What determines the simulation speed in a time domain simulation?
The Bladed integrator uses a variable step Runge-Kutta algorithm. This means that the step-size is changed from one step to the next based on an integration error estimate relative to the integrator tolerance. If there are high frequency or rapidly varying states, the integrator takes short steps; if all the states are low frequency or unchanging then the integrator takes larger steps. If on a particular step the integrator calculates the error of any state to be larger than the integrator tolerance, then the integrator repeats the step with a smaller step-size. This is called a step reduction.
If the calculation outputs option “Software performance” is selected, there are additional output groups in the results of dynamic simulations. In the output group called “Software performance”, it shows the number of step reductions caused by each state. If one state has a particularly large number of step reductions, then it is worth checking whether:
• The frequency of the mode related to this state is high.
• The related mode has been excited due to an instability.
In versions 4.4 and later, there is also a “Simulation step sizes” output group. Here, you can plot simulation time against step number. Each step can be assumed to take a similar amount of calculation time, which means that the steeper the line, the quicker the simulation. If the curve goes particularly flat at one point in the simulation, it’s worth taking a look in the results at what has happened at this time, which may in fact uncover real issues.