fluent用LES算水翼边界层选择中心差分引起的震荡如何消除
-
感谢李老师的及时回复!目前情况是这样
- https://aip.scitation.org/doi/pdf/10.1063/1.5045795 文献引用的文献[23]提到了一种filter,但是并不清楚怎样应用到fluent
- 试过dynamic也会震荡,貌似跟模型无关,因为要算边界层还是主要用WALE
3.目前fluent自带的差分格式包括upwind, 2nd upwind, 3ord MUSCL, quick, bound cd 和cd都试了,只有cd能比较精确的算准边界层的速度剖面和脉动强度,检索到涉及定量分析边界层的文献比如槽道流动,平板流动,绝大多数也是用的中心差分,只是雷诺数都偏低。
震荡并不影响边界层平均速度和脉动与实验的对照,但是瞬时流场会有这样的锯齿形状,所以Q涡显示时会有一些非物理的碎涡出现,怎么都感觉会对后处理有影响。。
-
有试过二阶中心差分格式中耦合小部分一阶迎风格式吗,比如给个0.02%的一阶迎风格式,之前用fluent算3900的圆柱绕流,用纯二阶中心出现过数值振荡容易发散的情况,改成混合之后收敛性好很多,精度也还行,没有很明显的下降
-
@cccrrryyy 直接用TUI命令:
solve/set/numerics/
一直回车到1-st order to higher order blending factor: 1就是纯二阶,0.98就是耦合0.02的一阶迎风
感觉有点像TVD格式,也不是很懂,导师之前教我的操作,具体也不是太懂 -
感谢老铁的建议,我查了查好像没有这种现成的耦合操作,bounded cd可能是最像的,但是经过计算发现在边界层应该是全部切换成迎风格式了,没有部分耦合这种机制。
在国外cfd论坛上找到一个相似的情况,幸运的是那个人好像不关注边界层就改成迎风就没有震荡了,我看有个人的回答也解释了为什么除了中心差分都算不太准边界层。。
The choice of the scheme for LES should be based on the numerical diffusion it introduces. Ideally you should use linear(中心差分), or cubic. You need good and prett uniform mesh anyways for LES, since you are assuming the filter operator and the differential operator are commutative, which is not true on non-uniform grids! The filtered schemes are a last resort scheme in my view, if you can't really do better with your mesh. Upwind and linearUpwind, but also limitedLinear and QUICK should be avoided because they are too dissipative.
可惜就是无法消除震荡 -
@李东岳 https://www.cfd-online.com/Forums/openfoam-solving/106700-dicretization-schemes-les-pitzdaily.html 李老师,这是钝体扰流那个原贴,里面好像发了几个算例文件,不过我Openfoam水平有限,您可以参考下帖子
-
试了下,把水的粘性增大了一百倍,也就是雷诺数降了100倍,震荡就没有了,但是边界层整个就加厚了几十倍。。难道高雷诺数下解析边界层是无解的吗
-
@李东岳 谢谢李老师 您有时间的话可以试试foam的中心差分格式,我用CFX和fluent都会有震荡。这是约450万的网格文件:https://cowtransfer.com/s/82c878e62de64d
介质为水,边界条件inlet 5m/s, outlet 静压45300 Pa,wall为滑移壁面,foilwall为无滑移壁面,symmetry面设为symmetry,LES模型用的WALE,时间步长用的5e-5 -
@gengchen 在 fluent用LES算水翼边界层选择中心差分引起的震荡如何消除 中说:
目前fluent自带的差分格式包括upwind, 2nd upwind, 3ord MUSCL, quick, bound cd 和cd都试了,只有cd能比较精确的算准边界层的速度剖面和脉动强度
Regarding this, do you have any plots?
-
李老师,可以先看下0.15弦长处的速度剖面,我是在这个位置插入一根沿z方向4mm线导出平均速度,x=0.0002m, y=-0.059182m, z=0.016964m到0.020964m,根据最大值无量纲化,图表附在文件里。velocity_profile_ldv_test_0.15C.zip
-
抱歉,李老师,那我直接把LDV的数据贴上吧,谢谢老师!
