As I make component changes in animation they reflect through all previous scenes.
A scene is a view state; it is not a separate model. There is only one model, and the same model, therefore, shows in all scenes (except to the extent that you artfully arrange a scene not to show everything). Still, the underlying geometry is the same in all scenes, and if you chop something in two in Scene 1 it will also be chopped in two in Scene 2.
The trick is to show different objects in each scene, so that, obviously, if you chop something in two in Scene 1, a completely different object in Scene 2 will be unaffected. Don’t use instances of components or you will defeat your own subterfuge.
So each scene is an arrangement of separate objects (mostly a bunch of copies) representing a frame in your animation. You can either string the scenes out along a line stretching to the horizon, so that the frame transitions would be the camera whizzing to each successive “stage setting,” or, you can superimpose all the elements of the animation in a single location, and the frame transition would be each successive scene switching certain layers on and others off to simulate motion.