How to Tell a Reset State From a Real Fault After a Power Interruption
Why this matters
Most modern equipment spends its first several minutes after a power interruption doing something that looks exactly like a failure. Anti-short-cycle timers hold a compressor off. Staged restarts bring loads back one at a time. Electronic controls relearn a valve position, a door position, or a flow baseline before they will act on a call. All of that presents to a customer as "it is on but it is not doing anything," which is word for word how they describe a dead compressor.
Getting this wrong costs in both directions. Condemn a part that was mid-recovery and you buy a return trip, a restock argument, and a customer who now reads every future diagnosis as a guess. Wait out a system that is genuinely damaged and you burn an hour, then do the same repair one visit later with a colder building behind you. The separation is not a feel. It is a clock and two measurements.
Lead with the isolation before any of it
If your check involves opening a panel, landing a probe on a terminal, or working on a branch circuit, de-energize at the disconnect, lock and tag it, and verify the conductors are dead before contact - the requirement to de-energize and apply lockout or tagout for work on electric circuit parts sits at 29 CFR 1910.333(b)(2), and on a construction site the electrical counterpart is 29 CFR 1926.417. Prove dead with the live-dead-live sequence in NFPA 70E-2021, 120.5: test your meter on a known live source, test the conductors, then re-test on the known live source to confirm the meter did not die between readings.
If the recovery you are watching involves a compressor, an accumulator, a spring-loaded damper, a pressurized vessel or any other stored energy, that is 29 CFR 1910.147 territory: isolate the energy source, apply your lock, and bleed or block the stored energy before you put a hand inside. A system that is mid-restart is one that intends to move. Treat "it has not started yet" as "it is about to."
Step 1: Fix the shape of the interruption before you touch the equipment
Three facts change everything downstream, and all three come from people and records, not from the unit.
- How long was power out? A three-second blip and a forty-minute outage put a control in different places. Short blips are worse for some controls, because the equipment may have been mid-cycle with a rotating load still spinning when supply returned.
- How many times did it cycle? Utility reclosers commonly make more than one attempt before a sustained outage. Two or three fast restore-and-drop events in a row are a different insult than one clean loss, and they are the ones that latch controls.
- Was it a clean loss or a sag? A brownout leaves voltage present but low, which can drop out a coil while leaving a board powered and confused. A clean loss is the gentler case.
Skip this step and you have no way to judge whether the recovery clock you are about to set is generous or absurd. You end up choosing a wait time out of the air, and a wait time you cannot justify is a wait time you will abandon halfway through.
Step 2: Set a recovery clock and write down the time it expires
State the rule as a number, per piece of equipment: the recovery window equals two times the longest published delay in that unit's own service literature, with a floor of 15 minutes, and the clock starts only when supply voltage has been continuously present. The doubling covers stacked delays, because a control that holds a compressor for one interval and a fan for another does not run them in parallel. The floor covers equipment whose literature publishes nothing.
Two conditions restart the clock rather than pausing it: another supply interruption, and any reset you perform. If you reset a control at minute 4, your window ends 15 minutes after that reset, not 15 minutes after restoration.
What you lose by skipping this: without a written expiry time, "waiting" becomes an argument between you and a customer standing over your shoulder, and you will lose that argument every time. A time written on the ticket converts the wait from hesitation into procedure.
Step 3: Sort the symptom into recovering, latched, or damaged
These three behave differently over the window, and the difference is observable without a single tool.
| State | What it looks like | What it does over the window | What clears it |
|---|---|---|---|
| Recovering | Powered, no output, no code or an informational code | Ends on its own at a time near the published delay | Time only |
| Latched | Powered, no output, a hard fault code, output stays off past the window | Nothing changes, ever, no matter how long you wait | A deliberate reset action |
| Damaged | Attempts to run and fails, or runs wrong | Repeats the same failure on every attempt | A repair |
The tell that separates latched from damaged is what happens after exactly one reset. A latch caused only by the interruption clears and stays clear. A latch protecting a real fault clears, the equipment attempts, and it locks out again on the same code. That is why the reset count matters: reset once to observe, never twice blind. A second blind reset destroys the only piece of evidence the first one produced, and repeated resets on a protective device are how a nuisance trip becomes a burned component.
