I tried to detect this issue in automated process and checked for 152 locklosses from DC lock state since September.
They seems to be detected by using the 1st‐order differentiation of oplev sum.

There were 27 obvious events in this term such as Fig.1.
In addition, there were several events which was slightly difficult to judge a jump in oplev sum was really a cause of lockloss such as Fig.2.

If we set the threshold for minimizing a false detection it's not so difficult. Threshold can be decided from past lockloss events.
On the other hand, it might be difficult to reduce false dismissal cases especially for the case that oplev light became unstable in long term.