How to Read an Insulation Resistance Result Over Time

Why this matters

Most shops that meg a motor annually are not trending anything. They are collecting a column of numbers taken at whatever temperature the machine happened to be, held for whatever count the tech happened to use, on whatever boundary of cable and leads happened to be connected that year. Then somebody lines the column up, sees it falling, and orders a motor.

Insulation resistance moves hard with temperature. A reading taken on a warm machine in August and one taken on a cold machine in February differ by a large factor with nothing wrong at all. An uncorrected column manufactures a decline every summer and hides one every winter, and it does both while looking exactly like data.

What makes a reading comparable to next year's reading is a record with enough fields in it that someone can normalize both to the same basis. This card builds that record field by field, then runs four years of one machine through it. The sibling card on what an insulation resistance test actually tests owns the mechanism and the limits of a single reading; this one owns what turns a series of them into evidence.

Before the test set comes out

The full isolation sequence is in the sibling card and it is not optional here: isolate and lock under 29 CFR 1910.333(b)(2) (which governs rather than 29 CFR 1910.147, whose 1910.147(a)(1)(ii)(C) carve-out sends electric utilization work to Subpart S), verify absence of voltage with a tester proved live-dead-live per NFPA 70E-2021, 120.5 in the edition your employer's electrical safety program has adopted, control the far end of every cable, and disconnect drives, boards and electronic overloads before you apply test voltage.

Three hazards belong specifically to trending, because trending asks you to do things a one-off test does not.

Taking a winding temperature means approaching a machine that was running. A frame at operating temperature burns skin on contact. Read it with a non-contact instrument or an embedded detector, keep the lock and tag on the whole time, and do not restore power to warm the machine for a reading.

A ten-minute test stores far more charge than a one-minute test. Polarization index holds test voltage for ten minutes with everything in the circuit charged throughout. Let the instrument discharge, apply a grounding device and confirm zero, and allow noticeably longer than after a one-minute reading before hands go anywhere.

A trend requires the same circuit boundary every year, which means lifting the same leads every year. A terminal box that has been wet has sharp edges, corroded lugs and an awkward reach. Support the leads rather than pulling on them, and re-terminate to the torque the manufacturer's literature specifies rather than to feel.

The two things that make readings comparable

Temperature. Leakage through and across insulation increases with temperature, so measured resistance falls as the machine warms. The long-standing field rule of thumb is that insulation resistance roughly halves for each 10 C rise and doubles for each 10 C fall, referenced to a 40 C base. That rule of thumb was derived on older absorptive insulation systems, and it is a rule of thumb, not a coefficient: modern epoxy and mica systems show a much weaker temperature dependence, and the correction curves published in IEEE 43, in the edition your specification or the machine manufacturer's literature references, are specific to insulation type and govern where they are available. A rule of thumb and a published curve are two different kinds of number. Where both exist, the curve wins and the rule of thumb is only for sanity-checking it.

Time under test. The reading climbs while you hold the button, because charging and absorption currents are decaying and only the steady leakage current remains at the end. A reading called at ten seconds and a reading called at sixty seconds are two different measurements. Pick one duration and never change it.

The ratio of two timed readings on the same machine is worth having precisely because both are taken at the same temperature, so the ratio carries no temperature error at all. Polarization index is the ten-minute reading divided by the one-minute reading; dielectric absorption ratio is the sixty-second divided by the thirty-second. A ratio near 1 means the reading stopped climbing, which is what a wet or contaminated winding does. IEEE 43 also cautions that the ratio loses meaning when the one-minute reading is already very high, because there is little absorption left to observe; check the edition your specification references for the threshold it sets rather than assuming one.

The record, field by field

Ten fields. Every one of them exists because leaving it out breaks a comparison someone will try to make later.

1. Equipment identity and location. Not "rooftop motor". The specific machine, so a replacement shows in the series as a discontinuity rather than as a recovery.
2. Circuit boundary. Exactly what the leads reached: motor only, motor plus machine leads, or the whole run back to the disconnect. Leakage paths add in conductance, so a year tested with the cable in is not comparable to a year tested without it.
3. Test voltage applied. Leakage current does not always scale linearly with voltage. Two years at two voltages are two series.
4. Duration at which the reading was called. Thirty seconds, sixty seconds, ten minutes. State it.
5. Raw reading, per phase or per conductor. One phase falling while two hold is a different finding from all three falling together.
6. Winding or frame temperature at the moment of test, and how it was obtained. Embedded detector, contact probe, or non-contact on a known surface. A guessed temperature makes the correction worse than no correction.
7. Corrected reading and the base you corrected to. A column of corrected values with no stated base is not comparable to anyone else's column.
8. Relative humidity and whether the machine or terminal box was visibly wet. Surface leakage across a damp terminal block is large, real, and completely reversible.
9. The acceptance figure and where it came from. The manufacturer's service literature or the standard your specification calls out. Not a number somebody remembered.
10. Instrument used and the date of its last calibration. A trend taken with three instruments over three years has an unknown step wherever the instrument changed.

