FAQ S01 - How do I use a fixed step integrator in Simcenter Amesim?

fixed step integrator

Category: Simcenter Amesim

Topic: Simulation

Access Level: End User

Publish Date: June 2008

Question

How do I use a fixed step integrator in Simcenter Amesim?

Since LMS Amesim 4.2, users have the possibility of using a fixed step integrator. Why was this facility provided? It is provided as an intermediate step towards running in a real time environment. The stages would be as follows:

• Start from a high fidelity model.

• Examine the eigenvalues.

• Progressively reduce the model by applying the following model reduction techniques:

State count

Eigenvalue analysis

Modal shapes

Activity index

• Reduce the number of state variables by removing extreme eigenvalues. As you do this compare the reduced model results with the high fidelity model results.

• When all extreme eigenvalues are gone, try running with a fixed step integrator. Set the fixed step to the time step you will be using on the real time platform.

When you have a low fidelity model which produces the same basis results as the high fidelity model and runs with the fixed step integrator in a reasonable time, you are ready to download for running in a real time environment.

Answer

My injection system with 1000 state variables will not run with a fixed step integrator. Why is this?

A fixed step integrator is totally inappropriate for a system like this. Use the standard integrator.

My system needs a step size of 1.0e-8 to run with fixed step Euler. Why is this?

Your system must have extreme eigenvalues. Use the standard integrator or if you are preparing the system to run in a real time environment, use model reduction techniques.

I get NaNs and Infs, or physically unrealistic values of 1.0e30 bar when I run with the fixed step integrator. Why?

You are operating outside of the stability region of the method. Why are you using a fixed step integrator? If you are doing general simulation, always use the standard integration method. If you are preparing a model for real time, you must reduce the step or more likely do some more model reduction.

I have no access to the fixed step integrators for my system. Why?

It is an implicit system and the simple fixed step methods provided are totally unsuitable for implicit systems.

I request linear analysis and it does not happen when I use a fixed step integrator. Why?

The whole object of running with the fixed step integrator is to do only the minimum of calculations. Hence LA is not supported. If you want to do LA, switch back to the standard integrator.

I use the RSTAT block and run with a fixed step integrator. Why are most of variables not set?

In the interest of speed, only those variables that are potentially useful in preparing a model for real time are computed.

The state count facility does not work for a run with a fixed step integrator. Why?

It takes CPU time to provide this facility. In the interest of speed it is not provided for fixed step integrators.