AJennings
Aerospace
- Jan 13, 2012
- 1
Hi everyone, this is my first post, so if I messed up on some form rules, just let me know.
I’m trying to solve for higher frequencies of a structure. The problem involves plates attached to a complicated base. From vibration testing, the bending modes were measured and I’m trying to measure them in the model for validation. I was given a model to use, so I don’t have a lot of flexibility with it.
The model has many nodes (10k’s across the plate, about 1M nodes), so I shouldn’t be getting to high frequencies relative to the total number. But these modes are going to be at least a hundred modes shapes in due to intra-plate motion and back structure, so I don’t want to solve for the first 300 modes (file size would also be a problem due to the size of the model, about 0.1GB per mode).
Right now I’m running NX Nastran generated by FEMAP for Normal Modes/Eigenvalues. I’ve been using Lanczos with lower and upper frequencies and number of modes to solve for. The solution is run on a linux cluster. All systems are 64bit. Sometimes the solution will be just wrong and re-exporting and running the analysis would fix it.
Are there better ways of getting these high frequencies? Such as other methods that offer better accuracy/efficiency (I’m doing a run on the modified Householder now)? The base interaction with the plate is assumed to be important, so doing the plate alone doesn’t seem to be an option.
Thanks a bunch!
I’m trying to solve for higher frequencies of a structure. The problem involves plates attached to a complicated base. From vibration testing, the bending modes were measured and I’m trying to measure them in the model for validation. I was given a model to use, so I don’t have a lot of flexibility with it.
The model has many nodes (10k’s across the plate, about 1M nodes), so I shouldn’t be getting to high frequencies relative to the total number. But these modes are going to be at least a hundred modes shapes in due to intra-plate motion and back structure, so I don’t want to solve for the first 300 modes (file size would also be a problem due to the size of the model, about 0.1GB per mode).
Right now I’m running NX Nastran generated by FEMAP for Normal Modes/Eigenvalues. I’ve been using Lanczos with lower and upper frequencies and number of modes to solve for. The solution is run on a linux cluster. All systems are 64bit. Sometimes the solution will be just wrong and re-exporting and running the analysis would fix it.
Are there better ways of getting these high frequencies? Such as other methods that offer better accuracy/efficiency (I’m doing a run on the modified Householder now)? The base interaction with the plate is assumed to be important, so doing the plate alone doesn’t seem to be an option.
Thanks a bunch!