0.51874 0.02 0.54523 0.04 0.57821 0.06 0.60027 0.08 0.61619 0.1 0.63168 0.12 0.65259 0.16 0.67519 0.2 0.69035 0.24 0.71312 0.28 0.72578 0.32 0.75144 0.38 0.76757 0.42 0.78385 0.46 0.79825 0.5 0.81624 0.54 0.83174 0.58 0.86305 0.66 0.89698 0.74 0.9205 0.82 0.94188 0.9 0.96355 0.98 0.98842 1.14 0.99863 1.3 1.00000 1.46 0.99797 1.62 0.99706 1.78
-
- is it time-averaged U predicted by LES?
- What is the maxima? the Umax in this line or in the computational domain?
Your results look quite good. Is it predicted by linear scheme? How about the other schemes, e.g., limited schemes?
I am runing LES simulation, but it is very slow. I try to limit Co under 0.6. Since the global Umax is around 10 m/s, it indicates
. What value is your global Umax?Courant Number mean: 0.00518159 max: 0.59921 deltaT = 2.67779e-06 Time = 0.0168822 GAMG: Solving for p, Initial residual = 0.0474207, Final residual = 0.00156874, No Iterations 2 time step continuity errors : sum local = 3.90116e-10, global = -6.12223e-14, cumulative = 5.73243e-11 DICPCG: Solving for p, Initial residual = 0.0105353, Final residual = 9.99036e-07, No Iterations 608 time step continuity errors : sum local = 2.48515e-13, global = -3.59741e-15, cumulative = 5.73207e-11 ExecutionTime = 401.48 s ClockTime = 421 s fieldMinMax minMaxp write: min/max(mag(U)) = 0 10.5464 Courant Number mean: 0.0051816 max: 0.601813 deltaT = 2.61829e-06 Time = 0.0168848 GAMG: Solving for p, Initial residual = 0.0479045, Final residual = 0.00156265, No Iterations 2 time step continuity errors : sum local = 3.72593e-10, global = 2.37179e-14, cumulative = 5.73444e-11 DICPCG: Solving for p, Initial residual = 0.0102726, Final residual = 9.97099e-07, No Iterations 152 time step continuity errors : sum local = 2.38352e-13, global = 2.12724e-14, cumulative = 5.73657e-11 ExecutionTime = 402.54 s ClockTime = 423 s
-
I did a rough calculation, my time step is 3.3e-6, Co = 0.6, my server takes 2 s to calculate one time step. If I want to calculate 0.5 s, it takes me
days to run, which looks too long for me.Another point we need to discuss is that, if Umax is 11 m/s, to ensure Co < 1, you should use similar time step like mine.
试了从1e-5到5e-5
No matter 1e-5 or 5e-5, the Co is larger than 1. Unless time step is smaller than 5e-6, Co is smaller than 1.
Meanwhile, I heard that commercial software can employ much larger Co number. Since the code is not public to the users, I dont understand why.
-
@李东岳 谢谢李老师的说明,我这边只试了中心差分Co<1的情况,确定是有震荡的,其他格式结果我还可以再减小时间步长测试下变化,您可以在确保稳定计算的情况增大下时间步长,简单看下foam的中心差分格式结果,有没有类似的不光滑速度等值面,或者破碎的震荡涡,不需要算那么长时间统计平均速度。有改善的话,我就全力研究争取用Openfoam计算。如果太麻烦的话就算啦,还是十分感谢李老师!
关于库朗数可以大于1的原因,我也不甚清楚,看有篇文献说测试平均库朗数到5还能保证计算结果,可能让我没有重视用小时间步计算。 -
Yes. My OpenFOAM results show that oscillation occurs if linear scheme is used. Although this can be explained by theory, you know that. You can try other schemes like
superbee
which is quite sharp to verify if it predicts better results.Now, I terminate my calculation.
