Industries / Industrial equipment and services / Industrial services / Motor rewinding
Testing and diagnosis
Testing and diagnosis is the work of finding out what is wrong with a motor or generator, and how close it is to failing, by measuring its insulation, its current, its vibration and its heat, in the shop or while it runs.
5
Milestones achieved, from 1977 to 2000.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1977
A guide for turn tests
A winding is made of coils, each of many turns of insulated wire, and it fails when the thin insulation between turns, between phases or between the winding and the steel core breaks down. A steady DC test does not put voltage between neighboring turns; a fast voltage surge does, and a Reliabilityweb article on motor testing calls the surge test "the only known method where instantaneous turn-to-turn voltage can be developed". In 1977 the IEEE issued a trial-use guide for testing turn-to-turn insulation on form-wound stator coils (coils shaped before they go into the slots of larger machines), after test methods such as the one James Oliver, Herbert Woodson and John Johnson described in IEEE Transactions on Power Apparatus and Systems in March 1968. The IEEE revised the guide in 1992, 2004 and 2023.
Credit
IEEE; James Oliver, Herbert Woodson and John Johnson
1986
Faults read from motor current
Oak Ridge National Laboratory (ORNL), studying wear in the motor-operated valves of US nuclear power plants, reported in 1986 that the motor's current during a valve stroke was a useful sign of trouble in the gears and parts it drives. Howard Haynes and David Eissenberg turned this into motor current signature analysis (MCSA): a clamp-on sensor around a power lead records the current, and frequency analysis picks out mechanical and electrical faults without opening the machine (US patent 4,965,513, 1990). In a program reported in 1990, the Electric Power Research Institute's (EPRI) Monitoring and Diagnostics Center used MCSA on motor-driven equipment at the Philadelphia Electric Company's Eddystone Generating Station.
Credit
David M. Eissenberg and Howard D. Haynes, ORNL; EPRI Monitoring and Diagnostics Center
1986
Discharges measured in service
Partial discharges are tiny sparks in gaps inside high-voltage insulation, and over years they eat it away. After some high-profile hydro generator failures in Canada in the mid-1970s, the Canadian Electrical Association paid for research at Ontario Hydro's Research Division on a test that utility staff could run while the generator was working. Mo Kurtz led the project and Greg Stone was its engineer; it produced an 80 pF sensor made from a short length of power cable and a way to tell discharge pulses from electrical noise, which earlier methods left to an expert reading an oscilloscope. Ontario Hydro and the association licensed the instrument to FES International in 1986, and in 2015 Stone, writing for Iris Power, which now sells it, counted more than 60,000 such sensors and more than 3,500 continuous monitors installed.
Lessonthe test spread once it could separate the fault signal from noise without an expert at the screen.
Credit
Mo Kurtz and Greg Stone, Ontario Hydro; Canadian Electrical Association
1995
Vibration judged by one standard
ISO 10816-1, published in December 1995, set general rules for measuring a machine's vibration on its non-rotating parts and judging the result, both for monitoring in service and for acceptance tests. It replaced ISO 2372 of 1974 and ISO 3945 of 1985, and ISO 20816-1 replaced it in 2016.
Credit
ISO technical committee TC 108/SC 2
2000
Resistance test revised
The insulation resistance test puts a DC voltage on a winding and reads, in megohms (millions of ohms), how well its insulation blocks current; the polarization index (PI) compares a reading taken after 10 minutes with the one-minute reading. In 2000 the IEEE revised its 1974 recommended practice for this test, IEEE 43, to accommodate "changes and improvements in insulating materials". Graeme Thomson and Jeff Jowett, summarizing the revision in 2003, note that some new insulation shows less absorption than older types, and that one-minute readings above 5,000 megohms come from currents so small that humidity, an unsteady test voltage and the test connections can sway them, which can make a PI from them unsuitable.
Lessona pass mark set for one generation of insulation can mislead on the next.
