How to Tell Imbalance From Misalignment Mechanically

Why this matters

Imbalance and misalignment both show up as a machine that shakes, and the two corrections take different skills and are ordered from different suppliers. Get it backwards and the wheel comes back from the balance house with a clean report while the seal keeps weeping, or the crew spends a morning on indicators for a fault in the rotor. The two faults are not distinguished by how bad the shaking is. They are distinguished by the shape of the force each one creates, and once you know that shape you can design physical tests that change one and leave the other alone. Reading the frequency signature is a real method the library covers elsewhere; this article separates them from the mechanics, with tools any shop already owns.

Stop before you touch anything

Every test below except the running-speed comparison happens on a stopped machine, and stopped means proven stopped. Open the disconnect, apply your own lock and tag, and confirm the rotor is not still coasting before a hand, an indicator base or a hammer goes anywhere near it, per the energy-control and verification requirements of 29 CFR 1910.147. If any part of the call takes you into the motor terminal box, the starter, or a variable-frequency drive enclosure, that is electrical work under 29 CFR 1910.333(b)(2) rather than 1910.147, which excludes electric utilization installations at 1910.147(a)(1)(ii)(C); prove the conductors dead with the live-dead-live sequence in NFPA 70E-2021, 120.5, and on a drive wait out the manufacturer's stated DC-bus discharge time and verify absence of voltage at the bus terminals before touching them. For the one test that requires the machine running, take the reading from outside an intact guard with a magnetic-base sensor on a stationary housing, no gloves and no loose sleeves near the shaft, because 29 CFR 1910.219 does not stop applying while you collect data.

What each fault actually does, mechanically

Unbalance is a mass-radius product that has not been corrected, and at speed it produces a force that rotates with the shaft. That force is radial, it points outward from the centerline in a fixed direction relative to the rotor, it is present whether or not the machine is coupled to anything, and its magnitude climbs with the square of speed. It is a property of the rotor alone.

Misalignment is a geometric mismatch between two shaft centerlines, and it produces nothing at all until the two shafts are connected. Once they are, the coupling is forced to accommodate that mismatch on every revolution: it flexes, it slides, or its elastomer compresses and relaxes, and the reaction to that flexing is fed back into both machines' bearings. Angular misalignment, where the centerlines meet at an angle, drives a component along the shaft axis, because the coupling is being opened and closed axially as it turns. Parallel offset, where the centerlines are parallel but displaced, drives a radial reaction. Most real misalignment is a mixture, and the library has separate articles on telling those two apart and on what the resulting load does to a bearing.

That gives you the lever: unbalance is a property of one rotor, misalignment is a property of the connection. Every test below exploits that.

The sequence, arranged so the most expensive omission comes first

Each step below is placed by what a shop loses when it skips that step, not by how the work flows. Skipping step 1 corrupts the evidence for all five that follow, so it goes first even though it feels like a detour.

1. Rule out amplification before you interpret anything

Strike a stationary bearing housing with a soft-face hammer, machine locked out and stopped, and see where the structure rings; or watch the amplitude through a coast-down after the disconnect is opened, reading from outside the guard. You are looking for a natural frequency of the machine and its support near running speed. The method and its limits are covered in the companion article on working out whether you are near a critical speed.

What you lose by skipping it: every subsequent observation. Near a resonance, a speed change produces amplitude changes that follow the magnification curve rather than the force, so step 4's arithmetic inverts. Techs who skip this step reliably conclude "it got worse when we slowed it down, so it cannot be unbalance", which is exactly what resonance does.

2. Uncouple and run the driver by itself

Break the coupling, secure the driver so its exposed shaft and coupling half are guarded, remove or positively retain the shaft key (an unrestrained key leaving a shaft at speed is a projectile), and run the motor alone briefly. Compare the reading at the motor bearings against what it read coupled.

What you lose by skipping it: half the machine. This single test cuts the problem in two. If the motor alone still shakes, the fault is in the motor rotor, its coupling half, or its own mounting, and the driven machine is innocent. If the motor alone runs smooth, the fault needs the connection, which means misalignment, unbalance in the driven rotor, or a coupling defect. Shops that skip this test are the ones that balance a fan wheel and then discover the motor was the source.

3. Read direction of motion at each bearing, with the geometry attached

With the machine running, its guards in place and a magnetic-base sensor set on each stationary bearing housing from outside the guard, compare motion along the shaft axis against motion across it. On a straddle-mounted rotor, meaning one carried between two bearings, unbalance produces a radial force and very little axial motion, so a substantial axial component points to angular misalignment or a bent shaft.

That inference is derived for the straddle geometry and does not survive a move to an overhung rotor, where the wheel hangs outboard of both bearings. There, an unbalance force acting on a wheel outside the bearing span creates a rocking moment about the bearings, and that moment can show up as axial motion at the housing through the pedestal's rocking mode. On an overhung fan or pump, the axial test does not separate the two faults, and you go to steps 4 and 5 instead.

What you lose by skipping it: the plane, not the machine. You still know which side has the fault from step 2; you have just lost the cheapest clue about which fault it is.