-
@李东岳 非常感谢李老师!如果方便的话,您可以把Openfoam的文件打包发给我吗,这个文件中转网站速度还行 https://cowtransfer.com/ 我看能不能改改差分格式,不方便也没关系,我这边摸索得比较慢,目前只能跑跑教程算例啥的。
-
Below is the plot predicted by steadyState solver
simpleFoam
. 10 mins is enough if you have 60/80 cores. You can see that it can be improved. I am not sure if LES can predict better results. Here is the test case, feel free to adjust the settings to see if you can obtain better results. Please also keep me updated. This is an excellent test case to investigate. -
@李东岳 十分感谢李老师!对OPENfoam不太熟,抱歉耽误了这么长时间回复您,现在使用了三种差分格式 1. filteredlinear 过滤系数取0.5 2. filteredlinear 过滤系数取1 3. LUST格式,25%迎风+75%中心
1.
2.
3.速度场wiggle
可以看到中心差分随着过滤强度从0.5-1,耗散性会变大,导致靠近边壁的涡结构被抹平,逐渐向迎风格式的结果过渡,但是都不能解决震荡问题,(过滤掉了一些高频的震荡(碎涡),但是速度场的wiggle依然存在)。LUST格式下没有震荡,虽然只有25%的迎风,但仍然会造成边界层计算不准确,速度梯度不对,结果和二阶迎风类似。目前我认为应该是这个高雷诺数问题(5e5)本身无法用千万量级的网格算边界层,网格雷诺数太高,(我看有人说cell Re <2才能稳定),中心差分耗散太小,导致震荡难以消除。现在准备暂时放弃边界层定量研究了。。。
-
@cccrrryyy 老师您好,用的WALE模型,我看近些年基本用这个的比较多
-
您好,我也用过LES做过边界层,请问这么高的
, 是多少,WALE模型的LES要求近壁面的 几乎为0吧。 -
@心里的小漩涡 在 fluent用LES算水翼边界层选择中心差分引起的震荡如何消除 中说:
有试过二阶中心差分格式中耦合小部分一阶迎风格式吗,比如给个0.02%的一阶迎风格式,之前用fluent算3900的圆柱绕流,用纯二阶中心出现过数值振荡容易发散的情况,改成混合之后收敛性好很多,精度也还行,没有很明显的下降
@李东岳 李老师,这个在OpenFOAM里面能直接设置吗?
-
@xiezhuoyu LUST格式是线性迎风与线性的混合,权重是0.25与0.75,不过可以相对简单的改一下权重
这个帖子挺老了。当时没有服务器来算。有时间最近应该继续算一下。库朗数减小一些,比如0.3或者0.2左右来尝试纯粹的linear。当时完全算不动。
-
-
@李东岳 在 fluent用LES算水翼边界层选择中心差分引起的震荡如何消除 中说:
库朗数减小一些,比如0.3或者0.2左右来尝试纯粹的linear。
这个倒是找到一个很有意思的格式("Blends two schemes based on the local face-based Courant number"),
div(phi,U) Gauss CoBlended 0.01 // Co below which scheme1 is used linear // scheme1 0.05 // Co above which scheme2 is used LUST grad(U); // scheme2
这个格式可以根据Co数自动调整格式,可以实现按区域混合各种格式(虽然不是按权重混合)。
-
@xiezhuoyu Nice! 挺有意思
Description 28 Two-scheme Courant number based blending differencing scheme. 29 30 Similar to localBlended but uses a blending factor computed from the 31 face-based Courant number and the lower and upper Courant number limits 32 supplied: 33 \f[ 34 weight = 1 - max(min((Co - Co1)/(Co2 - Co1), 1), 0) 35 \f] 36 where 37 \vartable 38 Co1 | Courant number below which scheme1 is used 39 Co2 | Courant number above which scheme2 is used 40 \endvartable 41 42 The weight applies to the first scheme and 1-weight to the second scheme. 43 44 Example of the CoBlended scheme specification using LUST for Courant numbers 45 less than 1 and linearUpwind for Courant numbers greater than 10: 46 \verbatim 47 divSchemes 48 { 49 . 50 . 51 div(phi,U) Gauss CoBlended 1 LUST grad(U) 10 linearUpwind grad(U); 52 . 53 . 54 } 55 \endverbatim 56 57 SourceFiles 58 CoBlended.C
发布 45 的 27