Fund Project:Project supported by the National Natural Science Foundation of China (Grant Nos. 11502292, 11572348).
Received Date:22 January 2019
Accepted Date:28 March 2019
Available Online:01 June 2019
Published Online:20 June 2019
Abstract:The harmonic balance method (HBM) is an efficient frequency-domain approach to computing periodically unsteady flows. The basic principle of this method is to decompose the flow variables into a Fourier series, and transform the unsteady flow into several steady problems coupled by a spectral time-derivative operator, from which the whole time history of a complete unsteady periodic flow can be reconstructed. In the present work, we investigate the ability of the HBM to be used for modeling the periodic unsteady vortex shedding behind a bluff body at low Reynolds numbers via solving the unsteady incompressible Navier-Stokes equations. For the periodic problem where the time period T of the unsteadiness is unknown, a variable-time-period method based on residual gradients is used to compute the exact time period iteratively starting from an initial guess T0. By simulating the two-dimensional laminar flows over a circular cylinder and a square cylinder, the accuracy and efficiency of the HBM are investigated and the effects of different parameters on the final results are analyzed. Comparisons with the results of fixed-time-period HBM using a constant time period are also implemented. Three practical methods of optimization are used to iterate the time period, and the values of accuracy and efficiency of different methods are compared with each other. The results show that the HBM can accurately capture the complex nonlinear flow field physics with only three harmonics. The Strouhal frequency and mean drag coefficient each as a function of the Reynolds number agree well with existing experimental and computational data. For both test cases, the computational efficiency of HBM is higher than that from the traditional time-domain method. For the square cylinder test case, the HBM offers speedup rate up to nearly 18 times. The real time period of vortex shedding can be predicted by the gradient based variable-time-period method, and the final result is insensitive to search step λ. The calculation result is sensitive to the initial T0, and when such a variable is greater than a certain value, the result will converge to an approximate integer multiple of the real one. Therefore, it deserves further exploration on how to specify this initial condition. The shedding time periods computed by different optimization methods are converged to the same value. The computational efficiency from the FR conjugate gradient method and that from Newton method are both equivalent to that from the steepest descent method with the maximum search step λ = 100. Avoiding prescribing parameters such as the search step λ, the Newton method possesses higher application value in engineering calculation than the other two schemes. Keywords:periodic unsteady flows/ numerical simulation/ harmonic balance method/ variable-time-period method
表3不同谐波数下的计算结果 Table3.Strouhal number and time-averaged coefficient computed by different number of harmonics.
图 12Re = 180, NH = 3条件下不同时刻的流线图 (a) t = T/3; (b) t = 2T/3; (c) t = T Figure12. Streamlines at various time instances over one period (Re = 180, NH = 3): (a) t = T/3; (b) t = 2T/3; (c) t = T.
图 18T = 11.43 时重建的升力系数曲线 Figure18. Variation of CL over one period with converged time period T = 11.43.
34.2.4.计算效率分析 -->
4.2.4.计算效率分析
图19比较了采用HBM与TDM计算得到的St随雷诺数的变化曲线. 当时域计算选择时间步长为?t = 0.01时, 在雷诺数$ 80 \leqslant Re \leqslant 180$之间两种计算方法的结果符合良好, 雷诺数低于80时, HBM的计算结果比TDM结果更接近实验值. 但当TDM计算采用较小步长?t = 0.001时, 较低雷诺数下的TDM计算结果与HBM计算保留前3阶谐波分量的结果基本重合, 不同的是此时的TDM付出的计算代价更大. 图20为雷诺数Re = 180和Re = 60的加速比. 由图20可知, 来流雷诺数越低, HBM相对TDM的计算效率优势越明显. 图 19 HBM计算的St与TDM计算结果的对比 Figure19. Comparison of the HBM St data results with TDM results.
图 20 不同雷诺下的加速比 Figure20. CPU time speedup of various Reynolds number.
