Electrical Causes of Motor Vibration: How to Separate Them from Mechanics

The motor is humming, there's a line in the spectrum that shouldn't be there, and someone suggests balancing the rotor. Sometimes that's the right call. And sometimes you'll spend a shift and end up with the same level, because the vibration was coming from an uneven air gap or a cracked rotor bar. You can tell one from the other with a single agreed shutdown. Below are the techniques, frequencies, and numbers for doing it right on site.

Updated 27 August 2026 · by AXILINE · Vila Nova de Gaia

Where a Motor Gets Vibration That Doesn't Care About Speed

The force with which the stator pulls on the rotor is proportional to the square of the magnetic flux density in the air gap. Square a 50 Hz sine wave and you get a constant term plus a pulsation at 100 Hz. On a 60 Hz supply that's 120 Hz. Every healthy motor has this pulsation; the only question is its level.

In a symmetric machine, the forces around the gap balance each other out, and almost nothing escapes to the outside. Break the symmetry and you're left with an uncompensated force that yanks the rotor to one side a hundred times a second. Three groups of causes break the symmetry: geometry (an uneven gap, a shifted stator, a bent rotor), the magnetic circuit (a loose stator core, a broken rotor bar, shorted turns), and the supply (phase imbalance, a burnt contact, an open phase).

Remember the key property of this vibration: its frequency is set by the mains supply, not the shaft. Load the motor, the speed sags, and the 100 Hz line stays exactly where it was. All the diagnostics that follow are built on this.

This creates two traps people regularly fall into. The first: on a two-pole motor, 2960 rpm gives 1x = 49.3 Hz and 2x = 98.7 Hz, while the supply gives 100 Hz. The difference is 1.3 Hz, and with typical measurement settings these lines merge into one. The second: the 100 Hz force itself can be perfectly normal, and what makes it high is a resonance in the frame or a foot. In that case you'll be hunting for a winding defect where what's actually needed is stiffening the structure.

• The electrical component sits at a frequency that's a multiple of the supply frequency: 100 Hz and 50 Hz on a 50 Hz supply. It doesn't shift along with speed.
• The mechanical component is tied to the shaft: 1x, 2x, vane-pass, bearing frequencies. Change the speed and the whole picture slides along the frequency axis.
• The electrical component usually grows with load, because current and field both grow.
• Imbalance doesn't care about load: the rotor's mass distribution doesn't change because of it.
• The electrical component disappears the instant the contactor opens. The mechanical one lives on as long as the shaft is turning.

A high 100 Hz line on its own isn't a verdict on the windings. It tells you only one thing: the excitation has an electromagnetic origin. The structure's response to that force can be amplified many times over by frame resonance or soft foot, and in that case repairing the motor will get you nowhere. Before hauling the machine off to the electrical shop, tap the frame and check the feet are tight. We have a separate article on resonance and the bump test.

The Coast-Down Test: One Shutdown Instead of Three Hypotheses

This is the cheapest and most convincing diagnostic technique there is. You cut the power and watch what happens to the suspect component in the first few seconds. The electromagnetic force disappears together with the current, within a fraction of a second. The mechanical one decays smoothly, because the shaft is still turning and gradually losing speed.

The key condition: recording has to be continuous through the moment of disconnection. Two separate spectra, before and after, won't settle the question, because you won't see exactly how the line disappeared.

1. 01

   Get the Shutdown Approved

   An unscheduled shutdown is a process risk and an extra start for the motor. Get clearance from operations, agree on who presses the button, and make sure the unit will actually restart afterward. Large machines have a limit on starts per hour, and that limit doesn't go away for your test.

2. 02

   Set Up the Measurement and Don't Touch It Again

   Sensor on the motor's bearing housing, rigidly mounted on a cleaned, flat spot, as close to the bearing as possible. Aim the laser phase sensor (which ties vibration to the shaft's angular position) at the reflective tape mark so you can see speed at the same instant. Don't change the point or the direction for the rest of the test.

3. 03

   Take the Baseline at the Running Condition

   Steady-state speed, working load, a warmed-up machine. A spectrum with enough resolution that 100 Hz and 2x sit as separate lines, plus a time-waveform recording. Record the actual speed with a tachometer — you'll need it to calculate slip.

4. 04

   Cut the Power and Keep Recording

   Start recording a few seconds before the shutdown and don't stop it until the shaft has fully stopped. If the instrument records spectra as a series, set the shortest possible acquisition time, otherwise the moment of disconnection will get smeared out by averaging.

