Normal Versus Abnormal: A Field Reference for Common Observations
Why this matters
Almost every observation a tech makes on a call is ambiguous on its own. Warm is normal until it is not. Noisy is normal until it is not. Wet, humming, cycling, drawing current, all of it sits in both columns depending on conditions you cannot see from the observation itself. A tech who tries to memorize which observations are normal will be wrong roughly half the time, because the observation is not what carries the information.
What carries the information is the discriminating test: the one additional thing you check that moves an ambiguous observation into a column. This card is organized around those tests, not around descriptions.
The four discriminators that do most of the work
Before the individual keys, learn these four. Between them they resolve the majority of ambiguous field observations across every trade, and you can apply them to an observation this card never lists.
Timing. Is the behavior tied to a repeatable point in the operating cycle, or does it wander? Normal behaviors are choreographed. They happen at startup, at shutdown, at a transition, on an interval, and they happen the same way every time. Defects keep poor time. If three occurrences are spaced within a few percent of the same interval, you are looking at a control. If the spacing wanders widely, you are looking at a fault.
Trend across a run. Take the reading cold, then take it again at heat soak, which is when component temperatures have stopped climbing, usually somewhere between roughly 20 minutes and about an hour depending on thermal mass. Normal readings settle. Defective ones walk in one direction and keep walking. A reading that is inside tolerance at both ends but moved substantially in one direction between them is a finding, even though neither endpoint condemns anything.
Comparison to an identical peer. Almost every system contains something that exists in pairs or sets: two legs, three phases, four terminations, six zones, two identical rooms. Compare the suspect against its twin under the same conditions. Asymmetry between things that should be identical is the fastest true finding in the trades, because it needs no published specification to interpret.
Persistence after shutdown. Turn it off and wait. Behaviors driven by operation stop when operation stops. Anything still happening well after shutdown has a different cause than the one you assumed. This single test separates condensation from a leak, residual heat from an active heat source, and normal drain-down from a real flow problem.
Sounds
| Observation | Normal when | Abnormal when | The discriminating test |
|---|---|---|---|
| Sharp tick or pop | Only in the first minutes of a run and the first minutes after shutdown | It continues at steady operating temperature | Time it against the cycle; expansion noise stops once temperature stops changing |
| Continuous rushing or whistling | It tracks flow, present whenever the system runs at that output | It appeared without any change in output setting | Change the flow path or output stage and see if the sound tracks it |
| A single clunk at a transition | Same point in every cycle, same character | Irregular, or a new sound after a component change | Cycle it three times and check whether the event lands identically |
| Hum or buzz | Steady, unchanged from commissioning, tracks load | It grows through the run or is louder than its identical peer | Compare against the twin component; check trend at heat soak |
| Grinding or rubbing | Never | Always | Shut down and inspect; do not run to reproduce |
That last row is not a hedge. Grinding is contact between things that should not touch, and running it to characterize the sound converts a repair into a replacement.
Water and moisture
| Observation | Normal when | Abnormal when | The discriminating test |
|---|---|---|---|
| Water on a cold surface | Surface is below the dew point and water reaches an intended drain path | The surface should be warm, or there is no path for the water | Touch-test the surface temperature and trace the drain path end to end |
| Standing water in a pan | Momentarily during heavy load, clearing on its own | It persists after the system has been off for hours | Shut down and return; condensation stops when the cold surface warms |
| A visible plume or steam | It coincides with a designed purge, defrost, or regeneration event on an interval | It occurs at random or outside the season it belongs to | Time the interval across three occurrences; check the documented sequence |
| Damp insulation | Never, though a small area can be a one-time event | Widespread or recurring | Check whether it dries out and stays dry over a full cycle after the source is addressed |
Heat
| Observation | Normal when | Abnormal when | The discriminating test |
|---|---|---|---|
| Warm component | It is within its rated operating temperature and matches its identical peer | It runs hotter than its twin under the same load | Peer comparison first, published rating second |
| Warm termination | It is not; terminations should be at or near the temperature of the conductor | Any measurable rise localized at the joint | Thermal comparison against the other terminations on the same load |
| Hot surface that cools slowly after shutdown | Thermal mass retaining heat | It stays hot or climbs after shutdown | Persistence test; anything still generating heat with no input is a fault |
