I still remember the day a brand-new blower came back to the bench with hairline cracks in its motor-mount ribs. The customer said the casting must be bad. The supplier said the steel must be cheap. Neither was right. The fan ran at 2900 rpm under load, which works out to a shaft speed of 48.3 Hz, and when I threw a logger on the frame the peak acceleration was sitting almost exactly where the first bending mode of that bracket lived. Nobody had thought to ask what happened between 40 and 60 Hz. That lesson I now spread to every junior modeler who puts a motor on a plate without checking the numbers first.
Resonance Is Not A Statistic, It Is A Date With Reality
Here is the thing people get wrong about modal analysis: it is not a stress calculation. The solver does not care how strong your material is. It only wants to know the natural frequencies and mode shapes of the structure as it sits right now, with the mass, stiffness, and boundary conditions you gave it. When an excitation frequency lands on a natural frequency, the displacement amplifies by the Q factor of the mode, which for a lightly damped steel bracket can be twenty or thirty times whatever the static deflection predicted. That is how a casting rated for 8 g of static load snaps a rib under what looks like smooth running.
| Quantity | What it means for the bracket | Where I get it |
|---|---|---|
| Natural frequency | Danger point for a given mode | Modal solver, first 50 modes |
| Mode shape | Where the part flexes and where the nodes are | Deformation animation |
| Mass participation | How much of the mass a mode actually drags along | Latest-modal output table |
| Q factor / damping | How violent the amplification gets | Measured or assumed, 1–3% for steel |
| Effective mass | Which modes matter for a real load direction | Mass participation report |
I usually pull forty to sixty modes for a machine sub-assembly even when I only care about the low end, because the mass-participation numbers tell you whether you have captured enough of the structure. If the first five rigid-body-ish modes swallow ninety percent of the effective mass in the vertical direction, the whole excitation story is dominated by them, and pointing a camera at the tenth mode is wasted effort. That table, not the pretty deformation plot, is where the engineering decision actually lives.
My 28 Hz Story, Or: What The First Prototype Taught Me
The bracket that bankrupted my confidence for a week was a welded steel motor mount with a baseplate 6 mm thick and two side gussets. The FEA told me the first bending mode was at about 46 Hz, which looked great against the 48.3 Hz running speed until I realized I had modeled the motor as a rigid slug of mass with no realistic bearing compliance, and the actual first mode came in at 44 Hz with a node line right through one gusset weld. The measured 28 Hz peak I keep quoting was the second operational amplifier of a torsional sway mode nobody in the first review believed existed. In the end the fix cost nothing but geometry: a third gusset rotated 90 degrees moved that mode up past 70 Hz and the cracks never came back.
If your modal model is stiffer than reality, every margin you report is a fiction. One loose bolt in a baseplate can drop the first natural frequency by more than a stiffener will ever raise it.
That bolt comment is not decorative. I have measured brackets where simply torquing the M8 hold-downs from hand-tight to 25 N·m pulled the first mode up by 8 Hz, because the joint stiffness, not the plate thickness, was the soft spring in the chain. Modal analysis only earns its keep when the boundary conditions in the model match the constraint in the real machine, and that sentence takes engineers years to really accept.
Boundary Conditions: The Place Where Modal Models Lie
In every modal review I sit in, ninety percent of the disagreement comes from the boundary assumptions. Ground the bracket as fully fixed and the first mode jumps 15 percent higher; model it as free-free and the numbers collapse. The truth for a bolted steel frame is somewhere in between, closer to the free end than anyone wants to admit. My working rule is to model the mounting feet with a soft spring of realistic joint stiffness rather than either extreme, and then to torture-test the model twice: once with stiff joints and once with soft ones, and to treat the spread between them as the honest uncertainty band on every frequency I report.
There is also the motor bearing question I keep harping on. A motor is not a rigid chunk; it has rotor mass spinning on bearings that flex with preload and temperature. If I throw the whole assembly on a plate and run modal analysis, I am computing the plate in vacuum. For a first pass that is fine, but before sign-off I run two more variants, one with the rotor as concentrated mass at the bearings and one split between rotor and stator, because that change alone shifted my critical mode by 6 Hz on a 22 kW machine. Nobody disputes that number in review, because the report shows the band.
Running The Analysis So The Report Survives Review
Start with the mesh. A modal mesh does not need the refinement of a contact stress model, but it does need enough elements across thin flanges to carry the bending waves, otherwise the solver happily reports a frequency that belongs to a mesh that never existed. I keep four elements across the thinnest plate, use solid elements for the bracket and lumped masses for the motor, and I check convergence by re-running with a denser mesh until the first-three-mode frequencies change by less than about one percent. Two runs is usually enough.
