With some experimentation I believe the constraint is not the issue, but simply the size of the model is too large for the current symbolic linearization. To check this I tried to linearize the SuspensionBeamEnergy model but with two changes:

1) setting world.enableAnimation to false in order to reduce the number of variables

2) replacing both flexureLower1 and flexureUpper1 with just direct connections between the frames they connect to

3) with the components above removed, I also replaced the equation for energyPE with

energyPE = 0;

in order to keep the system balanced.

The symbolic linearization made it a little farther beyond but unfortunately still failed.