I have had a look at the scheme and found two details to be "repaired":

1. There's no cap at the DC-DC-converter between pin2 and pin4 (GND and AGND)! This will inject enormous switching interference directly on the RS485 bus.

Solder a cap of about 10nF directly between pin2 and pin4 with very shortest connections (every millimeter counts!). Sometimes two or more caps in parallel at this place can decrease the switching noise even further.

2. ESD protection transzorbs add enormous capacitance (2...3nF per transzorb!) to the bus lines. This will destroy the proper 100R termination of 100R transmission line. Also, the number of participating bus partners will heavily influence the amount of mismatch.

A better approach is to use slew rate limited and ESD tolerant RS485 transceivers, like the LTC1487 e.g., and omit these ferrite beads and transzorbs with their huge capacitances. The LTC1487 is also fail-safe and doesn't need these R8 and R9 voltage divider resistors.

Kai Klaas