24.3.基于固定周期谐波平衡法的涡脱落周期辨识 -->
4.3.基于固定周期谐波平衡法的涡脱落周期辨识
对于固定周期HBM计算, 当给定的周期T不等于真实的物理周期T*时, 残差将下降到一定水平, 并且收敛后的残差和气动力会随着虚拟迭代步的增加呈现周期性变化, 如图21所示, 其中ρ为密度; u, v分别为x方向和y方向的速度; e为单位质量总能. 当Re = 180时, 采用可变周期HBM计算的圆柱绕流涡脱落周期为T* = 5.389. 固定周期HBM给定T = 4时, 计算得到的升力系数和残差都呈现周期性变化的趋势, 相邻迭代步重建的升力系数曲线之间存在相位差, 如图22所示. 对于给定的T, 当升力系数收敛后每一步虚拟迭代产生的相位差是常值. 而且这个相位差与给定的周期T呈近似线性的关系, 当且仅当给定的T为真实的周期T*时相位差为0, HBM计算得到各个时刻的升力系数曲线不再周期性振荡, 而是收敛到常值, 如图23所示. Spiker等[32]提出的相位差方法正是基于这一点来辨识涡脱落的频率. 图24给出了固定周期HBM计算时, 相位差随周期的变化. 由图24可知, 在一定范围内, 相位差的确与T成线性关系, 采用固定周期HBM辨识涡脱落周期时, 只需计算两个不同周期对应的相位差就可以通过线性关系推算相位差等于0时对应的T*. 值得注意的是, 采用3阶谐波进行计算时, 当T = 11.43时相位差也等于0, 在该值附近也存在线性关系, 说明11.43也是旋涡脱落的周期, 这与之前可变周期HBM的计算结果一致. 图 21 升力系数和t = T时刻的残差收敛曲线(Re = 180, T = 4, NH = 3) (a)升力系数; (b)残差 Figure21. Time history of lift coefficient CL at various time instances over one period and residual at t = T (Re = 180, T = 4, NH = 3): (a) Lift coefficient; (b) residual.
图 22 不同迭代步重建的升力系数随时间的变化(Re = 180, T = 4, NH = 3) (a)整体; (b)局部 Figure22. Variation of CL over one period at different iterations: (a) Overall; (b) local.
图 23T = 5.389时各个时刻升力系数收敛曲线(NH = 3) Figure23. Time history of lift coefficient CL at various time instances over one period with T = 5.389 (NH = 3).
图 24 相位差随周期T的变化(Re = 180, NH = 3) Figure24. Change in phase of unsteady lift versus time period for Re = 180 (NH = 3).
GBVTP的实现是基于残差在T = T*时最小的物理事实, 采用SDM沿着残差下降的方向进行搜索. 图25为采用固定周期HBM计算的残差随周期T的变化. 由图25可知, T的精度决定了HBM计算的精度. 当给定的周期远离T*时, 残差下降越来越困难, 当T = T*时残差迅速下降. 因为, 当且仅当给定的T为真实的涡脱落周期才能满足非定常流动控制方程, 使得残差接近于0. 因此, 采用3个谐波数进行计算时, T = 5.389和11.43都是旋涡脱落的周期, 这与可变周期计算的结果也相一致. 图 25 残差随周期T的变化(Re = 180, NH = 3) Figure25. HBM solution residual versus time period for Re = 180 (NH = 3).
分别采用以上方法对Re = 180的圆柱绕流进行数值模拟, 对比他们的计算精度和效率. 图26给出了牛顿法及SDM的计算结果, 可以看出, 牛顿法模拟的涡脱落周期与SDM的结果一致. 采用SDM时, 需要设置搜索步长λ, 步长越大收敛越快. 而采用牛顿法计算时不需要设置此参数, 其收敛曲线介于SDM采用λ = 1和λ = 100的收敛曲线之间, 且接近于λ = 100的计算结果. 与SDM类似, 共轭梯度法也需要设置搜索步长λ. 在本文算例中, 取不同时间步长计算时, 共轭梯度法计算得到的周期T收敛曲线基本重合, 步长对共轭梯度法计算结果的影响不大, 如图27所示. 共轭梯度法的收敛速度与SDM采用λ = 100时的相同. 图 26 采用牛顿法和SDM计算的周期T收敛曲线对比图 (a)初始T0 = 4; (b)初始T0 = 5.41 Figure26. Convergence of shedding time period computed by Newton method and SDM: (a) T0 = 4; (b) T0 = 5.41.
图 27 采用FR法计算的周期T收敛曲线(a)及其与SDM计算结果的比较(b) Figure27. Convergence of shedding time period computed by FR conjugate gradient method (a) and compared with the SDM results (b).
可变周期HBM采用三种不同优化方法计算得到的周期T收敛曲线对比如图28所示. 由图28可知, 三种方法的计算精度相同, 最终计算得到的涡脱落周期T均为5.389. 共轭梯度法和牛顿法的收敛速度与SDM采用最大搜索步长时的收敛速度一致. 由于牛顿法不需要设置搜索步长等参数, 在实际工程计算中更有优势. 图 28 采用三种不同优化方法计算得到的周期T收敛曲线图 Figure28. Convergence of shedding time period computed by three different methods of optimization.