Credit
Electric Machinery Committee, IEEE Power Engineering Society
Watch for
Motor faults from current alone
Eaton says its motor analytics software, an add-on to its Brightlayer software introduced on 22 April 2026, uses motor current signature analysis and machine learning to predict common motor and pump problems "months in advance" without sensors on the motor, and predicts issues "up to 30% earlier and 25% more accurately than currently available sensing solutions".
Achieved whena user or an independent tester publishes results on a named set of motors, comparing the software's warnings with what was found when those motors were opened.
Claimed by
Eaton, April 2026
Open
01
Surge tests from real failures
A public library of surge test waveforms that repair shops recorded on motors that came in failed or for overhaul, each tied to what the winder found when the motor was opened, and a classifier trained on it that flags turn faults and is scored on motors from a shop it never saw.
The briefCredit · unclaimed02
Name the failure from photos
An open set of photographs of failed stator windings, taken in repair shops as they are opened and labeled with the failure pattern a winder reads from the damage (for example one phase lost, called single-phasing; a short between turns or between phases; a winding grounded in the slot; overload; unbalanced voltage; a voltage surge), and a model that names the likely pattern from a phone photo.
The briefCredit · unclaimed
Open 01 · the brief
Surge tests from real failures
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public library of surge test waveforms that repair shops recorded on motors that came in failed or for overhaul, each tied to what the winder found when the motor was opened, and a classifier trained on it that flags turn faults and is scored on motors from a shop it never saw.
Why it is still open
Testers already score the difference between two phases' waveforms as an error area ratio (EAR), which EASA's Mike Howell wrote in 2016 was adopted to remove the test's subjectivity, but no published study was found that checks such scores against what winders found on teardown. Published work on reading surge waveforms by machine uses faults made on purpose: Hisahide Nakamura and Yukio Mizuno (Energies, 2020) soldered shorts between turns of 1.5 kW and 2.2 kW motors, because "collecting a sufficient number of such windings from actually damaged motors would have taken a long time and been expensive." A November 2026 GitHub search for "surge test winding" returned no repositories, and web searches found no labeled set of shop surge waveforms.
Built with what exists
Digital surge testers that already store each waveform (the shops run the tests; the builder handles only the files); Python with NumPy, SciPy and scikit-learn; a shared spreadsheet of teardown findings; phone photos of each opened winding.
First step
Ask one repair shop to export last year's stored surge test files with job numbers and the matching teardown notes, and label 30 by hand.
Done looks like
Waveforms from at least 300 motors from at least three shops, each with a teardown finding checked by a second winder; a classifier trained on all but one shop and tested on that shop's motors, with its hit rate and false-alarm rate for turn faults published alongside the data, set beside the EAR pass marks the shops use now.
Open 02 · the brief
Name the failure from photos
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of photographs of failed stator windings, taken in repair shops as they are opened and labeled with the failure pattern a winder reads from the damage (for example one phase lost, called single-phasing; a short between turns or between phases; a winding grounded in the slot; overload; unbalanced voltage; a voltage surge), and a model that names the likely pattern from a phone photo.
Why it is still open
Winders learn these patterns from reference pages such as EASA's "Failures in Three-Phase Stator Windings", which shows one photo for each of 13 patterns for people to compare, and whose photos may not be reproduced. Published image work covers other cases: Michael Farnsworth and colleagues at the University of Sheffield (2022) trained models on defects made while winding new coils in manufacturing, and an October 2025 arXiv paper classified thermal images of running motors. Web searches found no labeled photo set of failed windings from repair shops, and November 2026 GitHub searches for "stator winding fault images" and "stator winding failure" returned no repositories.
Built with what exists
Phone cameras and a fixed light; Label Studio for labeling; an image classifier fine-tuned on the photos, such as an Ultralytics YOLO classification model, on one graphics card; the Hugging Face Hub for publishing the dataset.
First step
Photograph 50 failed windings at one shop as they are opened, and have two winders label them separately to settle the list of patterns.
Done looks like
At least 2,000 photos from at least five shops, each labeled separately by two experienced winders; the winders' agreement published; the model's agreement with them measured on photos from a shop left out of training; photos, labels and code published under an open license.