I recently spoke with someone from a company that does finite element analysis on large steam turbines for the military.
They ran a simulation in CFX, using over 30 million nodes. Each node needed calculations for X, Y, Z, and temperature. These are key for structural or fluid dynamic simulations.
Without more details, we need to justify the need for a simulation of this complexity as the costs will be high. Since this was for a military project, funding wasn’t as restricted as it would be in private companies.
A simulation with 30 million nodes needs lots of RAM and microprocessors to handle it. Heavy parallel processing and a lot of RAM, likely well over 100 GB, is essential.
Mechanical hard drives access data much slower than RAM. Accessing data from RAM takes about 0.00000001 seconds, which is 1.3 million times faster than a mechanical drive.
Scratch Files
Running out of RAM is a situation to avoid. If that happens, the program has to store information on the hard drive and retrieve it many times. It’s easy to tell if your program runs out of RAM because ANSYS creates scratch files.
Solid-state drives can help supplement RAM. Their read-write speeds are about five times faster than mechanical drives. I designed my analysis computer with solid-state drives and a large amount of expandable RAM.
In the “old days,” solid-state drives didn’t exist, and computers had a maximum of 8 GB of RAM. Long run times were common, even for moderate tasks. Runs lasting several days or even weeks were not unusual. Any error that forced an analyst to discard results after such a long wait was frustrating and wasteful.
Simplifying the Model Using Symmetry
Adding symmetry features to symmetrical models can lead to a substantial reduction in the model’s size. This can cut the number of elements down to 1/4 or even 1/8 of the full model.
You can simplify 3-D models, even if they aren’t symmetrical, by reducing the number of parts you look at.
Break Up the Problem into Smaller Pieces
If you can’t reduce the model, you can achieve a reduction in execution time by running the analysis in segments. You can use the output from one part as input for another, lowering memory needs and total run time.
Model Simplification – Reducing the Number of Parts
A recent computer model I received had over 7,000 parts. I would never tackle an FEA problem of that complexity. Luckily, my customer and I focused on the design parts that needed analysis and agreed on a simpler model.
Supplement with Classical Methods
I recently helped a customer study a small industrial vehicle. This vehicle can carry several thousand pounds. The issue was the consequences of the vehicle coming to an abrupt stop, such as colliding with a wall or barrier. At first, I considered the impact analysis. The main concern was the containment structure. It could be a concrete wall, a steel barrier, or something else. If damage to this structure was a risk, we considered the effect of adding bumpers to the vehicle. This simplified the analysis. I still expected long calculation times due to the nonlinear impact analysis. I used basic equations for linear analysis. They include force, mass, acceleration, distance, and velocity. Solution times took only a few minutes, which pleased the customer, along with my billing. I’ll cover this topic in more detail in a future blog post.
Supplement Analysis with Test Data
For new designs, using test data from earlier versions can help. Calibrate the simulation. Check the FEA model output against test data from the current assembly. This method checks the analysis and helps build a precise model of the new design.
Norman Neher
Analytical Engineering Services, inc
Elko New Market, MN
www.aesmn.org