Many sites check their electrical connections with a scheduled infrared survey, often once a year. A thermographer opens panels on an agreed day and records whatever load the system happens to carry. It is a sound practice and it finds real faults. It also misses a predictable set of them: faults that heat only under peak load, faults that come and go, faults that develop between visits, and equipment nobody can safely open.
Continuous fixed thermal sensing watches the same connections under every load, around the clock. Neither approach replaces the other. This article covers what each catches, how to design continuous monitoring that does not bury you in false alarms, and how to combine the two into one program.
What a periodic IR survey catches
A competent survey with a radiometric camera finds loose and corroded connections, overloaded conductors, unbalanced phases, failing fuse clips and breakers, and overheating transformer components — provided they are hot on the day. A trained thermographer also sees the whole scene and notices the unexpected: a warm neutral, a blocked vent, a cooling fan that has stopped. And a survey covers a large asset base at a low cost per asset. For most of a facility’s electrical equipment, that is good value.
What it misses
Faults that depend on load
Heat generated at a resistive connection scales with the square of the current. At half load, a bad joint produces roughly a quarter of the heat it produces at full load. If survey day falls on a light-load day — a weekend, a slow season, a data hall half populated — a connection on its way to failure can look unremarkable.
A hypothetical example: a joint running 24 °C above its neighbors at full load would show only about 6 °C at half load, ignoring changes in cooling. Many survey programs would log the second reading as minor, or not log it at all.
Faults that come and go
Some faults only appear under particular conditions: a chiller plant starting, a production shift, a battery system charging or discharging hard, a heat wave. Thermal cycling also loosens connections progressively, so a joint can move from normal to dangerous well inside the interval between surveys.
Equipment you cannot safely open
Infrared does not see through metal covers. Surveying behind them means opening energized equipment, which exposes the thermographer to arc-flash hazards and requires the justification, PPE and qualified people your electrical safety program demands. Some equipment — enclosed medium-voltage switchgear, sealed battery containers — is impractical to open live at all, so it is either surveyed through closed covers, which tells you little, or not surveyed. IR windows help by allowing surveys with covers closed, but the survey is still a snapshot.
What continuous fixed sensing looks like
| Periodic handheld survey | Fixed radiometric thermal sensor | Contact point sensors | |
|---|---|---|---|
| Coverage | Whatever the thermographer can reach | Every surface in the field of view | One point per sensor |
| Frequency | Once per interval | Continuous | Continuous |
| Load conditions seen | Whatever occurs on the day | All of them | All of them |
| Installation | None | Mounting or IR windows, usually in an outage | On each joint, usually in an outage |
| Thermal image record | Yes, on the day | Yes, continuously | No |
A fixed radiometric sensor, such as a Lensnode T with 640 × 512 long-wave infrared resolution, measures temperature in every pixel, so one sensor can cover a row of cable terminations with a region of interest on each phase. Contact sensors are simpler and give one clean number per joint, but you need one for every point you care about. On large enclosures with many connections in view, imaging usually wins; for a few critical joints that cannot be seen, contact sensors may be the only option.
Fixed sensors do not need to stream video to be useful. An edge gateway can extract the maximum and average temperature for each region every few seconds, compute the phase differences locally, and keep a full radiometric frame only at intervals and whenever an alarm fires. That keeps network traffic small, keeps thermal images of your switch room on site, and still gives an engineer the picture they need to confirm a finding.
IR windows
A window must transmit in the sensor’s waveband. Some crystal window materials transmit poorly across parts of the long-wave band where most fixed thermal sensors operate, so check the window against the sensor, measure the actual transmission in place, and enter it as a correction. Windows also limit the field of view and get dirty; position them for the targets and add them to cleaning schedules. For switchgear, follow the equipment and window manufacturers’ guidance on ratings and placement.
Emissivity
Bare, bright copper and aluminum have low and variable emissivity. A thermal sensor looking at a polished busbar mostly sees reflections of its surroundings, and the computed temperature can be wrong by a wide margin — including false hot spots reflected from a warm neighbor. Aim regions of interest at surfaces with high, stable emissivity: insulated cable jackets, painted surfaces, or high-emissivity labels applied during an outage and rated for the location.
What to alarm on
Absolute temperature makes a poor primary alarm. Set it low and it fires every summer afternoon; set it high enough to stay quiet and it fires only when the fault is advanced. Keep an absolute limit as a backstop and build the main alarms on differences:
- Rise over ambient. Measure ambient inside or near the enclosure, not at a weather station.
- Difference between phases. The three phases of a circuit carry similar current at the same moment, so comparing A, B and C cancels most of the effect of load and ambient. It is the most informative signal in a three-phase lineup. Where loads are unbalanced, measure phase currents alongside — a clamp-on node such as Pulsenode C reports phase imbalance — so an electrical imbalance is not mistaken for a connection fault.
- Trend against load. Plot temperature rise against current squared. A healthy joint follows a stable line; a degrading one climbs above it over weeks. No periodic survey can show you that.
Published severity tables, such as those in the ANSI/NETA maintenance testing specifications, grade findings by temperature difference between similar components and over ambient. They are useful for deciding how urgently to respond. Continuous systems add something a table cannot: alerts on change.
Keeping false alarms down
Most false alarms come from a short list: reflections, an open enclosure door, sun on outdoor equipment, a sudden load step, a dirty window or a sensor knocked out of alignment. Each has a mitigation — persistence timers so a condition must hold for several minutes, masking during planned maintenance, sun-angle awareness on outdoor sites, and a periodic reference-image check for alignment and window condition. Then put a person in the loop to look at the thermal image and the load data before a technician is dispatched. Verified alerts are the difference between a system people trust and one they mute. The same principle runs through our predictive maintenance work on rotating equipment, where vibration spectra play the role thermal images play here.
When periodic is enough
Periodic surveys remain the right answer for a lot of equipment:
- Distribution where a failure is a nuisance rather than an outage or a safety event.
- Equipment that can be surveyed safely under representative load, for example through IR windows already fitted.
- Stable, steady loads, where load-dependent faults have little room to hide.
- Assets where the cost of fixed sensing exceeds the risk it reduces.
Continuous sensing is not cheap per point. The case rests on the consequence of a failure, its likelihood, and how much earlier continuous data would reveal it.
A combined program
- Rank assets by criticality and by how hard they are to survey safely.
- Monitor the top tier continuously: medium-voltage switchgear, main switchboards, UPS and battery systems, transformers feeding critical loads.
- Survey the rest periodically, on a high-load day that the continuous data helps you choose.
- Run one workflow. Findings from both go into the CMMS, follow the same repair and verification loop, and are re-checked after repair — immediately, on continuously monitored equipment.
Standards are moving the same way. The 2026 edition of NFPA 70B, Standard for Electrical Equipment Maintenance — which replaced the 2023 edition, the first issued as a standard rather than a recommended practice — permits permanently installed, continuously monitoring temperature devices to satisfy its infrared thermography requirements. Read the clause yourself, and confirm which edition your insurer, contracts or authority having jurisdiction reference.
Our switchgear thermal watch example engagement applies this approach to medium-voltage switchgear and battery containers; in that modeled scenario, connection faults are flagged weeks before a periodic survey would have found them. For the service itself, see continuous thermal and multispectral monitoring.