The State contributions tool

Launch a simulation, open the Performance Analyzer and check the Run statistics pane.

Figure 74: Performance Analyzer Run statistics

Look at the two plots. The upper plot shows the CPU time which has a linear trend and then a parabolic one. Use the cursor to identify the time at which the change occurs. The value is around 3.5 seconds.

Now, look at the Frequencies. The highest frequency for the oscillating modes at the initial time is about 3500 Hz, its damping is also quite low.

Figure 75: Performance Analyzer Frequencies

We are also interested in identifying the variables that are causing the slowing down. Open the State contributions pane. Sort the variables by the Controlled column and double-click on the highest one.

The most controlled variable is the velocity of the mass. Next is the pressure at port 2 in the actuator.

Figure 76: Performance Analyzer velocity state contribution

Figure 77: Performance Analyzer pressure state contribution

Now deselect all the variables by using the Select/Deselect all checkbox at the bottom of the variable list and check the two previous variables again.

Figure 78: State contributions of the two most controlled variables

The Cumulative contribution of these two variables remains level and equal to zero for a long time and it starts increasing at the same time as the trend of the CPU time curve is changing. It seems that this time has a particular meaning in the system.

The Instantaneous contribution is also a good indicator of the sudden activity of these variables after 3.5 seconds. This plot becomes very useful for large systems where many different variables can become active at different times of the simulation.

Plot the displacement of the 1 kg mass. We see that the characteristic time occurs when the mass stops at the actuator end stop. Nevertheless, it seems that there are no big oscillations on the displacement that could explain the slowing down.

Figure 79: 1 kg mass displacement

Now plot the velocity of the same mass which is the most controlled variable.

Figure 80: 1 kg mass velocity

The slowing down of the solver can now be explained. When the contact occurs, velocity starts oscillating very strongly. Plot the pressure at port 2 of the actuator.

Figure 81: Pressure at port 2 of the actuator

In this case the pressure oscillates a lot when the contact occurs and, since the pressure goes below 0 bar, cavitation occurs (and thus the calculations are harder). A more in depth study including a linear analysis could be interesting to see the oscillating modes at the time when contact occurs.

These observations tend to indicate that the contact between the mass and the actuator is badly configured. Go to Parameter mode and check for the parameters of the actuator.

Figure 82: Parameters of the actuator

The damping coefficient on endstops is set to zero, this is the reason for the numerical problems.

Here the damping coefficient (c) must be computed using this equation:

where:

• c is the damping coefficient in N/(m/s),

• z is the damping ratio,

• k is the spring rate of the endstop (1.109 N/m),

• M is the mass (1 kg).

We will fix the damping ration to 10 % because this value represents best the physical phenomenon.

This is why the value of the new damping coefficient is set in the experiment named "fast simulation". To access to the experiment, switch to Simulation mode and to select this experiment. Make sure that Experiment view is activated in the View > Show/hide menu.

Figure 83: Loading the new experiment

Figure 84: New parameters of the actuator

Now launch a new run and open the Performance Analyzer.

Figure 85: Performance Analyzer after modification

The new model is solved in 6 seconds instead of the 18 seconds before the modification, that is an impressive 300 % speed-up ratio.

Plot the velocity of the mass and the pressure at port 2 in the actuator.

Figure 86: Velocity and pressure before and after modification

Red curves: after modification Green curves: before modification

We can observe many fewer oscillations after the modification of the damping coefficient has been done.

This example illustrates again the importance of the parameterization of submodels and specifically the sizing of damping parameters which is often critical in terms of performance.