When the joint is the spring: preload scatter bites fatigue margins

I spent the last month chasing a 2x spread in rib attach fastener life between two ostensibly identical test articles. FEM said margins were healthy and shaker PSDs matched, yet one article cracked at 35% of life. The only consistent difference: clamp-up. A slightly different primer stack and a torque tool calibration shift gave about 20% lower preload and a big jump in joint slip.

Once we measured loss of preload across thermal cycling, the vibration damping fell and strain hotspots moved to the first engaged thread. Our linear model with bonded contact looked perfect on paper, but the joint was the spring and the damper. Switching to preloaded contact with friction and adding a small run-in cycle brought predictions within 10% of test.

Lesson learned: if the failure is at or near a joint, treat preload, friction, and the faying surface as first order. How are you quantifying preload scatter and degradation in your vib fatigue analyses, and are you baking it into your allowables or your test setup?