摘要
利用超级计算机资源求解雷诺平均N-S方程(RANS),计算评估大量外形方案性能,在可接受的时间周期内完成民机飞发集成构型下机翼多目标优化设计。搭建了集成机翼CST参数化、复杂外形网格变形、快速CFD流场解算和自动后处理等关键环节,用遗传算法全局寻优的优化系统。合理的流程设计使该系统可在"天河2号"超级计算机上同时对数百个方案实施计算评估。在并行计算、加速收敛等技术的综合运用下,使用包含800万单元的多块结构化网格,对NASA CRM (Common Research Model)机翼/机身/短舱/吊挂构型的计算分析可在15min内结束。在3点3目标优化案例中,用90个设计变量表达CRM机翼9个控制剖面的中弧线和扭转角,60h内完成了超过10000个外形方案的计算分析,遗传进化40代。与初始外形相比,PARETO前缘上选择的最优解的各设计点取得了2～10count(1count=阻力系数0.0001)的减阻效果。
Recent effort was presented on multi objective wing shape optimization of NASA Common Research Model （CRM） in wing body nacelle pylon （WBNP） configuration. The performance of a large number of designs was evaluated by solving Reynolds Averaged Navier Stokes （RANS） equations. Parallel computations and supercomputer resources were employed to make the optimization process completed in acceptable time cycle. An automated optimization framework was integrated, which consists of CST based wing parameterization, mesh deformation, flow solving, post processing, and genetic algorithm based global optimization. When running the framework on supercomputer, hundreds of designs can be evaluated at the same time. By using a multi block mesh with 8 million cells, the solution of CRM WBNP configuration can be obtained within 15 minutes with the help of parallel computation and convergence acceleration techniques. In a typical 3 objective optimization application, the evaluation of 128 candidates was carried out simultaneously with computational fluid dynamics on Tianhe 2 supercomputer. Total 90 design variables were used to represent camber line shapes and twist angles of 9 control sections. It took 60 hours to complete the evolution for near 40 generations. The drag reduction of 2 to 10 counts for each objective is achieved comparing with the baseline shape.
作者
薛帮猛
张文升
张志雄
XUE Bangmeng;ZHANG Wensheng;ZHANG Zhixiong(Beijing Key Laboratory of Simulation Technology for Civil Aircraft Design,Beijing Aeronautical Science and Technology Research Institute of COMAC,Beijing 102211,China)
出处
《空气动力学学报》
CSCD
北大核心
2018年第6期941-948,共8页
Acta Aerodynamica Sinica
基金
国家自然科学基金(11402306)
关键词
飞机/发动机集成
多目标优化设计
大规模并行计算
计算流体力学
网格变形
engine aircraft integration
multi objective optimization
massive parallel computation
computational fluid dynamics
mesh deformation
