Do anyone have any ideas? What this can be? How can I fix this?
At least judging from the pictures, that does seem like under-extrusion. Some ideas for further investigating the issue.
The problem may be due to the gcode being wrong. In this case, your printer is merely executing correctly... the wrong commands. To check if this is the case:
- The easiest, but inconclusive way, would be to re-slice a model that fails consistently, with a different slicer. If the second print came out good, than you would know that the problem is with the slicer. This method is inconclusive because you wouldn't not if the gcode is bad or if it simply happens that your printer cannot print well that specific sequence of commands (which may still be emitted by the other slicer under different conditions).
- The more conclusive analysis would be to look at the gcode of a failed print where the fail happens between two geometrical identical layers. This seem to be the case for the print in the picture, btw. You should then compare the gcode of the layer that printed good with that of the layer that printed poorly. If the gcode differs... then you positively know the slicer doesn't work as it should.
The problem may be due to a mechanical fault with the printer. Here the only idea I have to offer is overheating of the steppers and/or their controllers. This may in turn make the extruder servo skip some steps and therefore extrude less filament. If you perform the conclusive test above, you will know if this is the case.
The problem may be due to a firmware bug. This is difficult to investigate, my only suggestion would be: upgrade to the latest and greatest, if you haven't done it already.
The problem may be filament-related. This could be a number of things, but since the problem seems to happen at towards the end of the print, and your are printing at relatively high temp, an idea could be that too much heat reaches the cold end of the extruder, interfering with its extrusion. The easiest test here would simply be to re-print a failed print with a different filament. In your case I would suggest some PLA, just to decrease the temperature and change the chemical composition too.
These are more or less all shots in the dark, but - together with asking here - it would be what I would do to debug, had I the same problem. Keep up posted! :)