如何过滤多对多字段的模型？

I'm trying to implement a geofencing for a fleet of trucks. I have to associate a list of boundaries to a vehicle. On top of that one of the requirements is keep everything even once it is deleted for audit purposes. Therefore we have to implement soft delete on everything. This is where the problem lies. My many to many field does not conform to the soft delete manager, it includes both the active and the inactive records in the lookup dataset.I'm trying to implement a geofencing for a flee