Crash in standalone windows EXCEPTION_ACCESS_VIOLATION after LoadMap Ok

I have a mainly UI widget application, using version 4.15.3-3450819+++UE4+Release-4.15 and designed in blueprints with a minimal C++ functions added. UI include fade effects, blur and movie reproduction in some levels. My team were working ok from editor, but the last 3 weeks we do considerable adding to de proyect and didn’t tested from packaged. When we did it, the packged aplicacion started to sometimes crash without aplarentry reason and in diferent places, always with this error in log:

[2018.02.14-23.17.25:664][441]LogLoad: Took 0.447859 seconds to LoadMap(/Game/1821/Maps/Levels/L_PoliticalDialog)
[2018.02.14-23.17.25:772][441]LogWindows:Error: === Critical error: ===
[2018.02.14-23.17.25:772][441]LogWindows:Error: 
[2018.02.14-23.17.25:772][441]LogWindows:Error: Fatal error!
[2018.02.14-23.17.25:772][441]LogWindows:Error: 
[2018.02.14-23.17.25:772][441]LogWindows:Error: Unhandled Exception: EXCEPTION_ACCESS_VIOLATION reading address 0xffffffff
[2018.02.14-23.17.25:772][441]LogWindows:Error: 
[2018.02.14-23.17.25:772][441]LogWindows:Error: Vdj1821_Unreal4153.exe!0x000000003FD7927D
[2018.02.14-23.17.25:772][441]LogWindows:Error: Vdj1821_Unreal4153.exe!0x0000000041BA320D
...

Error happens usualy from 3th level ahead, usualy in the second movie play, but not always. When try removing parts of the levels, bluprints or actors, we only changed from rate of the crash (sometimes 9 from 10, others 1 from 10) or the place (somethimes starting at our L_map level, others our L_cinematic level) , always with the same signature (LogLoad: Took … === Critical error === Unhandle Exception: EXCEPTION_ACCESS_VIOLATION reading address). When executing the exe in other PC, same error ocurred.

We try removing objects, removing movieplayer, removing sound, adding garbage collector at the start and the end of every scene, removing plugins, with no success.

We aparently manage to solve the problem by updating de proyect to 4.18.3-3832480+++UE4+Release-4.18. After the update we tested 32 times in 2 diferent machines, so we hope the problem is gone. We asume was a internal engine bug that was remove in this version.