Reference data should be properly sampled to avoid aliasing issues

[casella]

Consider the negative regression report of OMC on the Buildings.Fluid.HydronicConfigurations.ActiveNetworks.Examples.InjectionTwoWayVariableReturn.diff.con.T2Sup.T variable.

The transient lasts one day and is sampled every 200 s, which in general seems a reasonable value to avoid generating too large simulation result files. However, around time = 60000, the temperature undergoes a relatively fast and poorly damped oscillation

which causes the verification test to fail because the results computed by OMC get (slighty) out of the tolerance tubes.

Result checkers such as CSV-compare usually have some tolerance both on the time and y axes, to accommodate for small mismatches in delays. This could be enough to accommodate the result of OMC's simulation if the reference tube was properly computed, i.e. as a damped sinusoidal oscillation. It is clear that a sampling interval of 200 s is too crude for this purpose, so the edgy tubes computed as a result are unnecessarily causing the verification to fail.

This test case may actually motivate the need for a different Interval for regular simulations and for testing simulations. In the first case, the 3-minutes interval is probably good enough for practical purposes, at the end of the day nobody cares about seeing those oscillations in great detail. However, for comparison purposes, one should probably take something like 20-40 s. Of course we could just set Interval = 40 in the experiment annotation, but this would unnecessarily cause much larger simulation results. If we had a custom annotation for verification tests only, that would be the right place to put this. BTW, reference simulations do not save all the variables, but typically only a few selected ones, so even if Interval is much shorter, that won't be a problem.

See discussion in modelica/ModelicaSpecification#3473