Oh, I see what you mean. The parts that "collide" are not actually part of the model. The thing is that I'm using rigid bodies to define constraints, so for instance the cylinder that you see in the middle of the gear is not part of the model, but it rather represents the axial rotation joint that imposes the rotation to the gear. This is the way I'm interfacing the CAD model with the multi body dynamics software. A simpler example is a joint for a rigid pendulum. To define the position and orientation of the rotation axis, you place a cylinder, and the pendulum will rotate around the main axis of the cylinder. Spherical joints can be defined by a sphere, but the sphere won't be part of the physical model, and will only define the joint. Does this make sense? You can see this better in this video:
This is also the way other MBD pre-post processors work.