When the TC access dropped from LE for me (should have mentioned it is on an RPi3), I was also trying access from Win 10 (1703) which would then not connect again until the TC was restarted, restarting LE would not help. The only difference was my mount file based on the existing sample file did not contain the additional explicit options "sec=ntlm,vers=1.0" which maybe made some form of difference to it's resiliance?
I also ran into this a few weeks ago with an old 2TB TC I was using as the media store. If all devices were power cycled it would seem to work for a certain time period as long as there was constant SMB traffic to/from the TC but if it went idle the connections broke after about 30s. My guess is something in the new SMB Client implementation in 8.1.x and above (as well as the 9.0 Milhouse alpha builds) appears to trigger a SMB/networking fault in the Time Capsule itself. I found once the TC shares dropped in LE/Kodi, shares also stopped being accessible by any other other device as well until it was power cycled. Ended up retiring the TC and replacing it with a Seagate Personal Cloud device that although noisier when spinning up, works perfectly with the updated LibreELEC builds SMB client.