| Burn-off smell on a seasonal first run | First heating cycle of the season, fading within a run or two | It returns after several cycles, or is accompanied by visible smoke | Run three cycles; a dust burn-off does not survive repetition |
Cycling and control behavior
| Observation | Normal when | Abnormal when | The discriminating test |
|---|---|---|---|
| Short cycles | Load is far below design, typical on mild days | Cycle length is identical regardless of load, or ends on a limit rather than on satisfying demand | Log three cycles and check whether length varies sensibly with conditions |
| Continuous operation | Conditions are at or near design load, or the equipment is a modulating type built to run long and low | It never satisfies even at mild conditions | Compare duty cycle against the load on the day, not against intuition |
| Refusal to restart immediately | A protective delay is holding it off for a fixed interval | The delay is longer than documented or it never restarts | Wait out the documented delay before touching anything |
| Reversal or interruption mid-run | It matches a documented periodic sequence and keeps time | Interval or duration wanders, or it occurs out of season | Interval timing across three events |
Electrical readings
| Observation | Normal when | Abnormal when | The discriminating test |
|---|---|---|---|
| Current below nameplate | The load is below design; most equipment rarely reaches nameplate | It is far below and output is also low | Correlate with actual output, not with the nameplate alone |
| Current above nameplate | Briefly at startup inrush | Sustained at steady operation | Read at heat soak, not at start; check the trend across the run |
| Voltage in range but the device will not run | Never conclusive; a reading under no load proves very little | The reading collapses when the load is applied | Measure under load, not open-circuit |
| Slight imbalance between legs | Small variation is expected in real installations | It grows across the run or exceeds the manufacturer's stated limit | Trend plus the published imbalance limit; do not eyeball it |
The worked example
A tech is called for a burning smell that has been intermittent for two weeks. On arrival, nothing smells. A panel is opened after de-energizing and verifying dead with live-dead-live, and everything looks clean. A snapshot inspection would close this no fault found.
Apply the discriminators instead.
Peer comparison. Four identical terminations on the same load. Under running load, three of them sit essentially at conductor temperature. The fourth is measurably warmer to a thermal reading. No specification was needed to make that finding, because three identical neighbours under the same load are the specification.
Trend. Current is logged at 4 minutes cold and again at 70 minutes at heat soak. The two legs read 4.8 and 5.1 amps against each other at the soaked reading, which is roughly a 3 percent imbalance against their average. On its own that is inside what many installations show and would not condemn anything.
Corroboration. The warm termination sits on the leg reading higher. Now the 3 percent imbalance stops being ambient noise and becomes the electrical signature of the same joint the thermal reading flagged. Two weak signals pointing at one physical location beat either one alone.
Persistence. The system is shut down and the joint checked again after twenty minutes. It has cooled at the same rate as its neighbours, which rules out a source generating heat independently and confirms the heat was resistive, produced by current crossing that joint.
The finding is a high-resistance termination, caught before it opened or ignited, on a call that presented with no symptom at all at the time of arrival. Not one step of that required a manufacturer specification. It required a peer, a runtime, and a shutdown.
Where this card fails you
Two ways, both worth knowing.
It fails when the system has no healthy peer to compare against, because everything of that kind on site was installed at the same time and is degrading together. A whole panel of warm terminations reads as symmetric and therefore normal. When you suspect this, the peer has to come from outside: another property, another identical system, the commissioning record, or the published rating.
It fails when a system has been operating out of spec long enough that its own history has normalized. A customer who says "it has always sounded like that" is reporting a real baseline, and the card will tell you the behavior is consistent, which it is. Consistency is not correctness. When the peer test and the trend test both come back clean but the output does not match design, go back to the design target rather than trusting the behavioral evidence.
The failure mode in the field looks like this: a tech applies the discriminators, everything reads symmetric and stable, the visit closes normal, and the whole set of components fails within a season. The tell that was available and got skipped was that nothing was ever compared against the design target, only against itself.
References
- Manufacturer documentation for rated operating temperature, nameplate current, and permissible imbalance limits
- NFPA 70E live-dead-live verification practice before contacting any conductor or terminal
- Trade-standard thermographic inspection practice for comparing like components under like load
- See related: Why Normal Noise, Cycling, and Condensation Alarm Customers; When Normal Behavior Signals a Real Problem Anyway; Baseline vs Faulty: The Comparison Diagnosis Method