Before the item closes
Two numbers on the report decide it
Consider a kiln fan booked for a motor current test on Saturday, when the process runs light and the switch room is quiet. Monday's report: no rotor sidebands. On the 1,470 rpm motor worked below, a one-second record at a quarter load cannot separate the expected broken-bar pair from the supply line.
Now put yourself in the planner's position: an ageing motor, a previous end-ring repair and a long wait for a replacement. A clean report would be welcome. Two numbers on the report decide whether that is safe: the load during the test and the length of the current record.
The useful question is not just "Did a peak appear?" It is "Could this test have separated it?"
The mechanism
The fault sits twice the slip frequency from the line
A cage rotor carries current because the stator's field turns slightly faster than the rotor does. The difference is the slip, a few percent at most at full load. When a bar breaks, the bars beside it carry more than their design current, and the cage's pattern of currents is no longer symmetrical.
Rotor currents create a field at slip frequency, and any asymmetry adds a backward-rotating component. The stator sees the result as a pair of sidebands at (1 ± 2s)f, twice the slip frequency either side of the supply line. They grow with the number of broken bars, and because slip is proportional to load, they move as the load moves.
In a 1 s record the 50 Hz line spreads 2 Hz either side, and on a 4-pole motor rated 1,470 rpm the broken-bar pair never gets clear of it.
The arithmetic needs nothing more than a nameplate. On a 4-pole, 50 Hz motor rated 1,470 rpm, synchronous speed is 1,500 rpm and full-load slip 30 rpm, or 2%. The sidebands sit at 48 and 52 Hz at full load, 49 and 51 Hz at half load, and 49.5 and 50.5 Hz at a quarter.
Load and size
Light load and a large motor close the gap
That nameplate belongs to the 5.5 kW laboratory motor in SEL's broken-bar paper. With two bars broken, the fault component in the paper's processed spectrum moved as the arithmetic says, from about 2 Hz at full load to 0.5 Hz at a quarter load. Unloaded, it could not be separated from the line at all.
Large motors usually run with less full-load slip, which makes this worse. The same paper lists a 1.656 MW motor rated 1,488 rpm: 0.8% slip, so its sidebands sit only 0.8 Hz from the line at full load and 0.2 Hz at a quarter.
A 1 hp motor in a published broken-bar dataset is rated 1,715 rpm at 60 Hz: 4.7% slip, with its full-load sidebands 5.7 Hz out. Same fault, same physics, and on the big motor the evidence sits seven times closer to the line.
What the record can separate
Record length sets the spacing between spectral lines
For an unpadded FFT, the spacing between spectral lines is set by the record length: T × Δf = 1. One second of data gives spectral lines 1 Hz apart; ten seconds, 0.1 Hz; a hundred seconds, 0.01 Hz. Sampling faster does not narrow that spacing. Separating a weak component beside a strong one also depends on the window, their relative levels and the noise.
The supply line has a width of its own. With a Hann window, the usual default, a strong component spreads about two lines either side of its centre, and its skirt can bury a weak component nearby. In a 1 s record that is 2 Hz either side, with the quarter-load pair deep inside it.
The same pair at three loads: a 1 s record separates it at none, and a 10 s record clears it at full and half load but leaves only a bump at a quarter.
A sideband 40 dB below the line starts to show at five lines and stands clear at about eight; one 50 dB down needs ten. A published rule of thumb reads a sideband 40 to 50 dB down as a likely single broken bar, so this note's working rule is ten lines of clearance.
The 1.656 MW motor's full-load pair sits only eight lines out in a 10 s record, so even at full load it needs 12.5 s. SEL's own relay, whose fixed window cannot tell apart components closer than 0.3 Hz, stops separating broken-bar components below a 0.15 Hz slip frequency: about 38% load on that nameplate.
The catch
A long record needs a steady speed
The obvious fix is to record for longer, and it works only if the speed holds. A Fourier transform assumes the signal does not change through the window. If the slip wanders during the record, the sideband wanders with it, and for any 4-pole motor a speed change of 15/T rpm, T in seconds, moves it by one line.
Ten lines of clearance for three published nameplates: the lighter the load and the larger the motor, the longer the record and the steadier the speed it needs.
A 10 s record needs the speed held within about 1.5 rpm, and every doubling of the record halves that allowance.