Then I request the outputs I actually need: the first, say, forty modes, the effective mass per direction, and the mass participation percentage per mode. I sort a little table in my head rather than trusting the PDF, and I flag any mode whose direction coincides with a real excitation. For a motor mount the excitations I care about are the running speed 1x, the blade-passing or pole-passing frequency if there is a fan or a switched reluctance driver, and the two-times line frequency if it is an AC machine on a flexibly mounted base.
| Excitation source | Frequency clue | Typical danger zone |
|---|---|---|
| Fan / blower 1x | RPM ÷ 60 | First bracket bending mode |
| Blade passing (BPF) | Number of blades × 1x | Duct and cover panel modes |
| Pole pass / 2× line | 100 or 120 Hz | Short stiff brackets, laminations |
| Gear mesh | Teeth × mesh frequency | High-order torsional modes |
| Start-transient sweep | 0 → running speed | Every mode in the ramp band |
That last row is the one that bites soft-start industrial fans: during ramp-up the rotor sweeps through every frequency between zero and running speed, so a mode that sits safely above 1x is still crossed twice per start. If the machine starts and stops often, you are not designing against a single frequency, you are designing against a band, which is why I aim for a 25 percent margin above the highest sustained excitation instead of the classic 10 to 15.
Moving A Mode: Stiffness, Mass, And The Order Of Attack
When a mode sits too close to an excitation, the fastest lever is almost always stiffness, not mass. Adding mass lowers frequency, which sometimes wins if you are tuning short, but on a motor bracket you usually want to push up, so I first attack the weak bending direction: a vertical gusset kills a vertical sway mode faster than doubling the plate thickness, and it adds grams instead of kilograms. After stiffness, I look for the cheapest geometric habit, orienting the first and second bending modes so the stiffest axis faces the highest excitation direction. On the bracket that failed at 28 Hz, rotating the motor 90 degrees on its base was the entire fix, and it did not cost a single gram of steel.
If damping is on the table, treat it as a relief valve, not a cure. A neoprene pad or a tuned mass damper can drop the response at a specific frequency, but dampers have to be tuned to a target and they drift with temperature and aging. I have mounted a 0.8 kg tuned mass damper on a fan cover before, tuned to 44.5 Hz, and it worked beautifully for eight months and then the rubber hardened and the peak came back. For production machines I use damping only when geometric separation is impossible, and I write the retune interval right into the maintenance schedule.
Verify With A Hammer, Not With Hope
No modal report leaves my desk until a hammer test agrees with it. The setup is embarrassingly cheap: an impact hammer with a force cell, a small accelerometer, and forty seconds of data. I tap the bracket in the direction of the mode I care about, ring the part on the bench in the same constraint the real install will see, and I compare the measured resonance peaks to the simulation. When the first measured mode lands within five percent of the predicted one I sign it off; when it does not, I stop trusting the model and go look at the boundary conditions again, because the physics of steel does not miss a mode by fifteen percent without lying somewhere in my assumptions.
One warning about hammer testing a motor bracket on a bench: if you clamp it in a vice with a force nobody will reproduce on site, you are measuring your vice. I bolt the bracket to the actual baseplate pattern, or a rigid proxy of it, and I record the torque I used, because as I said, joint stiffness moves everything. A colleague once caught a first-mode discrepancy that turned out to be the soft rubber feet under the test bench, which let him off the hook and let the real part almost kill a launch.
The Closing Numbers
So here is the summary I wish someone handed me the day that first prototype cracked. Identify every excitation frequency in the system before you mesh, including the start/stop sweep band for motors. Model boundary conditions as a range, not a single assumption, and report the band. Run enough modes to see the mass participation story, and trust the effective-mass table over the pretty animation. Aim for 25% separation margin when the machine cycles, and lower it to 15% only for a continuous non-starting duty, and never lower it because the static safety factor is high, because modal margin and stress margin are unrelated currencies. Then verify with an impact test and learn the one question that closes every review: where is the node, and what is it doing to my weld.
The bracket that taught me about 28 Hz now sits in my office with the cracked rib facing outward, and I lend it to every junior modeler who is about to bolt a motor to a plate. You can read about modal analysis in any textbook. But you will only believe in it once a number on a screen has matched a crack in the steel, and that is the day it stops being academic.
