When a 5% modal match is not enough

We signed off a bracket after updating the FE model to match test frequencies within 3 - 4%. Flight test later found a crack at a fillet. Root cause: the test fixture was much stiffer than the installed structure. Mode shapes looked fine by eye, but strain energy parked itself in a different spot. Our accelerometers missed the local rotation, MAC looked great overall, and only after teardown did strain gauges show roughly 2x strain at the fillet in flight.

Lesson learned: frequency correlation alone can be a false sense of security. For vibration critical parts, I care more about local mode shape fidelity at hotspots, strain gauge to FEA strain correlation, and realistic boundary compliance. I now push to measure fixture stiffness, place gauges near notches, and correlate PSD to stress transfer functions, not just FRFs.

What is your minimum evidence to freeze a vibration model? Do you use energy based metrics, local MAC, or modal strain energy comparisons? Any practical tricks to validate joint preload and contact conditions without blowing the budget?