Run just board 3 for an hour see if it works the whole hour
Next run just board 2 for an hour see if it works for an hour
last run just board 1 for an hour see if it works for an hour.
There is really no need for all that, line 463 in the kernel logs tells the whole story
driver-btm-c5.c:15886:re_open_core: Chain[J6] has 63 asic
driver-btm-c5.c:15886:re_open_core: Chain[J7] has 63 asic
driver-btm-c5.c:15886:re_open_core: Chain[J8] has 0 asic
Chain J8 ( most left ) is failing, in general, you can't do anything about it but watch it toast even more, soon enough it could start making the miner restart itself every now and then - and thus affecting the performance of the two working boards J6 and J7, however, there is hope and that would be by trying a different firmware such as BriianOs for Vnish (Asic.to / Awoesmeminer), and you want to play with the frequency of the bad board, start by going really low, test it for say 30 mins and then go higher slowly, sometimes increasing the voltage helps but keep that as your last resort.
if all the above fails, you should simply discount the bad board, and then separate the two boards, put on to the left and one to right (keep the middle empty) this way you ensure the boards get better cooling since the board in the middle usually runs hotter than the once on the sides.
