i****a
发帖数: 88 | 1 这里有没有做Model order reduction的？ 用ANSYS 或 Femlab建模得到系统矩阵，然
后Arnoldi method 得到 Krylov
matrix来project from n to r subspace.网上free的 mor4anays好像找不到了。。。 |
i****a
发帖数: 88 | 2 自己顶一下，已经搞定， 其实stiffness和 damping matrices都可以从ANSYS的 .full文件抽取出来，并不一定要用.emat
文件。
另外，ANSYS建模太麻烦， 今天搞定了用Fluent从Icepak里把model的elements&nodes导入ANSYS来提取系统矩阵。现
在原本Icepak里半个小时transient simulation在matlab里用MOR不到一分钟就得到结果了。坏处是现在每天工作量连15
分钟也没有了。
【在 i****a 的大作中提到】
: 这里有没有做Model order reduction的？ 用ANSYS 或 Femlab建模得到系统矩阵，然
: 后Arnoldi method 得到 Krylov
: matrix来project from n to r subspace.网上free的 mor4anays好像找不到了。。。
|
d**o
发帖数: 79 | 3 Are you working on fluid dynamics? What is the perpose of model order
reduction? In most cases, if we have the stiffness and damping matrices, we
can get the solution directly. |
i****a
发帖数: 88 | 4 MOR usually works on linear system. Assume you have n = 10k nodes in FEM
model, MOR projects the system
matrices into a subspace with order, say ~ 10. So instead of solving a 10k x
10k dimension ODEs in FEM
software, a system of equations with 10 x 10 dimension matrices would be
much faster to solve. Once you get
the solution on these 30 nodes in the subspace, you can project the results
back to the 10k nodes to get the
actual results.
we
【在 d**o 的大作中提到】
: Are you working on fluid dynamics? What is the perpose of model order
: reduction? In most cases, if we have the stiffness and damping matrices, we
: can get the solution directly.
|
d**o
发帖数: 79 | 5 Thanks. Sounds very attractive. I have tried models with 90k nodes. But once
the stiffness matrix is obtained, it takes only a few minutes to solve a
linear equation even with a order of 90k x 90k. So I never heared of MOR.
Also another thing should be noticed is the accuracy. How does the two
methods compare?
x
results
【在 i****a 的大作中提到】
: MOR usually works on linear system. Assume you have n = 10k nodes in FEM
: model, MOR projects the system
: matrices into a subspace with order, say ~ 10. So instead of solving a 10k x
: 10k dimension ODEs in FEM
: software, a system of equations with 10 x 10 dimension matrices would be
: much faster to solve. Once you get
: the solution on these 30 nodes in the subspace, you can project the results
: back to the 10k nodes to get the
: actual results.
:
|
i****a
发帖数: 88 | 6 Hehe, even minutes is so slow in my view as I am not a big fan of simulation
. MOR is a hot topic now - very
promising for fast simulation and real time prediction purposes. As a
comparison, simulation taking minutes
in FEM will only spend ~1s using MOR. Isn't it attractive? The accuracy
depends the order you choose to
approximate the original model in FEM. It tries to match the first several
moments of the system in the
frequency-domain. For typical thermal problem which is a low pass filter by
looking from signal processing
point of view, the error could be less than 0.5% even with 3~5 orders. But
it might not be suitable for high-
spatial frequency problems.
once
【在 d**o 的大作中提到】
: Thanks. Sounds very attractive. I have tried models with 90k nodes. But once
: the stiffness matrix is obtained, it takes only a few minutes to solve a
: linear equation even with a order of 90k x 90k. So I never heared of MOR.
: Also another thing should be noticed is the accuracy. How does the two
: methods compare?
:
: x
: results
|
d**o
发帖数: 79 | 7 不错啊，matlab里容易实现么，有没有reference推荐一下啊？
simulation
several
by
【在 i****a 的大作中提到】
: Hehe, even minutes is so slow in my view as I am not a big fan of simulation
: . MOR is a hot topic now - very
: promising for fast simulation and real time prediction purposes. As a
: comparison, simulation taking minutes
: in FEM will only spend ~1s using MOR. Isn't it attractive? The accuracy
: depends the order you choose to
: approximate the original model in FEM. It tries to match the first several
: moments of the system in the
: frequency-domain. For typical thermal problem which is a low pass filter by
: looking from signal processing
|
i****a
发帖数: 88 | 8 http://modelreduction.com/
【在 d**o 的大作中提到】
: 不错啊，matlab里容易实现么，有没有reference推荐一下啊？
:
: simulation
: several
: by
|
d**o
发帖数: 79 | |
p*******r
发帖数: 2 | 10 楼主还在吗?我现在也求一个MOR4ANSYS的CODE啊。。。好心人有吗。。。谢谢了 |
|
|
y*h
发帖数: 2121 | 11 同求这个code.
