I'm quite new in multibody modeling and I spent the most of time in debugging my model, trying to simplify as much as possibile the model. Is there any systematic way or best practices to debug big multibody models?
To be honest I tend to start with simple models, that work (or are at least easy to debug), and then add complexity. You might also want to break the system up into subsystems and build a rig to test each subsystem (that is almost the same as the first sentence, depending on your definition of rig).
As an example with a car I'd start by making sure each suspension works on a K&C test before assembling the suspension to a body and running full vehicle tests with tires. Then I'd add the powertrain and so on. Then I'd replace the smooth road with a 3d road surface. At some point I might need aero, so I'd start with still air, and then add a crosswind. Then I'd replace linear elastic engine mounts with hydraulic ones. And so on and so forth.
Also a modal analysis is a very useful check on your model.
Cheers
Greg Locock
New here? Try reading these, they might help FAQ731-376
It's been a few years, but I have to agree with Greg. Subsystems behaving in isolation (or in a simple test harness) is the first goal. Then connecting/merging them is the next. My example is building a complete working engine/transmission model (bit by bit). Then gradually moving it into place into a vehicle model, finally connecting the shafts. Building a big model in one go is a no-no.
When my vehicle was first connected up, it misbehaved really badly, but an animation of the first fraction of a second showed the problem: 180deg initial twist on the drive shafts - the car wanted to wheelie off backwards.
Also start with a stationary system, setting initial velocities is a complete pain. Also allow time for the system to settle into a stable configuration before driving off. I think ADAMS/Car starts with the vehicle somewhat above ground level with zero g and then turns g up and so it drops to the ground, that is a handy way of assembling the model and making sure your tire forces aren't inside out.
Cheers
Greg Locock
New here? Try reading these, they might help FAQ731-376
Initial conditions can take some sneaky treatment. I do like the approach of gradually applying gravity.
In my big model, the engine was installed at a small angle to the global vertical (15º). That was too much initial twist when multiplied through all the gears and shafts, which were all started at zero to the global vertical. So I installed the engine system in a "cradle", which gradually rotated it back by 15º while the starter motor was starting the engine. Happy days!
I use to do almost the same and also give insight the model, but it's very time consuming and I admit that I'm non that fast in modeling. Do you use the design of experiment for the debugging?
No. I use DoE for making small changes to the model. Perhaps what you call debugging is what I would call correlating though. Once I have a working model that more or less agrees with the test data then the interesting part of the job is over.
For instance I did use a DOE to get from blue curve to the green one in this case, trying to get better agreement to the red line which is the real vehicle with the same SWA time history.
Cheers
Greg Locock
New here? Try reading these, they might help FAQ731-376
thanks for your reply. For debugging I mean the analysis of checking mistakes inside the model. For improving the correlation with measured result I use the optimization and sometimes I use the DOE just to find a good starting point for the optimization.