If something momentarily disrupts the generator shaft rotation, the resulting voltage dips would be better represented by a graph like this...
View attachment 21015
...than by a graph like this.
View attachment 21016
In post 946, jaggedben raises a very important point.
I (and others) describe a phase shift as meaning a shift in time, but for many real physical systems things like transients hit all the phases at once, without a time shift.
This points to an error in how I am thinking about phase angle. Phase angle is _associated_ with a shift in time. It is a way of measuring the time offset between the zero crossings of a periodic signal. But phase angle does not mean that the underlying mechanism involves a time shift.
Clearly in a generator, the various phases are physically offset on the stator, and the phase angle is generated by the rotor poles passing at different times. But this is not the only way you could generate 3 phases; you could have an inverter where the different phases are produced by a table lookup by a microprocessor; everything happening at the same time but with an output that has a time offset in the zero crossings.
If, as jaggedben's diagram suggests, you wiggled the excitation control of a synchronous generator, you would modulate all of the phases at the same time. (Though I don't think that each phase would see the same modulation simply scaled, as suggested by jaggedben's first graph.)
I am sure that with another mechanism you could actually get a transient to 'roll' from one phase to the next.
IMHO this goes to the root of this whole discussion, distinguishing between the measurable phase angles in a system and the underlying mechanism that produces these phase angles.
Phase angle applies to single frequency sine waves and to periodic functions. Phase angle is _associated_ with a time shift, and _looks_ like a time shift, but could be created by an entirely different underlying mechanism. Phase angle must be understood simply as one of the parameters describing a sine wave, nothing more.
We can use phase angle to analyze transients and non-periodic functions, but only by first breaking the signal as a sum of single frequency sine waves (someone already mentioned Fourier analysis). So we can use phase angle to describe _each_ of the frequencies that make up a complex or transient signal, and _each_ of those frequencies will in general have a _different_ phase angle.
Back to our 'single phase center tapped transformer', if we use the neutral point as the reference, than the simple inversion is the equivalent of a 180 degree phase shift for each and every frequency going through that transformer.
If the mechanism of the transformer involved a time delay, then the phase shift would be _different_ for each different frequency going through the transformer. Such systems are possible, and have uses, but not for power distribution at the local level.
With regard to the question brought up of a pulse of 4 cycles of a sine wave, suggesting that inversion would give one result and a phase shift a different result. I believe that this is a terminology issue.
If you believe that a 180 degree phase shift necessarily requires a time delay, then there would be a 1/2 cycle delay difference between inversion and phase shift. If, as I now believe, that 'phase shift' simply describes sine waves, then you have to realize that a _pulse_ is not a single frequency sine wave, but is rather composed of lots of different frequencies. An inversion will result in a 180 degree phase shift for each of the component frequencies.
-Jon