Step 4: Take the measurement a recovery state cannot fake
A timer can hold an output off. It cannot make a running motor draw the wrong current, make a pressure sit where it should not, or make a temperature split that should be developing stay flat through a full cycle. So the measurement that settles the question is always taken on an attempt, not on the pause between attempts.
- Electrical: current on the load during an attempt, compared to the nameplate rating. A load drawing near zero with the contactor closed is an open circuit, not a delay. A load drawing several times rated for a couple of seconds before a protector opens is a mechanical or start problem, not a delay.
- Hydronic or plumbing: pressure or flow during an attempt. A pump that runs and moves nothing is not relearning.
- Combustion or thermal: whether the sequence advances between stages. A control that hits the same stage and stops, twice, at the same point, is failing there.
The reasoning matters more than the list. A recovery state is a decision the control is making. A fault is a physical fact the control is reacting to. Measure the physical fact and the control's opinion stops being relevant.
Worked example: a packaged unit after a two-blip outage
A restaurant loses power for 38 minutes. The customer reports two fast flickers before the sustained loss, so treat this as three interruptions, not one. Power has now been continuously present for 70 minutes when the tech walks in. The complaint is that the indoor fan runs and the air is not cold.
Set the clock first. The unit's service literature publishes one delay that matters, an anti-short-cycle hold of 5 minutes. Two times 5 is 10, which is below the 15-minute floor, so the recovery window is 15 minutes from restoration. Power has been up for 70 minutes, uninterrupted. The window closed 55 minutes ago. Whatever this is, it is not a recovery state, and the tech has just eliminated the entire wait-and-see branch in under a minute of conversation.
Now sort it. The control shows a hard lockout code rather than an informational one, so this is latched or damaged, not recovering. One reset. The unit attempts, the contactor pulls in, and roughly 3 seconds later it drops out and returns to the same code. That single observation moves it from latched to damaged: a latch caused only by the outage would have cleared and stayed clear.
Now measure on the attempt. Say the compressor nameplate rated-load current reads 12.0 and the clamp reads about 55 for the two to three seconds before the protector opens. That is roughly 4.6 times rated, which is the signature of a rotor that is not turning, not a timer that has not expired. The diagnosis is mechanical or start-related on that compressor, and the outage is context rather than cause: three interruptions in quick succession are a plausible trigger for a locked rotor on a unit that was already marginal, which is worth telling the customer, but the repair is the same either way.
Change one fact and the whole visit inverts. If the tech had arrived 8 minutes after restoration instead of 70, the identical no-cooling complaint would sit inside the recovery window, the correct action would be to wait out the remaining 7 minutes, and the same clamp reading would never have been taken because the unit would not have attempted yet. Same symptom, same unit, opposite call, decided entirely by a number the customer supplied at the door.
The failure mode here is specific and common: a tech who arrives at minute 8, sees no cooling, resets twice, gets no start, and writes up a compressor. That tech has three problems. The unit may have been fine, the second reset erased the evidence that would have proven it, and the repeated reset attempts loaded a possibly-healthy start component with back-to-back inrush.
How to verify you got this right
Three checks, all fast, all done before you write the quote.
- Read your own clock back. Confirm the window you set matches the literature you actually opened, that the clock started at continuous restoration and not at the first blip, and that any reset you performed restarted it. A window you set from memory is the most common quiet error in this whole method.
- Confirm the failure repeats on a fresh attempt. One attempt is an event. Two identical attempts, same stage, same code, same measurement, is a fault. If attempt two behaves differently from attempt one, you are still inside something changing on its own and you have not finished.
- Prove the recovery theory by letting it finish. If you concluded "recovery, not fault," stay until it completes a full cycle and produces output, and note that on the ticket. A recovery call you leave early is a callback you have scheduled for tomorrow with your own name on it.
References
- 29 CFR 1910.333(b)(2), OSHA electrical safe work practices: de-energize and apply lockout or tagout before work on electric circuit parts; 29 CFR 1926.417 is the construction counterpart for electrical lockout and tagging
- NFPA 70E-2021, 120.5, the live-dead-live process for establishing an electrically safe work condition
- 29 CFR 1910.147, OSHA control of hazardous energy, for isolating mechanical and stored energy before service
- Manufacturer service literature for published restart delays, staging intervals and lockout codes on the specific equipment
- See related: Correlating a Fault with a Recent Power Event; What a Brownout Does Differently Than a Full Outage; The Danger of Just Resetting a Safety Device Without Diagnosing Why