I would start by reducing the feed from 1800 to 900 and see if the path is closer.  Also if you are using servos what error limit do you have set, if its too big the path can be off but in tolerance.

The thing I do not understand is, that it does not happen every time at the same place. Its random...
Could really the feed be an issue.
What do you mean with error limit? It shows does and shows a complete different path than the one forcasted.

Hi,
servos have two parameters, depending on the manufacturer, called the Following Error Window and the Zero Error Window.

Taking the most important of the two, the Following Error Window.....it means that you specify/program an interval in so many encoder counts
and should the commanded position and the actual position deviate by more the the Following Error then the servo drive will fault Following Error.

As an example I have an older 1.8kW AC Allen Bradley servo, it has an 8000 count per rev encoder. The default Following Error window is 20 counts.
So provided the servo can keep to within 20 counts (54 arc min) then it is deemed that the servo is adequately following the desired toolpath.
You could program the Following Error window to be much smaller, say 4 counts. Now what happens if the servo lags a commanded move by as little as
10.8 arc min then it will fault. If you have a slowly varying toolpath you might get away with such a tight Following Error window but if your
toolpath has a lot of high acceleration moves then it will nuisance trip all the time.

Ideally you would set the Following Error to a very low value so that you can be assured that the servo is accurately following the commanded toolpath,
but there is a balance to be found.

When a servo comes from the factory it is common for the Following Error window to be set very wide, maybe up to half a servo turn or several thousand
encoder counts. This will allow you to tune the servo without it faulting Following Error all the time. It is anticipated that you will reduce the Following Error
window when your done. Many first time servos users don't even know that its there and so you could have the situation where the servo deviates from
the commanded toolpath by as much as half a turn and still not fault out!

The second parameter, Zero Error Window, is specified in a number of encoder counts. Under normal circumstances a servo drive will apply current to
the servo motor to reduce the error between the commanded position (from Mach/controller) and the actual position as measured by the encoder.
If the motor overshoots the drive will apply reverse current so the motor backs up a bit. When it gets very close to the commanded position the
motor will go one step forward, realize its overshot and so go one step backward, only now to realize its overshot in that direction ......so it goes one step
forward.....and so on. Its called 'servo dither' and can cause buzzing/vibration and keeps the servo from cooling.

As a consequence most servo manufacturers have a Zero Error window so that if the commanded position an the actual position are within a few counts the
servo stop trying to reduce the error anymore, its considered good enough to perfect. Usually the manufacturers default value is adequate, it is still advisable
to consider the value applied to see whether it suits your machine.

Following Error Window is a most important but often overlooked parameter, and it is so important it can't just be left to chance, you must set it.

Craig