5. 05

   Review the Recording

   The component that cuts off within a fraction of a second after the contactor opens is electrical. The component that creeps down in frequency together with the speed is mechanical. Look at both the spectra and the waveform shape: on the trace, the moment of disconnection is usually visible to the eye as a sharp change in the character of the oscillation.

6. 06

   If You Can't Shut Down, Change the Load Instead

   Take spectra at two or three load conditions at the same or similar speed. Electromagnetic components usually grow with current, while imbalance doesn't care about load. The conclusion is weaker than from a coast-down, but no shutdown is needed and there's no risk.

A variable-frequency drive breaks this test if you don't watch the stop mode. On most drives, a "stop" command means controlled deceleration: the motor is still under current, only the output frequency is falling. In that case the electromagnetic component won't disappear at all — it will just slide down along with the frequency, and you'll get a false "mechanical" conclusion. The test needs a free coast-to-stop mode or a break in the circuit at the drive's output.

Sidebands and Slip Frequency: Calculate Before You Go Looking

An induction motor's rotor always lags behind the stator's rotating field. That difference is slip, and it produces a second set of signs, far more informative than a lone 100 Hz line.

Work through it with real numbers. A four-pole motor on a 50 Hz supply has a synchronous speed of 1500 rpm. The tachometer reads 1480 rpm, so the running frequency f_r = 24.67 Hz, slip s = 20/1500 = 0.0133, slip frequency f_s = s · 50 = 0.67 Hz. Pole-pass frequency F_p = f_s · P = 0.67 · 4 = 2.67 Hz, where P is the number of poles. That's exactly the spacing at which sidebands appear around 1x when the rotor cage has a defect.

In other words, you shouldn't be looking for "something around 24.7 Hz" but for specific lines at 22.0 and 27.3 Hz. The difference between them and 1x is only 2.67 Hz, and this is exactly where typical measurement settings let you down.

Spectrum resolution equals Fmax (the spectrum's upper limit) divided by the number of lines, and the acquisition time for one record equals one divided by that resolution. Fmax of 1000 Hz over 400 lines gives 2.5 Hz per line: the sidebands physically can't separate, and you'll see one broad bump instead. Fmax of 200 Hz over 1600 lines gives 0.125 Hz and requires 8 seconds of acquisition. With settings like that, the sidebands are visible. There's a separate article with more detail on Fmax, line count, and windows.

The second requirement is dynamic range: the measurement's ability to show weak lines next to strong ones. Sidebands from a cracked bar sit 30–50 dB below 1x, so averaging over several records isn't a luxury here, it's the condition for pulling them out of the noise. On a lightly loaded motor, slip is small, the sidebands crowd right up against 1x, and separating them gets even harder. The takeaway is simple: take the spectrum at working load, not at no load.

• Sidebands around 1x spaced at F_p: a broken or cracked rotor bar, a defective end ring, loosened bar fits.
• Sidebands around 100 Hz spaced at F_p: dynamic eccentricity, meaning the gap changes as the rotor turns.
• Sidebands around the rotor slot-passing frequency (the number of bars times f_r) spaced at 100 Hz: the same cage defects, but the sign sits in the kilohertz range, and a velocity measurement over the 10–1000 Hz band simply won't see it.
• Sidebands around the stator slot-passing frequency (the number of slots times f_r) spaced at 100 Hz: a loose stator core, an uneven gap.
• Modulation of 1x itself, which makes the amplitude and phase drift with a period of 1/F_p (0.37 s in our example): the same rotor cage issue or dynamic eccentricity. Balancing under these conditions is pointless — the readings don't repeat.

Sources: ISO 13373-3:2015

Frequency and Sign, Probable Electrical Cause, How to Verify

The table gives you hypotheses, not a diagnosis. You can't choose a repair based on a single 100 Hz line: it has at least four different possible sources, and each one needs its own independent confirmation. The working rule is: coast-down test first, to separate electrical from mechanical, then current and temperature, to narrow the field, and only then teardown.

Six Electrical Defects, One by One

Variable-Frequency Drives: What They Add and What They Break in Diagnostics

A VFD changes the rules of the calculation. For the motor, the supply frequency is no longer 50 Hz but the drive's output frequency, and the electromagnetic pulsation sits at twice that value. If the drive is running at 35 Hz, look for the line at 70 Hz, not at 100 Hz. Calculate slip and pole-pass frequency from the output frequency too, otherwise the sidebands will show up somewhere other than where you expect them.

