When the model lies, check your boundary conditions

Recently chased a high-cycle fatigue crack in a simple bracket. FEM predicted 30% margin, shaker replication said otherwise. Root cause: test article had realistic stack-up, paint, sealant, aged isolators, and torque scatter, which changed joint stiffness and damping. Our model used ideal bolts and uniform contact. Strain gauges lit up at a lug we barely meshed.

Updating joint modeling (surface contact with micro slip, preload scatter, frequency-dependent damping) closed the gap. The life went from infinity to a number that matched the crack. Not glamorous, but getting the boundary right beat tweaking S-N curves.

How are you treating joints in vibration fatigue these days? Do you calibrate damping and preload from coupon tests, or lean on handbooks until flight test forces a correction? What small non-idealities have most moved your margins?