1-in-50 intermittent deviation with good parts — that pattern is almost always thermal. Here's how to confirm: log the deviation alarm timestamps for a shift. If they cluster 45-90 minutes after cold start and then again after lunch restart, it's thermal growth changing your BDC position by 0.01-0.03mm, just enough to occasionally exceed the deviation window.
Quick diagnostic: put a dial indicator on the slide at BDC, run 200 cycles, and watch the reading drift. On our Komatsu H1F the slide drops 0.025mm over the first hour as the frame warms up. That's enough to trigger Pn520 if it's set tight.
The real fix depends on your drive. On Yaskawa, widen Pn520 from the default (usually 200 pulses) to 400-500 for press work — the default is meant for CNC positioning, not forming. On Fanuc, it's parameter 2064. But don't just crank it open — if you're seeing >0.05mm deviation, something mechanical is actually moving and you want to know about it.
One more thing — check if the alarm correlates with specific die numbers. We had a progressive die where station 4 (a heavy coin) would occasionally spike the deviation because the nitrogen springs were losing pressure over the shift. Topped them off and the alarms disappeared. Sometimes it's not the press at all.