Why does the Point Mass feature in Workbench (WB) Simulation require so much memory?
|
This is due to the use of a consistent-mass matrix formulation. A remote mass in WB is simply a set of contraint equations (CEs) in ANSYS. When this remote mass is connected to many nodes (i.e., the number of terms in the remote mass CEs is very, very large), the memory requirements increase significantly. Using a consistent-mass matrix formulation causes the mass matrix to be very dense, and it is because of this dense matrix that the solution requires large amounts of memory. It becomes dense because the remote mass (or CE) requires a connection between many degrees of freedom (DOF). Each nonzero value in the matrix represents a connection between one DOF (row) and another DOF (col), so a CE that connects many, many DOFs together causes a lot of extra nonzeroes in the matrix. The factorization size is usually about 10x the matrix size, so when the matrix size is significantly increased, the size of the FULL file (where the matrix is stored) increases, and the sparse-solver factorization size increases. Basically, everything that uses the mass matrix downstream of the mass matrix assembly will require considerably more memory. A workaround is to use lumped-mass approximation (LUMPM,ON in a Commands Object) which avoids the large amounts of memory required for consistent mass because the off-diagonal terms are no longer stored for the mass matrix. Lumped-mass approximation uses a diagonal-matrix representation, only a single vector (i.e., the diagonal) needs to be stored. This can drastically reduce the memory requirements. The remote mass can still be applied (i.e., there is still a CE that ties many terms together); however, as the CEs are applied, any off-diagonal terms can be dropped, which maintains the diagonal matrix representation. Note that as these off-diagonal terms are dropped, some mass is lost, so a vibration analysis will typically yield higher frequencies. Usually, this is a small difference; however, when the remote mass is connected to many, many nodes and/or when the remote mass is a high percentage of the overall mass, then dropping these off-diagonal terms has a bigger impact on the final result. Please see Section 13.5.2 of the Theory Reference for some additional information. Also, please note that in v12, WB should be warning users about this high-memory usage when remote mass is used with consistent mass. This same effect can occur with other "remote" features in WB like remote displacement, remote loading, etc., although these items usually involve the stiffness matrix, rather than the mass matrix. These other remote features will also cause warnings in WB at v12. Since there is no diagonal-matrix representation of the stiffness matrix, there is no workaround other than to limit the scope of the remote feature (i.e., avoid tying the remote feature to a large number of nodes). |
||
|