At the same time, the 100 Hz line can still be present: the DC-link ripple originates from the mains and leaks through into the output voltage. So on a VFD-driven machine you'll see both lines at once, and you shouldn't confuse them.

The PWM carrier frequency (pulse-width modulation, the method the drive uses to synthesize its output voltage) typically lies in the 2 to 16 kHz range. It produces magnetostrictive noise and vibration at the carrier itself, its multiples, and sidebands spaced at the output frequency. A standard velocity measurement over the 10–1000 Hz band doesn't reach that far, and that's normal: you'll hear the carrier as a characteristic whine before you ever see it in the spectrum. It's fixed with drive parameters and an output filter, not with weights on the rotor.

Bearing currents are a separate topic. The drive induces a voltage on the shaft, it discharges through the bearing, spark erosion eats away at the raceways, and fluting appears. Bearing life calculated per ISO 281 is based on load, geometry, and lubrication, and electrical erosion doesn't appear in that model at all. So a bearing that's calculated to last years fails within a season on a VFD-driven machine with no shaft grounding ring. The protection is structural: an insulated bearing, a shaft grounding ring, and a shielded cable with proper grounding.

• Upside: with a VFD you can smoothly change speed and genuinely check for resonance without touching the mechanics.
• Upside: you can bring the machine to exactly the speed you need for a measurement and hold it steady from run to run, and a stable speed is the condition for a repeatable phase measurement.
• Downside: the setpoint range almost certainly includes a speed where running speed or vane-pass frequency hits a resonance. The machine will be calm at 42 Hz and start shaking at 47 Hz, and balancing won't fix that.
• Downside: until you've established the stop mode, the coast-down test proves nothing.
• Downside: without a grounding ring you'll get a bearing defect that looks like ordinary wear but will recur even on a new bearing.

Sources: ISO 281:2007

Current and Temperature: Two Confirmations Without Teardown

Vibration tells you that symmetry has been broken. It doesn't tell you exactly where. Two independent measurements close that gap, and both are done on the running machine.

Current spectrum analysis is, in essence, a cheap way of doing motor current signature diagnostics. Clamp a current probe with an output to an analyzer onto one phase and take a spectrum with the same high resolution as the vibration. With a rotor cage defect, sidebands appear around the supply frequency at f_line · (1 ± 2s). In our example, with a slip of 0.0133, we get 2s · f = 1.33 Hz, so the sidebands sit at 48.67 and 51.33 Hz. The depth of the sidebands relative to the main line is measured in dB. As a rough guide: above 50 dB is considered normal, 40–45 dB is suspicious, and below 40 dB suggests a likely broken bar. Different methodologies use different thresholds, so treat the number as a reason to investigate, not as a verdict.

A thermal image answers different questions. It instantly finds a weak connection in the terminal box and at the cable lugs, shows an uneven heating pattern across the starter's phases, and picks up a localized hot spot on the casing above a turn-to-turn short as well as an overheating bearing housing. The limitation is an honest one: a thermal camera sees the surface, while the rotor cage is hidden inside the machine and cooled by the air flow. Take the thermal image after a loaded run, not a minute after starting, and compare it against past images of the same motor at the same condition.

• Phase voltages and currents at the terminals, with imbalance calculated in percent, not eyeballed.
• Current spectrum with resolution no coarser than 0.1 Hz, with sidebands at f_line · (1 ± 2s) either found or clearly absent.
• Thermal image of the terminal box, starter, and cable lugs.
• Thermal image of the stator casing and both bearing housings after a loaded run.
• Insulation resistance and per-phase winding resistance, if 100 Hz has risen with no obvious geometric cause.
• Actual speed taken from a tachometer, not from the nameplate. The entire slip calculation rests on this number, and an error of 5 rpm shifts the sidebands to where you won't find them.

Balanset-1A measures vibration, speed, and phase, shows the overall level and 1x, builds the spectrum and time waveform across two channels, and keeps an archive of measurements and reports. It doesn't measure current, insulation resistance, or temperature, and it shouldn't claim to. Vibration shows that the cause is electrical; determining exactly which one takes a clamp meter, a thermal camera, and a megohmmeter, working together with an electrician.