4. Change the speed and check the response against the square law

If the machine has a variable-frequency drive or an easily swapped sheave, take a reading at running speed and at a meaningfully lower speed, staying inside the manufacturer's minimum continuous speed and, on a pump, above its minimum continuous flow, so that gathering the data point does not damage the machine. Unbalance force falls with the square of the speed ratio, so a 30 percent speed cut takes the force to about half. Misalignment reaction is set by the geometry of the mismatch and does not follow that curve; it typically softens far less.

What you lose by skipping it: the confirmation. Steps 2 and 3 identify the fault; this one proves it behaves like the mechanism you named.

5. Add a trial mass and see whether the rotor responds

If everything so far says unbalance, mount a known trial mass at a known radius on the rotor, positively secured with a fastener rated for the job and not with tape or a magnet, machine locked out while it goes on. A rotor whose problem is unbalance responds strongly and predictably to a trial mass; one whose problem is a constraint at the coupling barely notices it.

What you lose by skipping it: certainty before you spend. Most shops order the balance job without it. The step earns its place when the customer is being asked to pull a wheel, because it is the only cheap evidence that pulling it will change anything.

6. Measure the alignment cold, with indicators

Dial or laser indicators on a stopped, locked-out machine, removed before anything is restarted. Compare against the coupling and machine manufacturer's cold-setting target for that speed and coupling type, not against a general rule, and allow for thermal growth using the method in the companion article on aligning cold.

What you lose by skipping it: durability rather than diagnosis. By this point you already know which fault you have. What this step gets you is the number to correct to and the record that proves you hit it.

Worked example: the pump that kept eating seals

A direct-coupled end-suction pump at 1,780 rpm through an elastomeric coupling, third mechanical seal in fourteen months, and a maintenance tech who has already asked for a quote to balance the impeller.

Step 1. Locked out and stopped, a soft-face hammer on the pump bearing housing rings the structure at about 2,900 cpm. Running speed sits at a ratio of 1,780 / 2,900 = 0.61 to that, well below it, where the response is close to quasi-static and the magnification works out near 1 / (1 - 0.61 x 0.61) = about 1.6. Not resonance, and low enough that response will track force reasonably closely, which is what makes step 4 readable.

Step 2. Uncoupled, with the key retained and the shaft guarded, the motor alone reads about a quarter of what it read coupled. The fault needs the connection.

Step 3. Coupled again, axial motion at the pump inboard bearing runs about 80 percent of the radial motion at the same point. This is a straddle-mounted rotor, so that axial share is far too high for unbalance and points squarely at an angular component at the coupling.

Step 4. Turning the drive down 30 percent to 1,250 rpm, still above the pump's minimum continuous flow: unbalance force alone would fall to (1.25 / 1.78) squared = about 0.49 of its value, and the magnification also falls, from about 1.6 to about 1.23, so a pure unbalance would predict a reading near 0.49 x 0.77 = 38 percent of the original, a drop of roughly 60 percent. The machine actually dropped about 20 percent. That is nowhere near the square law, and it is the number that ends the argument.

The sequence stops here. Steps 2, 3 and 4 all point the same direction, so the trial mass in step 5 has nothing left to resolve, and the crew goes to indicators.

Step 6. Reverse-indicator readings work out to roughly 0.015 in of offset at the coupling with a clear angular component, several times the coupling maker's cold-setting target for this speed class. Loosening the discharge flange bolts drops the indicator reading noticeably before anything is re-shimmed, which names the cause: pipe strain pulling the pump off its aligned position, introduced when the flange gaskets were replaced.

What getting this wrong would have cost. The impeller goes out for balancing, comes back inside grade, gets reinstalled, and the fourth seal fails on the same schedule, because a coupling still flexing through 0.015 in every revolution is still driving the shaft against the seal faces. The shop has spent a wheel pull and a balance job to buy no change, and the customer now has a reason to doubt the next recommendation.

What flips the reading

Two conditions genuinely invert the method rather than just complicating it. A bent shaft behaves like both faults at once, producing a once-per-revolution radial force like unbalance and an axial component like angular misalignment, and it will not respond usefully to a trial mass; runout measured on the stopped shaft separates it. And a coupling with a failing element, a cracked elastomer or a spider chewed on one side, produces a per-revolution reaction that looks like misalignment on every test above while the shafts are aligned correctly; inspect the element itself before you re-shim anything.

References

• 29 CFR 1910.147 for isolation, lock and tag, and verification before handling a rotor or mounting indicators
• 29 CFR 1910.333(b)(2) and NFPA 70E-2021, 120.5 for the electrical fork when the work moves into a terminal box, starter or drive enclosure
• 29 CFR 1910.219 for guarding of couplings, shafts and belt drives while a running reading is taken
• Coupling and machine manufacturer documentation for cold-setting alignment targets by speed and coupling type
• See related: How a Small Imbalance Becomes a Large Force; How to Tell Angular Misalignment From Parallel Offset; What Misalignment Does to a Bearing in Mechanical Terms; How to Allow for Thermal Growth When You Align Cold