还有ansys .full文件不需要计算么？
一直是在Ansys里建模计算，想了解下LZ用的这方法是不是把线性方程组导入到matlab
计算啊？
有什么优缺点么？谢谢。
full文件抽取出来，并不一定要用.emat
nodes导入ANSYS来提取系统矩阵。现
结果了。坏处是现在每天工作量连15
【在 i****a 的大作中提到】
: 自己顶一下，已经搞定， 其实stiffness和 damping matrices都可以从ANSYS的 .full文件抽取出来，并不一定要用.emat
: 文件。
: 另外，ANSYS建模太麻烦， 今天搞定了用Fluent从Icepak里把model的elements&nodes导入ANSYS来提取系统矩阵。现
: 在原本Icepak里半个小时transient simulation在matlab里用MOR不到一分钟就得到结果了。坏处是现在每天工作量连15
: 分钟也没有了。
|
i****a
发帖数: 88 | 12 这里有没有做Model order reduction的？ 用ANSYS 或 Femlab建模得到系统矩阵，然
后Arnoldi method 得到 Krylov
matrix来project from n to r subspace.网上free的 mor4anays好像找不到了。。。 |
i****a
发帖数: 88 | 13 自己顶一下，已经搞定， 其实stiffness和 damping matrices都可以从ANSYS的 .full文件抽取出来，并不一定要用.emat
文件。
另外，ANSYS建模太麻烦， 今天搞定了用Fluent从Icepak里把model的elements&nodes导入ANSYS来提取系统矩阵。现
在原本Icepak里半个小时transient simulation在matlab里用MOR不到一分钟就得到结果了。坏处是现在每天工作量连15
分钟也没有了。
【在 i****a 的大作中提到】
: 这里有没有做Model order reduction的？ 用ANSYS 或 Femlab建模得到系统矩阵，然
: 后Arnoldi method 得到 Krylov
: matrix来project from n to r subspace.网上free的 mor4anays好像找不到了。。。
|
d**o
发帖数: 79 | 14 Are you working on fluid dynamics? What is the perpose of model order
reduction? In most cases, if we have the stiffness and damping matrices, we
can get the solution directly. |
i****a
发帖数: 88 | 15 MOR usually works on linear system. Assume you have n = 10k nodes in FEM
model, MOR projects the system
matrices into a subspace with order, say ~ 10. So instead of solving a 10k x
10k dimension ODEs in FEM
software, a system of equations with 10 x 10 dimension matrices would be
much faster to solve. Once you get
the solution on these 30 nodes in the subspace, you can project the results
back to the 10k nodes to get the
actual results.
we
【在 d**o 的大作中提到】
: Are you working on fluid dynamics? What is the perpose of model order
: reduction? In most cases, if we have the stiffness and damping matrices, we
: can get the solution directly.
|
d**o
发帖数: 79 | 16 Thanks. Sounds very attractive. I have tried models with 90k nodes. But once
the stiffness matrix is obtained, it takes only a few minutes to solve a
linear equation even with a order of 90k x 90k. So I never heared of MOR.
Also another thing should be noticed is the accuracy. How does the two
methods compare?
x
results
【在 i****a 的大作中提到】
: MOR usually works on linear system. Assume you have n = 10k nodes in FEM
: model, MOR projects the system
: matrices into a subspace with order, say ~ 10. So instead of solving a 10k x
: 10k dimension ODEs in FEM
: software, a system of equations with 10 x 10 dimension matrices would be
: much faster to solve. Once you get
: the solution on these 30 nodes in the subspace, you can project the results
: back to the 10k nodes to get the
: actual results.
:
|
i****a
发帖数: 88 | 17 Hehe, even minutes is so slow in my view as I am not a big fan of simulation
. MOR is a hot topic now - very
promising for fast simulation and real time prediction purposes. As a
comparison, simulation taking minutes
in FEM will only spend ~1s using MOR. Isn't it attractive? The accuracy
depends the order you choose to
approximate the original model in FEM. It tries to match the first several
moments of the system in the
frequency-domain. For typical thermal problem which is a low pass filter by
looking from signal processing
point of view, the error could be less than 0.5% even with 3~5 orders. But
it might not be suitable for high-
spatial frequency problems.
once
【在 d**o 的大作中提到】
: Thanks. Sounds very attractive. I have tried models with 90k nodes. But once
: the stiffness matrix is obtained, it takes only a few minutes to solve a
: linear equation even with a order of 90k x 90k. So I never heared of MOR.
: Also another thing should be noticed is the accuracy. How does the two
: methods compare?
