In the photo of mbouwer it can be seen that the armature has 20 poles and that the stator has 24 poles. So the armature pole angle is 360 / 20 = 18° and the stator pole angle is 360 / 24 = 15°. So the difference in pole angle is 3°. Assume that the armature has a preference position if an armature pole is just opposite to a stator pole. This changes every 3° and so the armature should have 360 / 3 = 120 preference positions per revolution. However, in practice it appears that it has only 20 preference positions per revolution and the peak on the cogging torque is therefore much higher than what if should be theoretically. I found the following reason for this phenomenon.
The flat magnets are glued in the armature at a certain distance of each other. There are no machined flat grooves in the steel armature ring which position the magnets at exactly the correct place. So it might be that the magnets arn't positioned exactly at an angle of 18°. It might also be that the thickness of the glue layer isn't the same for all magnets. This can make that there is one of six preference postions for which the magnetic flux flows easier from armature to stator than for the other five preference positions. This makes one of six preference positions stronger than the other five. If you rotate the armature, you only feel the strongest preference positions. I have tried to find the five less stronger preference positions in between the strongest one and they can be felt. So inaccurate manufacture is part of the problem.