Mike Tocci <email@example.com> wrote in message <firstname.lastname@example.org>... > Kelvin Hales wrote: > > I find when solving my present stiff Simulink 5.1 (R13SP1) system > > with ODE15s, that the solution tends to grind to a halt when > > approaching a steady-state, i.e. most annoyingly at the very point > > where it should start to take bigger steps and zoom along. If, > > however, I stop the simulation and re-start it by initialising the > > states with the end state of the previous run, then the solution > > immediately picks-up and shoots along. > > > > Any ideas anyone? > > Its as though the solver is getting bogged down in the detail > > because it cannot forget the previous small time-steaps. > > A problem with a zero-crossing detection, maybe? > > Is there something in the Simulink debugger/profiler that would > > reveal what's happening? > > > > > > Kelvin B. Hales > > Kelvin Hales Associates Limited > > Consulting Control Engineers > > Web: www.khace.com > > Hi Kelvin, > > I'm not 100% sure, but it could be zero-crossings that are causing > this. Whenever you stop a simulation and use the final state as an > initial state, there is some information not stored in the state > vector that is lost. This could be affecting the zc detection in > some blocks. > > I do recommend using the Simulink debugger to try to find the problem. > If you use the command-line debugger (sldebug), there are commands to > print out solver information. I'd try setting a time break point > (tbreak) so you get to the point where the model nears steady state, > and then use the command "strace 4" to print out all the information > in the solver. If there's any information in the output that's not > clear, let me know and I'll see if I can help. There can be alot of > information printed to the screen, so I recommend taking a few time > steps using "step top" to see if you can tell why the time steps are > so small. > > > Hope that helps, > > Mike >
I too have the same issue. I'm simulating a highly dynamic system with lots of power setpoint jumps with ODE15s. The model matches all the extreme dynamic changes fine until a large drop in power after being held at a steady state value. At this large drop in power, the model starts to progress at a snail pace. The step size it progresses at is at about 8e-8 seconds.
I am playing around with the minimum step size, maximum step size, relative error tolerance, and absolute error tolerance, but nothing is getting me past this point.
I am attempting to use a fixed step solver, but am not partial to this because it is so slow.
I will look into the suggestion posted by Mike. Any other suggestions are appreciated.