Sources: Balanset-1A operation manual

Why Balancing Doesn't Cure Electrical Faults, and How the Attempt Ends

Balancing only works with mass, and only at running speed. The instrument installs a trial weight, watches how the 1x amplitude and phase changed, and derives an influence coefficient — the specific "rotor, supports, foundation" system's response to a known added mass. It then solves the inverse problem and gives you the correction mass and angle.

A 100 Hz line doesn't fit into this scheme at all. It's not at 1x, it's not a multiple of running speed, and its source isn't mass in the first place. No matter how much weight you install, nothing in the air gap changes. The instrument will honestly reduce 1x to the target value and report "within tolerance," while the overall level — in mm/s RMS (root-mean-square) over the 10–1000 Hz band — stays almost the same, because the main contribution was coming from the 100 Hz line. The customer will see the same number as before and reasonably ask what they paid for.

It's worse when 1x really is high, but inflated by dynamic eccentricity or a cage defect. In that case the 1x amplitude and phase drift with a period of 1/F_p — a bit over a third of a second in our example — and drift more slowly along with load and warm-up. The measurement doesn't repeat from run to run. An influence coefficient calculated from a measurement like that is invalid: it describes not the system, but the random instant you happened to press the button.

• The instrument will produce a weight that lands at an effectively arbitrary angle. Vibration is roughly as likely to increase as to decrease.
• You'll add real mechanical imbalance to compensate for an electromagnetic force. That force depends on load, while mass doesn't. At a different operating condition the machine will shake harder than it did before you arrived.
• You'll spend a shift and three or four extra starts on a motor that may already have a cracked cage. Every start is a thermal shock to that same crack.
• You'll lose time. A broken bar and a turn-to-turn short don't stand still — they progress, and the cost of repair grows along with them.
• A report saying "within tolerance" will create a false sense that the machine has been sorted out, and the next person will start the diagnosis from scratch a month later.

There's also an opposite bias people forget about. An electrical defect and imbalance coexist just fine, especially on a motor with a coupling half or a pulley. Finding a 100 Hz line doesn't mean you should throw out the imbalance hypothesis: first eliminate the electrical cause, then take a fresh measurement, and only then decide whether balancing is needed. Choose the residual-imbalance standard by G class from the applicable part of ISO 21940, and assess the machine's condition by overall vibration and the zones in ISO 20816, checking the applicable part and edition for your machine.

Sources: ISO 21940-11:2016 · ISO 20816-1:2016

Who to Hand the Task To, and What We Bring When We Come Out

Balanset instruments are designed and manufactured by engineers who use them to do field balancing themselves. That's where this article's practical framework comes from: first prove the vibration comes from imbalance, and only then install weights. Balanset-1A gives you everything you need for that proof: two channels on the bearing housings, a laser phase sensor using a reflective tape mark, 1x amplitude and phase, speed, spectrum and time waveform, an archive of measurements, and a report where the before-and-after numbers land.

If 1x is confirmed, you remove it with the same instrument: single- or two-plane balancing using the influence coefficient method, fixed positions instead of a protractor, drilled-hole calculation, recalculating weights for other planes, and saved influence coefficients for a quick trim balance at the next service. There's a Balanset-1A OEM version without the case, for building into machine tools and test stands. If 1x isn't confirmed, you've saved a shift and a few extra starts, and that's a result too.

AXILINE comes out to the site with this equipment: measurement, breaking vibration down into its components, finding the cause, in-situ balancing in the machine's own bearings, a report with the numbers. If the picture points to electrical causes, we'll say so directly and hand the task to an electrician, instead of installing weights on the rotor. We also provide consulting support on the instrument itself: helping with measurement settings and spectrum interpretation.

• To mechanics: the 100 Hz line is amplified by frame resonance, there's a soft or settled foot, the bedplate has twisted, a worn bearing has let the gap drift. This is adjustment and tightening, often on site and the same day.
• To the site electrician: phase current and voltage imbalance, a weak connection in the terminal box or starter, questions about VFD parameters, a missing shaft grounding ring. The cheapest group of causes, and the sensible one to start with.
• To the motor repair shop: a broken cage bar, a turn-to-turn short, a loose stator core, static eccentricity from worn fits. This is teardown, not adjustment, and gets resolved by shutting down and replacing or repairing the motor.
• To the vibration technician: the signs contradict each other, a shutdown can't be arranged, a high-resolution spectrum and a properly recorded coast-down are needed, or after the electrical cause has been eliminated a high 1x remains and needs to be removed.

Sources: Balanset-1A manufacturer specification

Frequently asked questions

Related content