I have not used [analyze aggregation] before, but tried a simple test. See your script - modified and attached.
Looking at the output files with the distance criteria 1.0, 1.5 and 2.0, obviously the result is not always a single aggregate. But it is very sensitive to the distance and none of these results is compatible with the visual inspection, which gives 2 aggregates plus a number of free chains (each of them is counted as an "aggregate" by itself). This could indicate a possible bug in the implementation, but I have not had time to check the code.
Please be aware that for a meningful simulation of aggregation, you need to simultaneously satisfy two things:
1. average aggregation number much smaller than the number of chains per box (rule of thumb >10 aggregates per box), otherwise you get strong finite size effects
2. simulation runtime much longer (rule of thumb > 100times) than the timescale of the exchange of chains between different aggregates. Otherwise you can't be sure whether you are not simulating a metastable state. This time scale drastically increases with aggregate size and LJ epsilon.
I am afraid that meeting these criteria with Langevin dynamics will be a tough job.
Best regards,
peter