When PSDs Lie: Non-Gaussian Vibration and Fatigue Damage

Random vibe specs still lean on Gaussian assumptions. We qualify to a PSD, GRMS, and notches look fine on the shaker. Then a fleet bracket cracks in months. Pull the flight recorder: same RMS, very different story. Spiky content, high crest factor, kurtosis around 7-9.

We repeated the test with kurtosis control and strain gages at the hot spot. Same PSD, same GRMS, but damage doubled. The strain rainflow histogram shifted right, and Miner's sum went from 0.4 to 0.9 for the same hours. The fix wasn't exotic: small fillet change, clamp load audit on the joint, and a touch more damping tape moved the mode and cut the peaks.

I'm curious how others handle non-Gaussian environments. Do you require kurtosis-controlled vibe for fragile attachments? Any rules of thumb for matching time-domain damage rather than just PSD equivalence? And how are you convincing certification folks when the spec only names RMS?