:
: x
: results
|
d**o
发帖数: 79 | 18 不错啊，matlab里容易实现么，有没有reference推荐一下啊？
simulation
several
by
【在 i****a 的大作中提到】
: Hehe, even minutes is so slow in my view as I am not a big fan of simulation
: . MOR is a hot topic now - very
: promising for fast simulation and real time prediction purposes. As a
: comparison, simulation taking minutes
: in FEM will only spend ~1s using MOR. Isn't it attractive? The accuracy
: depends the order you choose to
: approximate the original model in FEM. It tries to match the first several
: moments of the system in the
: frequency-domain. For typical thermal problem which is a low pass filter by
: looking from signal processing
|
i****a
发帖数: 88 | 19 http://modelreduction.com/
【在 d**o 的大作中提到】
: 不错啊，matlab里容易实现么，有没有reference推荐一下啊？
:
: simulation
: several
: by
|
d**o
发帖数: 79 | |
|
|
p*******r
发帖数: 2 | 21 楼主还在吗?我现在也求一个MOR4ANSYS的CODE啊。。。好心人有吗。。。谢谢了 |
y*h
发帖数: 2121 | 22 同求这个code.
还有ansys .full文件不需要计算么？
一直是在Ansys里建模计算，想了解下LZ用的这方法是不是把线性方程组导入到matlab
计算啊？
有什么优缺点么？谢谢。
full文件抽取出来，并不一定要用.emat
nodes导入ANSYS来提取系统矩阵。现
结果了。坏处是现在每天工作量连15
【在 i****a 的大作中提到】
: 自己顶一下，已经搞定， 其实stiffness和 damping matrices都可以从ANSYS的 .full文件抽取出来，并不一定要用.emat
: 文件。
: 另外，ANSYS建模太麻烦， 今天搞定了用Fluent从Icepak里把model的elements&nodes导入ANSYS来提取系统矩阵。现
: 在原本Icepak里半个小时transient simulation在matlab里用MOR不到一分钟就得到结果了。坏处是现在每天工作量连15
: 分钟也没有了。
|
i****a
发帖数: 88 | 23 其实是把线性方程组导入ANSYS计算。1Mx1M的矩阵Matlab根本解不动, 一些开源的
linear system solver,例如MUMPS, 速度比ANSYS慢一倍以上。感兴趣的可以看一下
ANSYS APDL MATH.
matlab
【在 y*h 的大作中提到】
: 同求这个code.
: 还有ansys .full文件不需要计算么？
: 一直是在Ansys里建模计算，想了解下LZ用的这方法是不是把线性方程组导入到matlab
: 计算啊？
: 有什么优缺点么？谢谢。
:
: full文件抽取出来，并不一定要用.emat
: nodes导入ANSYS来提取系统矩阵。现
: 结果了。坏处是现在每天工作量连15
|
y*h
发帖数: 2121 | 24 在ANSYS里建模然后求解不是本来就用的Ansys解的线性方程组么？
【在 i****a 的大作中提到】
: 其实是把线性方程组导入ANSYS计算。1Mx1M的矩阵Matlab根本解不动, 一些开源的
: linear system solver,例如MUMPS, 速度比ANSYS慢一倍以上。感兴趣的可以看一下
: ANSYS APDL MATH.
:
: matlab
|
i****a
发帖数: 88 | 25 其实是把线性方程组导入ANSYS计算。1Mx1M的矩阵Matlab根本解不动, 一些开源的
linear system solver,例如MUMPS, 速度比ANSYS慢一倍以上。感兴趣的可以看一下
ANSYS APDL MATH.
matlab
【在 y*h 的大作中提到】
: 同求这个code.
: 还有ansys .full文件不需要计算么？
: 一直是在Ansys里建模计算，想了解下LZ用的这方法是不是把线性方程组导入到matlab
: 计算啊？
: 有什么优缺点么？谢谢。
:
: full文件抽取出来，并不一定要用.emat
: nodes导入ANSYS来提取系统矩阵。现
: 结果了。坏处是现在每天工作量连15
|
y*h
发帖数: 2121 | 26 在ANSYS里建模然后求解不是本来就用的Ansys解的线性方程组么？
【在 i****a 的大作中提到】
: 其实是把线性方程组导入ANSYS计算。1Mx1M的矩阵Matlab根本解不动, 一些开源的
: linear system solver,例如MUMPS, 速度比ANSYS慢一倍以上。感兴趣的可以看一下
: ANSYS APDL MATH.
:
: matlab
|
l*****b
发帖数: 1 | 27 我要做的是在ansys提取机床大件的质量、刚度矩阵，
在matlab中对大件分别进行降阶处理，
再通过多点约束（MPC）综合到一起，
用缩减后的模型进行瞬态动力学分析，谐响应分析。
网上看有人用MOR4ANSYS实现了，
不过现在这个程序收费 5000EUR
您用您的方法看能行的通吗？
谢谢 |
