem, Surrogate model-assisted multi-objective evolutionary algorithm, Surrogate-based optimization, Multidisciplinary design optimization