What else puts a peak there
Load and supply can imitate a bar
A load that pulses puts its own sidebands at the pulse frequency from the line. SEL ran a motor whose load oscillated at 1 Hz and got a signature like that of three broken bars at 50% load.
A supply with 1.5% ripple at 2 Hz produced much the same picture. Hence the first defence: if several motors on the same duty or feeder show the same pattern, it most likely belongs to the load or the supply, not to one rotor.
The second defence is a measured speed, not the nameplate's. NEMA allows manufacturers a 20% tolerance on nameplate full-load speed, reckoned on the slip: 6 rpm on the 1,470 rpm motor, or 0.4 Hz of sideband spacing.
Worked on two published nameplates
What a quarter-load record can show
The 5.5 kW and 1.656 MW nameplates from SEL's paper, both at a quarter load.
From each sideband to the supply line, on the 5.5 kW and the 1.656 MW motor.
Both pairs sit within the supply line's own spread. Nothing can be separated.
The 5.5 kW pair is only a bump on the skirt, and in some records not even that; the large motor's pair sits at the lobe's edge.
Ten lines of clearance for each motor at the same quarter load.
15/T over those records; any wider drift moves the sideband more than a line.
Use in the field
Specify the test before it is booked
Qualified people collect the record under the site's electrical safety programme. What goes into it is the planner's part, and it belongs in the booking.
- LoadAs close to full load as the process allows, held steady. Record the estimate and its method. At light load, prefer measured input power to current alone; it remains an estimate of shaft load.
- SpeedMeasured by tachometer or strobe during the record, with the supply frequency, so the sideband positions are known in advance.
- Record lengthAt least 10 / (2 × slip × f) seconds, with slip as a fraction at the test load (full-load slip × load) and f the supply frequency: 25 s for the 1.656 MW motor at half load.
- WindowHann, named in the booking; the ten-line rule assumes it.
- SteadinessSpeed within 60 / (poles × T) rpm for the whole record (15/T on a 4-pole motor, 10/T on a 6-pole), and the record rejected if the motor was not steady, as SEL's relay does.
- ComparisonThe motor's own earlier record at a similar load, or a sister motor on the same duty measured the same way, supply voltage noted.
- EvidenceRaw three-phase current waveforms, line voltage if available, nameplate data, measured speed, load, sampling and window settings.
Technical sources
Primary references behind this field note
These references inform the measurement and reasoning principles in this field note. They do not imply ConditionCore certification or conformity with a particular standard.
- Pezzani et al. 2014Sideband growth, nameplates, the 0.3 Hz relay limit and look-alikes; Sections I to VIII.
- SEL-710-5 data sheetBroken rotor bar detection "under sufficient motor load conditions"; p. 14.
- Maharjan et al. 2025Sidebands, full-load records, the 40 to 50 dB rule, stationarity; Sections 2 and 3.
- DOE motor-load fact sheetSynchronous speed, slip proportional to load and the NEMA tolerance; pp. 5 to 7.
- DOE motor selection guideDesign B slip at rated load and full-load speed against motor size; p. 4-11.
- Siemens DSP primerRecord length and frequency resolution; Section 3.
- NI, FFTs and windowingLeakage, main and side lobes, and the Hann recommendation; Windowing section.
- Treml et al. 2020 datasetThe 1 hp, 1,715 rpm motor and its 12.5 to 100% loads; public description only.
- ISO 20958:2013Signature analysis of three-phase induction motors; catalogue abstract only.
- ConditionCore MCA pageAccepted inputs, the sideband-family wording and the safety scope; full page.
The ten-line working rule is this note's own, from a Hann-window calculation checked over sideband phase and the supply line's position between spectral lines.
Back at the fan
Run the test where it can see
The Saturday window will come round again. With the test specified this way, the question changes from whether the analyst found anything to whether the record could have shown it, and the planner can answer that before the test rather than after.
ConditionCore's Motor Current Analysis workflow is designed to take records like these, with speed, slip and load attached, and treats a changed sideband family as grounds for a targeted review rather than a broken-bar declaration. The 15 October field note covers what three RMS readings can and cannot say.
A long lead time makes a capable test more valuable. It does not make an inconclusive test reassuring.
Review a machine with ConditionCore