一种面向离散伴随气动优化的高效混合时间推进策略

An efficient hybrid time integration method for discrete adjoint aerodynamic optimization

  • 摘要: 为提高伴随优化方法在飞行器宽速域优化问题中的适用性,本文通过结合LU-SGS(lower-upper symmetric Gauss-Seidel)、近似牛顿-克雷洛夫(approximate Newton-Krylov, ANK)以及牛顿-克雷洛夫(Newton-Krylov, NK)方法,提出了一种新型混合时间推进策略。该策略采用隐式LU-SGS方法开展初始流场迭代,并在残差下降一定范围后先后切换为ANK与NK方法,以实现计算快速收敛。为验证该混合时间推进策略的计算效率,本文针对某宽速域乘波体开展了多工况流场模拟与气动优化测试。测试结果表明,与传统隐式LU-SGS方法相比,本文所提出的混合推进策略可将残差收敛效率提升75%以上,并改善其容易在复杂流动问题中残差收敛停滞的缺点。与显式龙格库塔+ANK+NK混合策略及D3ADI(diagonalized diagonally dominant alternating direction implicit)+ANK+NK混合策略相比,该策略填补了两者难以实现超声速/高速流场问题稳定求解的不足。此外,在宽速域气动优化问题中,较之于使用单一LU-SGS方法的优化方法,应用该策略的优化方法可通过减少每次流场迭代时间将整体优化效率提升70%左右。

     

    Abstract: To improve the efficiency of the aerodynamic shape optimization based on the discrete adjoint method in a wide speed range, a hybrid time integration method is proposed. The hybrid method consists of the lower-upper symmetric Gauss-Seidel (LU-SGS) scheme, approximate Newton-Krylov (ANK) scheme, and Newton-Krylov (NK) scheme. The LU-SGS scheme is first employed to start the iteration, and then the ANK and NK schemes are sequentially used to accelerate the convergence rate. The relative convergence of the residual norm is employed to switch the three methods above. The ANK method is selected until the residual norm decreases by 2–4 orders of magnitude. The NK method is specified when the residual norm is 102 to 103 times the target residual. By this switching strategy, the hybrid scheme is able to accelerate the convergence rate and improve the efficiency of aerodynamic optimization. The results of the test cases demonstrate that the solver employing the LU-SGS+ANK+NK scheme is 75% faster than that using the LU-SGS scheme. Moreover, compared to the Runge-Kutta (RK)+ANK+NK and diagonalized diagonally dominant alternating direction implicit (D3ADI)+ANK+NK schemes, the LU-SGS+ANK+NK scheme exhibits greater robustness. Furthermore, in a wide-speed-range aerodynamic optimization case, the optimizer based on the hybrid integration scheme is about 70% faster than that based on the LU-SGS method by accelerating each flow field simulation.

     

/

返回文章
返回