Thermal camera electrical inspection: finding hotspots in panels

Loose lugs, overloaded poles and failing contactor contacts announce themselves as heat long before they announce themselves as smoke: a joint that will fail at the next plant peak is already running kelvin above its healthy siblings today. This guide applies thermal imaging to electrical panels, MCC buckets and switchgear — the pixel count a lug needs, the NETD worth paying for, one worked example with the arithmetic, the emissivity trap of bright bus bar, and a scan routine you can repeat. One frame also reads hundreds of connections at once, which is why a thermal imaging camera catches faults a spot pyrometer walks past. And because thermal cameras are passive, there is no lighting and no controller in this build: camera, lens, cable, software.

How many pixels do I need across a lug or fuse?

Three pixels across the smallest target is the floor for seeing that a hotspot exists, and roughly ten pixels across is what you want when the temperature number itself goes in the report. Both come from one datasheet figure — the angular resolution, or IFOV.

Divide the horizontal field of view by the horizontal pixel count and you have the instantaneous field of view of a single pixel. An 88.5° horizontal FOV over 640 pixels is 88.5 ÷ 640 = 0.138° = 2.4 mrad per pixel, so one pixel covers 2.4 mm of panel at 1 m and 1.2 mm at 0.5 m. Datasheets often sell the same figure as a D:S ratio: 2.4 mrad is about 400:1, per pixel.

The calculator below runs the same sum from the other end: field of view and smallest feature in, sensor pixels out.

How far away can I scan and still see a hotspot?

With a 640 × 512 camera covering an 88.5° × 73.2° field of view, 0.5 m is the working distance: the frame holds a 600 × 700 mm MCC bucket whole, and a 12 mm lug lands on ten pixels. That is report grade.

The arithmetic, step by step:

1. IFOV = 88.5° ÷ 640 = 0.138° = 2.4 mrad per pixel.
2. Spot at 0.5 m = 2.4 mrad × 0.5 m ≈ 1.2 mm per pixel.
3. Lug coverage = 12 mm ÷ 1.2 mm = 10 pixels — enough to log the temperature.
4. Frame width = 2 × 0.5 m × tan(44.25°) = 0.97 m; frame height = 2 × 0.5 m × tan(36.6°) = 0.74 m.

The decision falls out of the table. Scan at 0.5 m when the lug temperature goes in the report; 1.0 m still clears the three-pixel floor for detection; 2.0 m does not, although a 20 mm breaker pole keeps 4 pixels there. A wide-FOV datasheet IFOV is a field average, so hold targets near frame centre, where the pixel footprint matches spec. The general method, with the trig for other lens angles, is in thermal camera working distance and spot size for safe scans.

The camera behind the example is the LS-IRC-0044: its data prints 640 × 512 pixels, an 88.5° × 73.2° field of view, 8-14 μm, 50 Hz and GigE.

What temperature rise means a loose connection?

Compare phases, not absolutes: 4 to 15 K above the sibling phase means schedule investigation, above 15 K — or climbing between scans — means urgent, and below 3 K you trend it and move on. Always write the measured load next to the image.

A healthy termination sits within a kelvin or two of the conductor it feeds; resistance heating scales with the square of current. That square matters when you read the scan: a 5 K rise at 40% load predicts 5 ÷ 0.4² = 5 ÷ 0.16 ≈ 31 K at full load — a work order, not a note. Those bands are the maintenance practice of electrical thermography, and the fault signatures behind them, component by component, are collected in the infrared electrical inspection troubleshooting knowledgebase.

NETD — the sensor's temperature noise floor — should be 50 mK or better for this work; the figure is quoted against a 25-30 °C blackbody, usually through f/1.0 optics. Keep the rise you act on at least five times NETD: 5 × 50 mK = 0.25 K, met with margin by any connection fault, and the 50 mK class is what keeps 1-2 K gradients on fuse clips visible at all. Lens speed moves the number: NETD scales with the square of the f-number, so f/1.2 glass multiplies an f/1.0 baseline by (1.2 ÷ 1.0)² = 1.44 — a 50 mK sensor reads about 72 mK. Still a tenth of a 5 K rise. The spec and its trade against lens choice are covered in thermal camera NETD and lens choice for process monitoring.

