Showing results for 
Search instead for 
Did you mean: 

What is wrong with this MPC formulation?


I have a pretty simple model here with the goal of exporting a state space model to MATLAB between an input node and the center of a body. If you look at the included deck, the input is at node 10 (the RBE2 higher in Z) and the output is node 2420 (relative motion between 2419 - the center of the body - and 10, or so I'd think).


If I take the output using node 2419 and input a step force at node 10, I would expect to see a step response where node 2419 moves in a predictable way towards the expected displacement at the step. For instance, put a step of 10000 pounds in, node 10 moves to 1 inch, and node 2419 then moves to 1 inch in time.


I would expect that if node 2420 is my output and is set up to be (2419) - (10), for the same input, it should jump to -1 and then move to 0 over time, as the two nodes grow closer together and then spread apart. However, I do not see this.


What am I doing wrong?


Re: What is wrong with this MPC formulation?

Siemens Phenom Siemens Phenom
Siemens Phenom

A couple of things that I notice that would be considered "not recommended practice", while they may not be the exact source of your issue.

Some guidelines for MPC's:

1) Only connect coincident nodes. If you must connect non coincident nodes then remember geometry effects are NOT included unless YOU include the proper coefficients. (rotation angle x moment arm = translation). If you include rotations and leave out these terms, then your equations can lead to grounding or other "non intuitive" behavior.

2) When using MPC to "calculate" a value like average motion or relative motion, consider using SPOINT instead of GRID, SPOINT is a single value instead of automatically worrying about 6 values for a single GRID and it does not have a physical location to worry about. Eliminates connecting to other elements by mistake since SPOINT is not allowed for element connections.


Have you tried some static cases to make sure you understand how all of these connections are behaving and have you reviewed the mode shapes to make sure things are as expected before exporting the state-space model?


I'll try to look at your model in a litle more detail later.





Re: What is wrong with this MPC formulation?


Alright, I switched over to using SPOINTs for all of the relative motions. It wasn't too hard, so I'll stick with that in the future. I believe that I don't need to worry about the geometry effects, as you mentioned, because it is the relative motion and we use SPOINTs which explicitly don't have any geometry. I attached an updated deck with SPOINTs.


When I've tested using other output nodes that aren't supposed to be giving relative motion, it is very much what I'd expect. I input a step of 10000 pounds to node 10 (RBE2 center). This moves it 1 inch immediately. Then, I see the output node rise to 1 inch displacement in what looks like a first order step response, although I think it is just exhibiting damping. The time constant is near 1/15 seconds, which would fit the expected performance. The CBUSH elements that connect to the mass should have a natural frequency of about 15 Hz when taken together in parallel. The unexpected result is that setting an MPC-constrained SPOINT to be the output shows the exact same response curve.




If I set node 10 to be both input and output, I have a response on the order of 1e-7 which I would attribute to roundoff and numerical errors and accept as correct. A plot of that is below.