The record filled in: one motor, four annual tests

Same machine, same technician, motor leads lifted at the terminal box each year so the boundary is the motor alone, test voltage per the manufacturer's literature, reading called at sixty seconds every time, corrected to a 40 C base using the halving-per-10-C rule of thumb because no published curve for this machine's insulation class was available. That last choice is stated in the record so the next reader knows the correction is approximate.

Read the raw column first, the way a shop without this record would. It falls every year: 620, then 310, then 240, then 96. Year 2 is exactly half of year 1, which reads as a dramatic first-year collapse and is the point at which an uncorrected shop starts budgeting for a rewind.

Now read the corrected column. It held at 219, held at 219, fell to 158, then fell hard to 36. Two different stories.

The entire apparent 50 percent drop between year 1 and year 2 was the 10 C temperature difference, which is precisely the halving the rule of thumb predicts. Nothing happened to that machine in year 2. Year 3 is the first real movement: 219 down to 158 is a drop of about 28 percent, small enough to be inside the noise of an approximate correction and a frame-temperature estimate, and large enough to warrant tightening the interval rather than waiting another twelve months. Year 4 is the finding: 158 down to 36 is a drop of about 77 percent in one interval, on a corrected basis, on the same boundary, at the same voltage, at the same duration.

Why the year-3 call is "shorten the interval" and not "act". A single 28 percent step, corrected with a rule of thumb rather than a published curve and from a frame temperature rather than an embedded detector, sits inside the uncertainty the record itself declares. The year-4 step does not. The right response to an ambiguous step is a shorter interval, which converts one ambiguous point into two and resolves it.

What the corrected column still does not tell you. It does not locate the leakage, because the boundary was the motor alone and everything inside it is in parallel. It does not evaluate turn insulation. And it does not say the machine will fail, only that its ground-wall leakage is rising steeply on a consistent basis.

The failure mode. A shop keeping only the raw column orders a motor at the end of year 2 on a change that was entirely temperature, and then has two years of readings on a new machine with no way to compare them to the old series. The uncorrected trend does not merely mislead once; it destroys the baseline, so the next real decline has nothing to be measured against.

What would change the answer. If year 4 had been taken with the cable included and the earlier three without it, the 36 megohms would be a boundary change and not a finding, and the correct next action would be to re-test on the original boundary before drawing any conclusion. Equally, if a published correction curve for this insulation class had been available and showed a much flatter temperature dependence, the year 1 to year 2 step would not fully collapse, and part of that apparent decline would become real. The correction you use is part of the finding, which is why field 7 records it.

Turning the series into a decision

• A flat corrected series is the answer you want and the one most machines give. Keep the interval.
• A steady, shallow corrected decline over several intervals is aging. Plan for it; do not chase it.
• A single steep corrected step is an event, not aging. Water entry, a failed seal, a wash-down, a season change. Look for the event before you look at the machine.
• A corrected value approaching the acceptance figure from the governing document is the trigger for action, and the trigger is the corrected value, never the raw one.
• A ratio that has collapsed toward 1 while the absolute value still looks acceptable is a moisture or contamination finding arriving before the absolute number does, which is the main reason to take the ratio at all.

How to verify the trend is real before you spend on it

• Re-read the boundary field for every year. If any year differs, that year is not in the series.
• Re-read the duration field. A sixty-second reading compared against a ten-second reading overstates the decline.
• Re-compute one correction by hand. Confirm the direction: a machine COOLER than the base was reading high and corrects DOWNWARD; a machine WARMER than the base was reading low and corrects UPWARD. Every row in the table above is a cooler-than-base machine, which is why every corrected value sits below its raw value.
• Confirm each row's temperature was measured, not assumed. A row with an assumed temperature is a row with an assumed conclusion.
• Take one immediate repeat reading before acting on a steep step. If the repeat matches, the step is real. If not, you have an instrument, connection or boundary problem to find first.

References

• 29 CFR 1910.333(b)(2) - lockout and tagging of circuits for work on electrical installations
• 29 CFR 1910.147(a)(1)(ii)(C) - the carve-out sending electric utilization work to Subpart S rather than the general lockout standard
• NFPA 70E-2021, 120.5 - verification of an electrically safe work condition, binding through your employer's electrical safety program or your contract in the edition adopted
• IEEE 43, in the edition referenced by your specification or the machine manufacturer's service literature, for temperature correction curves, acceptance values and polarization index guidance
• See related: What an Insulation Resistance Test Actually Tests; Trend vs Snapshot Readings; The Calibration Schedule Worth Keeping