Which thermal camera suits a fixed panel install?

An 8-14 μm LWIR camera with 640 × 512 pixels on GigE at 50 Hz is the right default for fixed panel and switchgear monitoring; a USB3 module suits a compact station beside one cabinet.

LWIR is the band for panels: room-temperature targets radiate there, the atmospheric window is clean at building distances, and uncooled sensors live happily in an enclosure. The parts list stays short — camera, lens, cable, capture software. On cable, GigE covers 100 m on copper back to a control cabinet; USB3 is a few metres before it needs retimers. Filter the camera finder by LWIR and GigE to see the current range.

Frame rate: LWIR sensors faster than 9 Hz fall under export control in many jurisdictions, which is why handheld units are often 9 Hz. Fixed industrial cameras run higher — the LS-IRC-0044 prints 50 Hz — and 50 Hz makes a walk-down sweep read smoothly on screen.

Three LWIR options from the infrared cameras category:

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

When the spot must shrink without moving the camera, buy a narrower field:

The MV-CI003-GL-N25 prints a 25 mm EFL, a 0.3 MP LWIR sensor, GigE and an M34x0.75 mount. A longer EFL narrows the field, and the spot size falls in proportion: double the focal length, half the spot at the same distance.

On the lens side, match image circle to the sensor diagonal and mount to the camera before ordering; LWIR optics from wide angle to telephoto sit in the thermal, SWIR and UV lenses category. One LWIR example:

The LS-SWL-0006 prints 17 mm, F1.2, an 11-16 mm image circle and 1/1.2" sensor format — weigh that F1.2 speed against the NETD tax worked out above.

Why does the bus bar read the wrong temperature?

Because bright copper emits almost nothing — emissivity 0.05 to 0.15 — the camera mostly reports the reflected room, not the bar; set emissivity to the surface you are actually looking at, or read high-emissivity tape stuck on the joint.

A surface of emissivity ε sends the camera ε of its own thermal radiation and 1 − ε of whatever it reflects; at ε 0.10 that is 90% background. Typical values from published emissivity tables: bright copper or plated bus 0.05-0.15, oxidized copper 0.6-0.8, paint, anodising and PVC electrical tape 0.9-0.96.

Three habits keep the readings honest:

1. Compare identical phases with identical surface finish — a real ΔT survives the emissivity error that wrecks absolute numbers.
2. Set emissivity to 0.95 and read tape patches on the joints, applied de-energized.
3. Keep the viewing angle within about 60° of surface normal; emissivity falls away at grazing angles.

Never tune emissivity on a suspect joint until it agrees with a contact probe — you would be calibrating in exactly the error you are hunting.

Do I need to open the panel to scan it?

No, if the panel has an infrared window: a window that transmits 8-14 μm lets you scan through bolted covers, while a steel door is opaque at those wavelengths and reads only its own surface. An open-cover scan is a different job entirely.

Ordinary acrylic and polycarbonate viewing windows are also opaque in LWIR, so a visible-light inspection window does not pass thermal — the window must be specified for the band. Material choice, transmission loss and mounting are covered in choosing an infrared window material for a thermal sensor enclosure. Enter the window's transmission correction in the camera software, or every reading through it runs low.

What makes a panel scan repeatable?

Scan above 40% of rated load, from a marked floor position at a fixed distance, with unchanged emissivity settings, and save the radiometric file every time — the trend in phase-to-phase rise is what catches the failing joint, not any single reading.

1. Load at 40% of rated or more; record amps with the image. Below that, heating has fallen with the square of current and faults hide.
2. Fixed distance and angle: tape an X on the floor — 0.5 m in the worked example above.
3. Same emissivity settings and the same tape patches, scan after scan.
4. Radiometric files, not screenshots: radiometric jpg thermal images keep the temperature data re-readable for trending.
5. Log the ΔT between phases, not just the hottest spot, and plot a line across the bus — temperature vs position line profiles from radiometric pixel data show the gradient a single point reading misses.
6. Re-scan after any breaker, cable or bus work; at least annually, quarterly on critical switchgear.

Build the station in the configurator from the camera step, or send the panel sizes and distances to the applications team and we will run the numbers with you.